gh actions: upgrade to 1.2.11-1

Added: 增加IPv6支持

.gitignore

@@ -31,3 +31,4 @@ src/Makefile
 
 # other
 *.swp
+*.swo

debian/changelog

@@ -1,3 +1,196 @@
+libserverframe (1.2.11-1) unstable; urgency=medium
+
+  * upgrade to 1.2.11-1
+
+ -- YuQing <384681@qq.com>  Sun, 23 Nov 2025 10:48:22 +0000
+
+libserverframe (1.2.11-1) unstable; urgency=medium
+
+  * upgrade to 1.2.11-1
+
+ -- YuQing <384681@qq.com>  Sun, 23 Nov 2025 10:00:56 +0000
+
+libserverframe (1.2.11-1) unstable; urgency=medium
+
+  * upgrade to 1.2.11-1
+
+ -- YuQing <384681@qq.com>  Sun, 23 Nov 2025 09:06:43 +0000
+
+libserverframe (1.2.8-1) unstable; urgency=medium
+
+  * upgrade to 1.2.8-1
+
+ -- YuQing <384681@qq.com>  Sat, 16 Aug 2025 16:32:03 +0000
+
+libserverframe (1.2.7-1) unstable; urgency=medium
+
+  * upgrade to 1.2.7-1
+
+ -- YuQing <384681@qq.com>  Sun, 06 Apr 2025 16:56:40 +0000
+
+libserverframe (1.2.5-1) unstable; urgency=medium
+
+  * upgrade to 1.2.5-1
+
+ -- YuQing <384681@qq.com>  Sun, 29 Sep 2024 15:24:39 +0000
+
+libserverframe (1.2.4-1) unstable; urgency=medium
+
+  * upgrade to 1.2.4-1
+
+ -- YuQing <384681@qq.com>  Sat, 15 Jun 2024 14:45:44 +0000
+
+libserverframe (1.2.3-1) unstable; urgency=medium
+
+  * upgrade to 1.2.3-1
+
+ -- YuQing <384681@qq.com>  Sun, 17 Mar 2024 15:11:04 +0000
+
+libserverframe (1.2.2-1) unstable; urgency=medium
+
+  * upgrade to 1.2.2-1
+
+ -- YuQing <384681@qq.com>  Wed, 31 Jan 2024 12:00:10 +0000
+
+libserverframe (1.2.1-1) unstable; urgency=medium
+
+  * upgrade to 1.2.1-1
+
+ -- YuQing <384681@qq.com>  Mon, 01 Jan 2024 11:24:45 +0000
+
+libserverframe (1.2.0-3) unstable; urgency=medium
+
+  * upgrade to 1.2.0-3
+
+ -- YuQing <384681@qq.com>  Tue, 21 Nov 2023 14:36:16 +0000
+
+libserverframe (1.2.0-2) unstable; urgency=medium
+
+  * upgrade to 1.2.0-2
+
+ -- YuQing <384681@qq.com>  Mon, 20 Nov 2023 13:24:02 +0000
+
+libserverframe (1.2.0-1) unstable; urgency=medium
+
+  * upgrade to 1.2.0-1
+
+ -- YuQing <384681@qq.com>  Sun, 19 Nov 2023 14:46:16 +0000
+
+libserverframe (1.1.29-1) unstable; urgency=medium
+
+  * upgrade to 1.1.29-1
+
+ -- YuQing <384681@qq.com>  Sun, 06 Aug 2023 07:22:46 +0000
+
+libserverframe (1.1.28-1) unstable; urgency=medium
+
+  * upgrade to 1.1.28-1
+
+ -- YuQing <384681@qq.com>  Sun, 23 Jul 2023 14:28:20 +0000
+
+libserverframe (1.1.27-1) unstable; urgency=medium
+
+  * upgrade to 1.1.27-1
+
+ -- YuQing <384681@qq.com>  Sat, 24 Jun 2023 06:51:30 +0000
+
+libserverframe (1.1.26-1) unstable; urgency=medium
+
+  * upgrade to 1.1.26-1
+
+ -- YuQing <384681@qq.com>  Sun, 04 Jun 2023 10:52:11 +0000
+
+libserverframe (1.1.25-1) unstable; urgency=medium
+
+  * upgrade to 1.1.25-1
+
+ -- YuQing <384681@qq.com>  Sat, 18 Feb 2023 05:44:50 +0000
+
+libserverframe (1.1.24-1) unstable; urgency=medium
+
+  * upgrade to 1.1.24-1
+
+ -- YuQing <384681@qq.com>  Sun, 15 Jan 2023 13:50:15 +0000
+
+libserverframe (1.1.22-1) unstable; urgency=medium
+
+  * upgrade to 1.1.22-1
+
+ -- YuQing <384681@qq.com>  Mon, 21 Nov 2022 14:55:50 +0000
+
+libserverframe (1.1.21-1) unstable; urgency=medium
+
+  * upgrade to 1.1.21-1
+
+ -- YuQing <384681@qq.com>  Sat, 08 Oct 2022 13:28:40 +0000
+
+libserverframe (1.1.20-1) unstable; urgency=medium
+
+  * upgrade to 1.1.20-1
+
+ -- YuQing <384681@qq.com>  Thu, 22 Sep 2022 12:22:39 +0000
+
+libserverframe (1.1.18-1) unstable; urgency=medium
+
+  * upgrade to 1.1.18-1
+
+ -- YuQing <384681@qq.com>  Wed, 07 Sep 2022 13:36:38 +0000
+
+libserverframe (1.1.17-1) unstable; urgency=medium
+
+  * upgrade to 1.1.17-1
+
+ -- YuQing <384681@qq.com>  Mon, 15 Aug 2022 13:31:54 +0000
+
+libserverframe (1.1.16-1) unstable; urgency=medium
+
+  * upgrade to 1.1.16-1
+
+ -- YuQing <384681@qq.com>  Mon, 25 Jul 2022 13:52:09 +0000
+
+libserverframe (1.1.15-1) unstable; urgency=medium
+
+  * upgrade to 1.1.15-1
+
+ -- YuQing <384681@qq.com>  Wed, 15 Jun 2022 14:26:27 +0000
+
+libserverframe (1.1.14-1) unstable; urgency=medium
+
+  * upgrade to 1.1.14-1
+
+ -- YuQing <384681@qq.com>  Thu, 28 Apr 2022 11:54:26 +0000
+
+libserverframe (1.1.13-1) unstable; urgency=medium
+
+  * add files: sf_iov.[hc]
+  * iova_slice error detect
+  * sf_iova_memset impl.
+  * sf_iova_memset_ex for iov and iovcnt
+  * sf_iova_memset_ex: add const modifier
+  * sf_iov.[hc] add function sf_iova_memcpy_ex
+  * simple_hash rename to fc_simple_hash
+  * change log level to debug
+  * add function sf_load_global_base_path
+  * support function sf_sharding_htable_delete
+  * sf_load_global_config_ex: server_name can be NULL
+  * upgrade version to 1.1.13
+  * make.sh: change DEBUG_FLAG to 0
+  * add function sf_binlog_writer_get_index_filename
+
+ -- YuQing <384681@qq.com>  Sun, 13 Mar 2022 16:46:17 +0800
+
+libserverframe (1.1.12-1) unstable; urgency=medium
+
+  * upgrade version to 1.1.12
+
+ -- YuQing <384681@qq.com>  Sat, 15 Jan 2022 20:00:21 +0800
+
+libserverframe (1.1.11-1) unstable; urgency=medium
+
+  * upgrade version to 1.1.11
+
+ --  YuQing <384681@qq.com>  Sun, 26 Dec 2021 21:02:05 +0800
+
 libserverframe (1.1.10-1) unstable; urgency=medium
 
   * fixed somebugs

debian/control

@@ -3,7 +3,7 @@ Section: net
 Priority: optional
 Maintainer: YuQing <384681@qq.com>
 Build-Depends: debhelper (>=11~)
- , libfastcommon-dev (>= 1.0.52)
+ , libfastcommon-dev (>= 1.0.56)
 Standards-Version: 4.1.4
 Homepage: https://github.com/happyfish100/libserverframe
 
@@ -12,7 +12,7 @@ Section: net
 Architecture: any
 Multi-Arch: same
 Pre-Depends: ${misc:Pre-Depends}
-Depends: ${misc:Depends}, ${shlibs:Depends}, libserverframe (= ${binary:Version})
+Depends: ${misc:Depends}, libserverframe (= ${binary:Version})
 Description: libserverframe (development files)
   This package contains header files.
 
@@ -21,5 +21,5 @@ Section: net
 Architecture: any
 Multi-Arch: same
 Pre-Depends: ${misc:Pre-Depends}
-Depends: ${misc:Depends}, ${shlibs:Depends}, libfastcommon (>= 1.0.52)
+Depends: ${misc:Depends}, ${shlibs:Depends}, libfastcommon (>= ${libfastcommon:Version})
 Description: this network service framework library extract from FastDFS

debian/copyright

@@ -2,45 +2,7 @@ Format: https://www.debian.org/doc/packaging-manuals/copyright-format/1.0/
 Upstream-Name: libserverframe
 Source: https://github.com/happyfish100/libserverframe
 
-Files:     src/idempotency/client/client_channel.c
+Files:     *
-           src/idempotency/client/client_channel.h
-           src/idempotency/client/client_types.h
-           src/idempotency/client/receipt_handler.c
-           src/idempotency/client/receipt_handler.h
-           src/idempotency/client/rpc_wrapper.h
-           src/idempotency/server/channel_htable.c
-           src/idempotency/server/channel_htable.h
-           src/idempotency/server/request_htable.c
-           src/idempotency/server/request_htable.h
-           src/idempotency/server/server_channel.c
-           src/idempotency/server/server_channel.h
-           src/idempotency/server/server_handler.c
-           src/idempotency/server/server_handler.h
-           src/idempotency/server/server_types.h
-           src/sf_binlog_writer.c
-           src/sf_binlog_writer.h
-           src/sf_cluster_cfg.c
-           src/sf_cluster_cfg.h
-           src/sf_configs.c
-           src/sf_configs.h
-           src/sf_connection_manager.c
-           src/sf_connection_manager.h
-           src/sf_define.h
-           src/sf_func.c
-           src/sf_func.h
-           src/sf_global.c
-           src/sf_global.h
-           src/sf_nio.c
-           src/sf_nio.h
-           src/sf_proto.c
-           src/sf_proto.h
-           src/sf_service.c
-           src/sf_service.h
-           src/sf_sharding_htable.c
-           src/sf_sharding_htable.h
-           src/sf_types.h
-           src/sf_util.c
-           src/sf_util.h
 Copyright: 2020 YuQing <384681@qq.com>
 License:   AGPL-3.0+
  This program is free software: you can use, redistribute, and/or modify
@@ -54,39 +16,6 @@ License:   AGPL-3.0+
  You should have received a copy of the GNU Affero General Public License
  along with this program. If not, see <https://www.gnu.org/licenses/>.
 
-Files:     .gitignore
-           README.md
-           libserverframe.spec
-           make.sh
-           sample.conf
-           src/Makefile
-           src/Makefile.in
-           src/idempotency/client/client_channel.lo
-           src/idempotency/client/receipt_handler.lo
-           src/idempotency/server/channel_htable.lo
-           src/idempotency/server/request_htable.lo
-           src/idempotency/server/server_channel.lo
-           src/idempotency/server/server_handler.lo
-           src/libserverframe.so
-           src/sf_binlog_writer.lo
-           src/sf_cluster_cfg.lo
-           src/sf_configs.lo
-           src/sf_connection_manager.lo
-           src/sf_func.lo
-           src/sf_global.lo
-           src/sf_nio.lo
-           src/sf_proto.lo
-           src/sf_service.lo
-           src/sf_sharding_htable.lo
-           src/sf_util.lo
-Copyright: __NO_COPYRIGHT_NOR_LICENSE__
-License:   __NO_COPYRIGHT_NOR_LICENSE__
-
-#----------------------------------------------------------------------------
-# Files marked as NO_LICENSE_TEXT_FOUND may be covered by the following
-# license/copyright files.
-
-#----------------------------------------------------------------------------
 # License file: LICENSE
                      GNU AFFERO GENERAL PUBLIC LICENSE
                         Version 3, 19 November 2007

debian/libserverframe.install

@@ -1 +1 @@
-usr/lib64/libserverframe.so* usr/lib/
+usr/lib/libserverframe.so*

debian/rules

@@ -11,3 +11,7 @@ override_dh_auto_build:
 override_dh_auto_install:
 	./make.sh install
 	dh_auto_install
+
+.PHONY: override_dh_gencontrol
+override_dh_gencontrol:
+	dh_gencontrol -- -Tdebian/substvars

debian/substvars

@@ -0,0 +1 @@
+libfastcommon:Version=1.0.83

libserverframe.spec

@@ -2,7 +2,7 @@
 %define CommitVersion %(echo $COMMIT_VERSION)
 
 Name: libserverframe
-Version: 1.1.11
+Version: 1.2.11
 Release: 1%{?dist}
 Summary: network framework library
 License: AGPL v3.0
@@ -12,9 +12,9 @@ Source: http://github.com/happyfish100/libserverframe/%{name}-%{version}.tar.gz
 
 BuildRoot: %{_tmppath}/%{name}-%{version}-%{release}-root-%(%{__id_u} -n)
 
-BuildRequires: libfastcommon-devel >= 1.0.54
+BuildRequires: libfastcommon-devel >= 1.0.83
 Requires: %__cp %__mv %__chmod %__grep %__mkdir %__install %__id
-Requires: libfastcommon >= 1.0.54
+Requires: libfastcommon >= 1.0.83
 
 %description
 common framework library 

make.sh

@@ -5,18 +5,32 @@ TARGET_PREFIX=$DESTDIR/usr
 TARGET_CONF_PATH=$DESTDIR/etc
 
 LIB_VERSION=lib64
-DEBUG_FLAG=1
+DEBUG_FLAG=0
 
 if [ -f /usr/include/fastcommon/_os_define.h ]; then
-  OS_BITS=$(fgrep OS_BITS /usr/include/fastcommon/_os_define.h | awk '{print $NF;}')
+  OS_BITS=$(grep -F OS_BITS /usr/include/fastcommon/_os_define.h | awk '{print $NF;}')
+  USE_URING=$(grep -F IOEVENT_USE_URING /usr/include/fastcommon/_os_define.h | awk '{print $NF;}')
 elif [ -f /usr/local/include/fastcommon/_os_define.h ]; then
-  OS_BITS=$(fgrep OS_BITS /usr/local/include/fastcommon/_os_define.h | awk '{print $NF;}')
+  OS_BITS=$(grep -F OS_BITS /usr/local/include/fastcommon/_os_define.h | awk '{print $NF;}')
+  USE_URING=$(grep -F IOEVENT_USE_URING /usr/local/include/fastcommon/_os_define.h | awk '{print $NF;}')
 else
   OS_BITS=64
+  USE_URING=''
 fi
 
+uname=$(uname)
 if [ "$OS_BITS" -eq 64 ]; then
+  if [ $uname = 'Linux' ]; then
+    osname=$(cat /etc/os-release | grep -w NAME | awk -F '=' '{print $2;}' | \
+            awk -F '"' '{if (NF==3) {print $2} else {print $1}}' | awk '{print $1}')
+    if [ $osname = 'Ubuntu' -o $osname = 'Debian' ]; then
+      LIB_VERSION=lib
+    else
       LIB_VERSION=lib64
+    fi
+  else
+    LIB_VERSION=lib
+  fi
 else
   LIB_VERSION=lib
 fi
@@ -31,13 +45,16 @@ CFLAGS="$CFLAGS -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE"
 if [ "$DEBUG_FLAG" = "1" ]; then
   CFLAGS="$CFLAGS -g -O1 -DDEBUG_FLAG"
 else
-  CFLAGS="$CFLAGS -O3"
+  CFLAGS="$CFLAGS -g -O3"
 fi
 
 LIBS=''
 uname=$(uname)
 if [ "$uname" = "Linux" ]; then
   CFLAGS="$CFLAGS"
+  if [ -n "$USE_URING" ]; then
+    LIBS="$LIBS -luring"
+  fi
 elif [ "$uname" = "FreeBSD" ] || [ "$uname" = "Darwin" ]; then
   CFLAGS="$CFLAGS"
   if [ "$uname" = "Darwin" ]; then
@@ -66,19 +83,19 @@ elif [ "$uname" = "HP-UX" ]; then
   fi
 elif [ "$uname" = "FreeBSD" ]; then
   if [ -f /usr/lib/libc_r.so ]; then
-    line=$(nm -D /usr/lib/libc_r.so | grep pthread_create | grep -w T)
+    line=$(nm -D /usr/lib/libc_r.so | grep -F pthread_create | grep -w T)
     if [ $? -eq 0 ]; then
       LIBS="$LIBS -lc_r"
       have_pthread=1
     fi
   elif [ -f /lib64/libc_r.so ]; then
-    line=$(nm -D /lib64/libc_r.so | grep pthread_create | grep -w T)
+    line=$(nm -D /lib64/libc_r.so | grep -F pthread_create | grep -w T)
     if [ $? -eq 0 ]; then
       LIBS="$LIBS -lc_r"
       have_pthread=1
     fi
   elif [ -f /usr/lib64/libc_r.so ]; then
-    line=$(nm -D /usr/lib64/libc_r.so | grep pthread_create | grep -w T)
+    line=$(nm -D /usr/lib64/libc_r.so | grep -F pthread_create | grep -w T)
     if [ $? -eq 0 ]; then
       LIBS="$LIBS -lc_r"
       have_pthread=1
@@ -87,7 +104,7 @@ elif [ "$uname" = "FreeBSD" ]; then
 fi
 
 if [ $have_pthread -eq 0 ] && [ "$uname" = "Linux" ]; then
-   /sbin/ldconfig -p | fgrep libpthread.so > /dev/null
+   /sbin/ldconfig -p | grep -w libpthread.so > /dev/null
    if [ $? -eq 0 ]; then
       LIBS="$LIBS -lpthread"
    else
@@ -113,8 +130,15 @@ sed_replace()
     fi
 }
 
-cd src
+cd src/include
+link=$(readlink sf)
+if [ $? -ne 0 ] || [ "$link" != '..' -a "$link" != '../' ]; then
+  ln -sf .. sf
+fi
+
+cd ..
 cp Makefile.in Makefile
+sed_replace "s#\\\$(CC)#gcc#g" Makefile
 sed_replace "s#\\\$(CFLAGS)#$CFLAGS#g" Makefile
 sed_replace "s#\\\$(LIBS)#$LIBS#g" Makefile
 sed_replace "s#\\\$(TARGET_PREFIX)#$TARGET_PREFIX#g" Makefile

src/Makefile.in

@@ -6,16 +6,20 @@ LIB_PATH = $(LIBS) -lfastcommon
 TARGET_LIB = $(TARGET_PREFIX)/$(LIB_VERSION)
 
 TOP_HEADERS = sf_types.h sf_global.h sf_define.h sf_nio.h sf_service.h \
-              sf_func.h sf_util.h sf_configs.h sf_proto.h sf_binlog_writer.h \
+              sf_func.h sf_util.h sf_configs.h sf_proto.h sf_cluster_cfg.h \
-              sf_cluster_cfg.h sf_sharding_htable.h sf_connection_manager.h  \
+              sf_sharding_htable.h sf_connection_manager.h sf_serializer.h \
-              sf_serializer.h sf_binlog_index.h sf_file_writer.h \
+              sf_binlog_index.h sf_file_writer.h sf_binlog_writer.h \
-              sf_ordered_writer.h
+              sf_ordered_writer.h sf_buffered_writer.h sf_iov.h \
+              sf_shared_mbuffer.h
+
+IDEMP_COMMON_HEADER = idempotency/common/idempotency_types.h
 
 IDEMP_SERVER_HEADER = idempotency/server/server_types.h \
                       idempotency/server/server_channel.h  \
                       idempotency/server/request_htable.h  \
                       idempotency/server/channel_htable.h  \
-                      idempotency/server/server_handler.h
+                      idempotency/server/server_handler.h  \
+                      idempotency/server/request_metadata.h
 
 IDEMP_CLIENT_HEADER = idempotency/client/client_types.h    \
                       idempotency/client/receipt_handler.h \
@@ -24,16 +28,18 @@ IDEMP_CLIENT_HEADER = idempotency/client/client_types.h    \
 
 ALL_HEADERS  = $(TOP_HEADERS) $(IDEMP_SERVER_HEADER) $(IDEMP_CLIENT_HEADER)
 
-SHARED_OBJS = sf_nio.lo sf_service.lo sf_global.lo \
+SHARED_OBJS = sf_nio.lo sf_iov.lo sf_service.lo sf_global.lo \
         sf_func.lo sf_util.lo sf_configs.lo sf_proto.lo \
-        sf_binlog_writer.lo sf_sharding_htable.lo  \
+        sf_sharding_htable.lo sf_cluster_cfg.lo   \
-        sf_cluster_cfg.lo sf_connection_manager.lo \
+        sf_connection_manager.lo sf_serializer.lo \
-        sf_serializer.lo  sf_binlog_index.lo   \
+        sf_binlog_index.lo sf_file_writer.lo  \
-        sf_file_writer.lo sf_ordered_writer.lo \
+        sf_binlog_writer.lo sf_ordered_writer.lo \
+        sf_shared_mbuffer.lo \
         idempotency/server/server_channel.lo   \
         idempotency/server/request_htable.lo   \
         idempotency/server/channel_htable.lo   \
         idempotency/server/server_handler.lo   \
+        idempotency/server/request_metadata.lo \
         idempotency/client/receipt_handler.lo  \
         idempotency/client/client_channel.lo
 
@@ -55,16 +61,18 @@ libserverframe.so: $(SHARED_OBJS)
 install:
 	mkdir -p $(TARGET_LIB)
 	mkdir -p $(TARGET_PREFIX)/lib
+	mkdir -p $(TARGET_PREFIX)/include/sf/idempotency/common
 	mkdir -p $(TARGET_PREFIX)/include/sf/idempotency/server
 	mkdir -p $(TARGET_PREFIX)/include/sf/idempotency/client
 
 	install -m 755 $(ALL_LIBS) $(TARGET_LIB)
 	cp -f $(TOP_HEADERS) $(TARGET_PREFIX)/include/sf
+	cp -f $(IDEMP_COMMON_HEADER) $(TARGET_PREFIX)/include/sf/idempotency/common
 	cp -f $(IDEMP_SERVER_HEADER) $(TARGET_PREFIX)/include/sf/idempotency/server
 	cp -f $(IDEMP_CLIENT_HEADER) $(TARGET_PREFIX)/include/sf/idempotency/client
 
 	@BUILDROOT=$$(echo "$(TARGET_PREFIX)" | grep BUILDROOT); \
-	if [ -z "$$BUILDROOT" ] && [ ! -e $(TARGET_PREFIX)/lib/libserverframe.so ]; then ln -s $(TARGET_LIB)/libserverframe.so $(TARGET_PREFIX)/lib/libserverframe.so; fi
+	if [ -z "$$BUILDROOT" ] && [ "$(TARGET_LIB)" != "$(TARGET_PREFIX)/lib" ]; then ln -sf $(TARGET_LIB)/libserverframe.so $(TARGET_PREFIX)/lib/libserverframe.so; fi
 clean:
 	rm -f $(ALL_OBJS) $(ALL_LIBS) $(ALL_PRGS)
 

src/idempotency/client/client_channel.c

@@ -31,11 +31,11 @@
 #include "fastcommon/pthread_func.h"
 #include "fastcommon/sched_thread.h"
 #include "fastcommon/fc_queue.h"
-#include "../../sf_util.h"
+#include "sf/sf_util.h"
-#include "../../sf_func.h"
+#include "sf/sf_func.h"
-#include "../../sf_nio.h"
+#include "sf/sf_nio.h"
-#include "../../sf_global.h"
+#include "sf/sf_global.h"
-#include "../../sf_service.h"
+#include "sf/sf_service.h"
 #include "client_channel.h"
 
 typedef struct {
@@ -135,7 +135,7 @@ static int idempotency_channel_alloc_init(void *element, void *args)
         return result;
     }
 
-    if ((result=init_pthread_lock_cond_pair(&channel->lc_pair)) != 0) {
+    if ((result=init_pthread_lock_cond_pair(&channel->lcp)) != 0) {
         return result;
     }
 
@@ -171,17 +171,30 @@ void client_channel_destroy()
 }
 
 static struct fast_task_info *alloc_channel_task(IdempotencyClientChannel
-        *channel, const uint32_t hash_code, const char *server_ip,
+        *channel, const uint32_t hash_code, const FCCommunicationType comm_type,
-        const uint16_t port, int *err_no)
+        const char *server_ip, const uint16_t port, int *err_no)
 {
     struct fast_task_info *task;
+    SFAddressFamilyHandler *fh;
+    SFNetworkHandler *handler;
 
-    if ((task=sf_alloc_init_task(&g_sf_context, -1)) == NULL) {
+    if (is_ipv6_addr(server_ip)) {
+        fh = g_sf_context.handlers + SF_IPV6_ADDRESS_FAMILY_INDEX;
+    } else {
+        fh = g_sf_context.handlers + SF_IPV4_ADDRESS_FAMILY_INDEX;
+    }
+
+    if (comm_type == fc_comm_type_sock) {
+        handler = fh->handlers + SF_SOCKET_NETWORK_HANDLER_INDEX;
+    } else {
+        handler = fh->handlers + SF_RDMACM_NETWORK_HANDLER_INDEX;
+    }
+    if ((task=sf_alloc_init_task(handler, -1)) == NULL) {
         *err_no = ENOMEM;
         return NULL;
     }
 
-    snprintf(task->server_ip, sizeof(task->server_ip), "%s", server_ip);
+    fc_safe_strcpy(task->server_ip, server_ip);
     task->port = port;
     task->arg = channel;
     task->thread_data = g_sf_context.thread_data +
@@ -190,7 +203,8 @@ static struct fast_task_info *alloc_channel_task(IdempotencyClientChannel
     channel->last_connect_time = g_current_time;
     if ((*err_no=sf_nio_notify(task, SF_NIO_STAGE_CONNECT)) != 0) {
         channel->in_ioevent = 0;   //rollback
-        sf_release_task(task);
+        __sync_sub_and_fetch(&task->reffer_count, 1);
+        free_queue_push(task);
         return NULL;
     }
     return task;
@@ -200,6 +214,15 @@ int idempotency_client_channel_check_reconnect(
         IdempotencyClientChannel *channel)
 {
     int result;
+    char formatted_ip[FORMATTED_IP_SIZE];
+
+#if IOEVENT_USE_URING
+    struct fast_task_info *task;
+    task = channel->task;
+    if (SF_CTX->use_io_uring && FC_ATOMIC_GET(task->reffer_count) > 1) {
+        return 0;
+    }
+#endif
 
     if (!__sync_bool_compare_and_swap(&channel->in_ioevent, 0, 1)) {
         return 0;
@@ -210,11 +233,16 @@ int idempotency_client_channel_check_reconnect(
         channel->last_connect_time = g_current_time;
     }
 
+    if (FC_LOG_BY_LEVEL(LOG_DEBUG)) {
+        format_ip_address(channel->task->server_ip, formatted_ip);
         logDebug("file: "__FILE__", line: %d, "
-            "trigger connect to server %s:%u",
+                "trigger connect to server %s:%u", __LINE__,
-            __LINE__, channel->task->server_ip,
+                formatted_ip, channel->task->port);
-            channel->task->port);
+    }
 
+    if (channel->task->event.fd >= 0) {
+        channel->task->handler->close_connection(channel->task);
+    }
     __sync_bool_compare_and_swap(&channel->task->canceled, 1, 0);
     if ((result=sf_nio_notify(channel->task, SF_NIO_STAGE_CONNECT)) == 0) {
         channel->last_connect_time = g_current_time;
@@ -226,8 +254,8 @@ int idempotency_client_channel_check_reconnect(
 }
 
 struct idempotency_client_channel *idempotency_client_channel_get(
-        const char *server_ip, const uint16_t server_port,
+        const FCCommunicationType comm_type, const char *server_ip,
-        const int timeout, int *err_no)
+        const uint16_t server_port, const int timeout, int *err_no)
 {
     int r;
     int key_len;
@@ -239,8 +267,11 @@ struct idempotency_client_channel *idempotency_client_channel_get(
     IdempotencyClientChannel *current;
     IdempotencyClientChannel *channel;
 
-    key_len = snprintf(key, sizeof(key), "%s_%u", server_ip, server_port);
+    key_len = strlen(server_ip);
-    hash_code = simple_hash(key, key_len);
+    memcpy(key, server_ip, key_len);
+    *(key + key_len++) = '-';
+    key_len += fc_itoa(server_port, key + key_len);
+    hash_code = fc_simple_hash(key, key_len);
     bucket = channel_context.htable.buckets +
         hash_code % channel_context.htable.capacity;
     previous = NULL;
@@ -277,8 +308,8 @@ struct idempotency_client_channel *idempotency_client_channel_get(
             break;
         }
 
-        channel->task = alloc_channel_task(channel,
+        channel->task = alloc_channel_task(channel, hash_code,
-                hash_code, server_ip, server_port, err_no);
+                comm_type, server_ip, server_port, err_no);
         if (channel->task == NULL) {
             fast_mblock_free_object(&channel_context.
                     channel_allocator, channel);
@@ -323,8 +354,8 @@ int idempotency_client_channel_push(struct idempotency_client_channel *channel,
     receipt->req_id = req_id;
     fc_queue_push_ex(&channel->queue, receipt, &notify);
     if (notify) {
-        if (__sync_add_and_fetch(&channel->in_ioevent, 0)) {
+        if (FC_ATOMIC_GET(channel->in_ioevent)) {
-            if (__sync_add_and_fetch(&channel->established, 0)) {
+            if (FC_ATOMIC_GET(channel->established)) {
                 sf_nio_notify(channel->task, SF_NIO_STAGE_CONTINUE);
             }
         } else {

src/idempotency/client/client_channel.h

@@ -22,6 +22,7 @@
 #include "fastcommon/pthread_func.h"
 #include "fastcommon/sched_thread.h"
 #include "fastcommon/fc_atomic.h"
+#include "sf/sf_types.h"
 #include "client_types.h"
 
 #ifdef __cplusplus
@@ -40,13 +41,14 @@ void idempotency_client_channel_config_to_string_ex(
         char *output, const int size, const bool add_comma);
 
 struct idempotency_client_channel *idempotency_client_channel_get(
-        const char *server_ip, const uint16_t server_port,
+        const FCCommunicationType comm_type, const char *server_ip,
-        const int timeout, int *err_no);
+        const uint16_t server_port, const int timeout, int *err_no);
 
 static inline uint64_t idempotency_client_channel_next_seq_id(
         struct idempotency_client_channel *channel)
 {
-    return __sync_add_and_fetch(&channel->next_req_id, 1);
+    return SF_IDEMPOTENCY_NEXT_REQ_ID(channel->server_id,
+            channel->id, FC_ATOMIC_INC(channel->next_seq));
 }
 
 int idempotency_client_channel_push(struct idempotency_client_channel *channel,
@@ -74,13 +76,28 @@ static inline void idempotency_client_channel_set_id_key(
 static inline int idempotency_client_channel_check_wait_ex(
         struct idempotency_client_channel *channel, const int timeout)
 {
-    if (__sync_add_and_fetch(&channel->established, 0)) {
+    if (FC_ATOMIC_GET(channel->established)) {
         return 0;
     }
 
     idempotency_client_channel_check_reconnect(channel);
-    lcp_timedwait_sec(&channel->lc_pair, timeout);
+    lcp_timedwait_sec(&channel->lcp, timeout);
-    return __sync_add_and_fetch(&channel->established, 0) ? 0 : ETIMEDOUT;
+    if (FC_ATOMIC_GET(channel->established)) {
+        return 0;
+    } else {
+        /*
+        char formatted_ip[FORMATTED_IP_SIZE];
+        format_ip_address(channel->task->server_ip, formatted_ip);
+        logInfo("file: "__FILE__", line: %d, "
+                "channel_check fail, server %s:%u, in_ioevent: %d, "
+                "canceled: %d, req count: %"PRId64, __LINE__,
+                formatted_ip, channel->task->port,
+                __sync_add_and_fetch(&channel->in_ioevent, 0),
+                __sync_add_and_fetch(&channel->task->canceled, 0),
+                channel->task->req_count);
+         */
+        return ETIMEDOUT;
+    }
 }
 
 #ifdef __cplusplus

src/idempotency/client/client_types.h

@@ -21,6 +21,7 @@
 #include "fastcommon/fast_mblock.h"
 #include "fastcommon/fc_list.h"
 #include "fastcommon/fc_queue.h"
+#include "sf/idempotency/common/idempotency_types.h"
 
 typedef struct idempotency_client_config {
     bool enabled;
@@ -40,11 +41,12 @@ typedef struct idempotency_client_channel {
     volatile char in_ioevent;
     volatile char established;
     int buffer_size;  //the min task size of the server and mine
+    uint32_t server_id;
+    volatile uint32_t next_seq;
     time_t last_connect_time;  //for connect frequency control
     time_t last_pkg_time;      //last communication time
     time_t last_report_time;   //last report time for rpc receipt
-    pthread_lock_cond_pair_t lc_pair;  //for channel valid check and notify
+    pthread_lock_cond_pair_t lcp;  //for channel valid check and notify
-    volatile uint64_t next_req_id;
     struct fast_mblock_man receipt_allocator;
     struct fast_task_info *task;
     struct fc_queue queue;
@@ -61,6 +63,14 @@ typedef struct idempotency_receipt_thread_context {
     } last_check_times;
 } IdempotencyReceiptThreadContext;
 
+typedef struct idempotency_receipt_global_vars {
+    struct {
+        int task_padding_size;
+        sf_init_connection_callback init_connection;
+    } rdma;
+    IdempotencyReceiptThreadContext *thread_contexts;
+} IdempotencyReceiptGlobalVars;
+
 #ifdef __cplusplus
 extern "C" {
 #endif

src/idempotency/client/receipt_handler.c

@@ -41,23 +41,35 @@
 #include "client_channel.h"
 #include "receipt_handler.h"
 
-static IdempotencyReceiptThreadContext *receipt_thread_contexts = NULL;
+static IdempotencyReceiptGlobalVars receipt_global_vars;
 
-static int receipt_init_task(struct fast_task_info *task)
+#define RECEIPT_THREAD_CONTEXTS  receipt_global_vars.thread_contexts
+#define TASK_PADDING_SIZE        receipt_global_vars.rdma.task_padding_size
+#define RDMA_INIT_CONNECTION     receipt_global_vars.rdma.init_connection
+
+static int receipt_init_task(struct fast_task_info *task, void *arg)
 {
-    task->connect_timeout = SF_G_CONNECT_TIMEOUT; //for client side
+#if IOEVENT_USE_URING
-    task->network_timeout = SF_G_NETWORK_TIMEOUT;
+    FC_URING_IS_CLIENT(task) = true;
+#endif
+
+    if (RDMA_INIT_CONNECTION != NULL) {
+        return RDMA_INIT_CONNECTION(task, arg);
+    } else {
         return 0;
+    }
 }
 
 static int receipt_recv_timeout_callback(struct fast_task_info *task)
 {
     IdempotencyClientChannel *channel;
+    char formatted_ip[FORMATTED_IP_SIZE];
 
+    format_ip_address(task->server_ip, formatted_ip);
     if (SF_NIO_TASK_STAGE_FETCH(task) == SF_NIO_STAGE_CONNECT) {
         logError("file: "__FILE__", line: %d, "
                 "connect to server %s:%u timeout",
-                __LINE__, task->server_ip, task->port);
+                __LINE__, formatted_ip, task->port);
         return ETIMEDOUT;
     }
 
@@ -65,11 +77,13 @@ static int receipt_recv_timeout_callback(struct fast_task_info *task)
     if (channel->waiting_resp_qinfo.head != NULL) {
         logError("file: "__FILE__", line: %d, "
                 "waiting receipt response from server %s:%u timeout",
-                __LINE__, task->server_ip, task->port);
+                __LINE__, formatted_ip, task->port);
     } else {
         logError("file: "__FILE__", line: %d, "
-                "communication with server %s:%u timeout",
+                "%s server %s:%u timeout, channel established: %d",
-                __LINE__, task->server_ip, task->port);
+                __LINE__, task->nio_stages.current == SF_NIO_STAGE_SEND ?
+                "send to" : "recv from", formatted_ip, task->port,
+                FC_ATOMIC_GET(channel->established));
     }
 
     return ETIMEDOUT;
@@ -78,22 +92,27 @@ static int receipt_recv_timeout_callback(struct fast_task_info *task)
 static void receipt_task_finish_cleanup(struct fast_task_info *task)
 {
     IdempotencyClientChannel *channel;
+    char formatted_ip[FORMATTED_IP_SIZE];
 
     if (task->event.fd >= 0) {
         sf_task_detach_thread(task);
-        close(task->event.fd);
-        task->event.fd = -1;
     }
 
-    channel = (IdempotencyClientChannel *)task->arg;
+    sf_nio_reset_task_length(task);
+    task->req_count = 0;
+    task->pending_send_count = 0;
 
+    channel = (IdempotencyClientChannel *)task->arg;
     fc_list_del_init(&channel->dlink);
     __sync_bool_compare_and_swap(&channel->established, 1, 0);
     __sync_bool_compare_and_swap(&channel->in_ioevent, 1, 0);
 
+    if (FC_LOG_BY_LEVEL(LOG_DEBUG)) {
+        format_ip_address(task->server_ip, formatted_ip);
         logDebug("file: "__FILE__", line: %d, "
                 "receipt task for server %s:%u exit",
-            __LINE__, task->server_ip, task->port);
+                __LINE__, formatted_ip, task->port);
+    }
 }
 
 static void setup_channel_request(struct fast_task_info *task)
@@ -103,14 +122,15 @@ static void setup_channel_request(struct fast_task_info *task)
     SFProtoSetupChannelReq *req;
 
     channel = (IdempotencyClientChannel *)task->arg;
-    header = (SFCommonProtoHeader *)task->data;
+    header = (SFCommonProtoHeader *)task->send.ptr->data;
     req = (SFProtoSetupChannelReq *)(header + 1);
     int2buff(__sync_add_and_fetch(&channel->id, 0), req->channel_id);
     int2buff(__sync_add_and_fetch(&channel->key, 0), req->key);
 
     SF_PROTO_SET_HEADER(header, SF_SERVICE_PROTO_SETUP_CHANNEL_REQ,
             sizeof(SFProtoSetupChannelReq));
-    task->length = sizeof(SFCommonProtoHeader) + sizeof(SFProtoSetupChannelReq);
+    task->send.ptr->length = sizeof(SFCommonProtoHeader) +
+        sizeof(SFProtoSetupChannelReq);
     sf_send_add_event(task);
 }
 
@@ -137,10 +157,10 @@ static int check_report_req_receipt(struct fast_task_info *task)
         return 0;
     }
 
-    header = (SFCommonProtoHeader *)task->data;
+    header = (SFCommonProtoHeader *)task->send.ptr->data;
     rheader = (SFProtoReportReqReceiptHeader *)(header + 1);
     rbody = rstart = (SFProtoReportReqReceiptBody *)(rheader + 1);
-    buff_end = task->data + channel->buffer_size;
+    buff_end = task->send.ptr->data + channel->buffer_size;
     last = NULL;
     receipt = channel->waiting_resp_qinfo.head;
     do {
@@ -170,8 +190,9 @@ static int check_report_req_receipt(struct fast_task_info *task)
 
     count = rbody - rstart;
     int2buff(count, rheader->count);
-    task->length = (char *)rbody - task->data;
+    task->send.ptr->length = (char *)rbody - task->send.ptr->data;
-    int2buff(task->length - sizeof(SFCommonProtoHeader), header->body_len);
+    int2buff(task->send.ptr->length - sizeof(SFCommonProtoHeader),
+            header->body_len);
     header->cmd = SF_SERVICE_PROTO_REPORT_REQ_RECEIPT_REQ;
     sf_send_add_event(task);
     return count;
@@ -185,18 +206,18 @@ static void close_channel_request(struct fast_task_info *task)
     channel = (IdempotencyClientChannel *)task->arg;
     idempotency_client_channel_set_id_key(channel, 0, 0);
 
-    header = (SFCommonProtoHeader *)task->data;
+    header = (SFCommonProtoHeader *)task->send.ptr->data;
     SF_PROTO_SET_HEADER(header, SF_SERVICE_PROTO_CLOSE_CHANNEL_REQ, 0);
-    task->length = sizeof(SFCommonProtoHeader);
+    task->send.ptr->length = sizeof(SFCommonProtoHeader);
     sf_send_add_event(task);
 }
 
 static void active_test_request(struct fast_task_info *task)
 {
     SFCommonProtoHeader *header;
-    header = (SFCommonProtoHeader *)task->data;
+    header = (SFCommonProtoHeader *)task->send.ptr->data;
     SF_PROTO_SET_HEADER(header, SF_PROTO_ACTIVE_TEST_REQ, 0);
-    task->length = sizeof(SFCommonProtoHeader);
+    task->send.ptr->length = sizeof(SFCommonProtoHeader);
     sf_send_add_event(task);
 }
 
@@ -224,17 +245,22 @@ static void report_req_receipt_request(struct fast_task_info *task,
         if (update_lru) {
             update_lru_chain(task);
         }
+        task->pending_send_count++;
     }
 }
 
 static inline int receipt_expect_body_length(struct fast_task_info *task,
         const int expect_body_len)
 {
-    if ((int)(task->length - sizeof(SFCommonProtoHeader)) != expect_body_len) {
+    int body_len;
+    char formatted_ip[FORMATTED_IP_SIZE];
+
+    body_len = task->recv.ptr->length - sizeof(SFCommonProtoHeader);
+    if (body_len != expect_body_len) {
+        format_ip_address(task->server_ip, formatted_ip);
         logError("file: "__FILE__", line: %d, "
-                "server %s:%u, response body length: %d != %d",
+                "server %s:%u, response body length: %d != %d", __LINE__,
-                __LINE__, task->server_ip, task->port, (int)(task->length -
+                formatted_ip, task->port, body_len, expect_body_len);
-                    sizeof(SFCommonProtoHeader)), expect_body_len);
         return EINVAL;
     }
 
@@ -247,6 +273,7 @@ static int deal_setup_channel_response(struct fast_task_info *task)
     IdempotencyReceiptThreadContext *thread_ctx;
     SFProtoSetupChannelResp *resp;
     IdempotencyClientChannel *channel;
+    char formatted_ip[FORMATTED_IP_SIZE];
     int channel_id;
     int channel_key;
     int buffer_size;
@@ -258,28 +285,30 @@ static int deal_setup_channel_response(struct fast_task_info *task)
     }
 
     channel = (IdempotencyClientChannel *)task->arg;
-    if (__sync_add_and_fetch(&channel->established, 0)) {
+    if (FC_ATOMIC_GET(channel->established)) {
+        format_ip_address(task->server_ip, formatted_ip);
         logWarning("file: "__FILE__", line: %d, "
                 "response from server %s:%u, unexpected cmd: "
                 "SETUP_CHANNEL_RESP, ignore it!",
-                __LINE__, task->server_ip, task->port);
+                __LINE__, formatted_ip, task->port);
         return 0;
     }
 
-    resp = (SFProtoSetupChannelResp *)(task->data + sizeof(SFCommonProtoHeader));
+    resp = (SFProtoSetupChannelResp *)SF_PROTO_RECV_BODY(task);
     channel_id = buff2int(resp->channel_id);
     channel_key = buff2int(resp->key);
     buffer_size = buff2int(resp->buffer_size);
+    channel->server_id = buff2int(resp->server_id);
     idempotency_client_channel_set_id_key(channel, channel_id, channel_key);
     if (__sync_bool_compare_and_swap(&channel->established, 0, 1)) {
         thread_ctx = (IdempotencyReceiptThreadContext *)task->thread_data->arg;
         fc_list_add_tail(&channel->dlink, &thread_ctx->head);
     }
-    channel->buffer_size = FC_MIN(buffer_size, task->size);
+    channel->buffer_size = FC_MIN(buffer_size, task->send.ptr->size);
 
-    PTHREAD_MUTEX_LOCK(&channel->lc_pair.lock);
+    PTHREAD_MUTEX_LOCK(&channel->lcp.lock);
-    pthread_cond_broadcast(&channel->lc_pair.cond);
+    pthread_cond_broadcast(&channel->lcp.cond);
-    PTHREAD_MUTEX_UNLOCK(&channel->lc_pair.lock);
+    PTHREAD_MUTEX_UNLOCK(&channel->lcp.lock);
 
     if (channel->waiting_resp_qinfo.head != NULL) {
         bool notify;
@@ -298,6 +327,7 @@ static inline int deal_report_req_receipt_response(struct fast_task_info *task)
     IdempotencyClientChannel *channel;
     IdempotencyClientReceipt *current;
     IdempotencyClientReceipt *deleted;
+    char formatted_ip[FORMATTED_IP_SIZE];
 
     if ((result=receipt_expect_body_length(task, 0)) != 0) {
         return result;
@@ -305,13 +335,15 @@ static inline int deal_report_req_receipt_response(struct fast_task_info *task)
 
     channel = (IdempotencyClientChannel *)task->arg;
     if (channel->waiting_resp_qinfo.head == NULL) {
+        format_ip_address(task->server_ip, formatted_ip);
         logWarning("file: "__FILE__", line: %d, "
                 "response from server %s:%u, unexpect cmd: "
                 "REPORT_REQ_RECEIPT_RESP", __LINE__,
-                task->server_ip, task->port);
+                formatted_ip, task->port);
-        return 0;
+        return EINVAL;
     }
 
+    task->pending_send_count--;
     current = channel->waiting_resp_qinfo.head;
     do {
         deleted = current;
@@ -328,40 +360,46 @@ static inline int deal_report_req_receipt_response(struct fast_task_info *task)
 static int receipt_deal_task(struct fast_task_info *task, const int stage)
 {
     int result;
+    SFCommonProtoHeader *header;
+    char formatted_ip[FORMATTED_IP_SIZE];
 
     do {
         if (stage == SF_NIO_STAGE_HANDSHAKE) {
             setup_channel_request(task);
             result = 0;
             break;
-        } else if (stage == SF_NIO_STAGE_CONTINUE && task->length == 0) {
+        } else if (stage == SF_NIO_STAGE_CONTINUE) {
+            if (task->pending_send_count == 0) {
                 if (((IdempotencyClientChannel *)task->arg)->established) {
                     report_req_receipt_request(task, true);
-            } else {
+                } else if (task->req_count > 0) {
                     sf_set_read_event(task);  //trigger read event
                 }
+            }
+
             result = 0;
             break;
         }
 
-        result = buff2short(((SFCommonProtoHeader *)task->data)->status);
+        header = (SFCommonProtoHeader *)task->recv.ptr->data;
+        result = buff2short(header->status);
         if (result != 0) {
             int msg_len;
             char *message;
 
-            msg_len = task->length - sizeof(SFCommonProtoHeader);
+            msg_len = SF_RECV_BODY_LENGTH(task);
-            message = task->data + sizeof(SFCommonProtoHeader);
+            message = SF_PROTO_RECV_BODY(task);
+            format_ip_address(task->server_ip, formatted_ip);
             logError("file: "__FILE__", line: %d, "
                     "response from server %s:%u, cmd: %d (%s), "
-                    "status: %d, error info: %.*s",
+                    "status: %d, error info: %.*s", __LINE__,
-                    __LINE__, task->server_ip, task->port,
+                    formatted_ip, task->port, header->cmd,
-                    ((SFCommonProtoHeader *)task->data)->cmd,
+                    sf_get_cmd_caption(header->cmd),
-                    sf_get_cmd_caption(((SFCommonProtoHeader *)task->data)->cmd),
                     result, msg_len, message);
             break;
         }
 
-        switch (((SFCommonProtoHeader *)task->data)->cmd) {
+        switch (header->cmd) {
             case SF_SERVICE_PROTO_SETUP_CHANNEL_RESP:
                 result = deal_setup_channel_response(task);
                 break;
@@ -369,29 +407,36 @@ static int receipt_deal_task(struct fast_task_info *task, const int stage)
                 result = deal_report_req_receipt_response(task);
                 break;
             case SF_PROTO_ACTIVE_TEST_RESP:
+                task->pending_send_count--;
                 result = 0;
                 break;
             case SF_SERVICE_PROTO_CLOSE_CHANNEL_RESP:
                 result = ECONNRESET; //force to close socket
+                if (FC_LOG_BY_LEVEL(LOG_DEBUG)) {
+                    format_ip_address(task->server_ip, formatted_ip);
                     logDebug("file: "__FILE__", line: %d, "
                             "close channel to server %s:%u !!!",
-                    __LINE__, task->server_ip, task->port);
+                            __LINE__, formatted_ip, task->port);
+                }
                 break;
             default:
+                format_ip_address(task->server_ip, formatted_ip);
                 logError("file: "__FILE__", line: %d, "
                         "response from server %s:%u, unexpect cmd: %d (%s)",
-                        __LINE__, task->server_ip, task->port,
+                        __LINE__, formatted_ip, task->port, header->cmd,
-                        ((SFCommonProtoHeader *)task->data)->cmd,
+                        sf_get_cmd_caption(header->cmd));
-                        sf_get_cmd_caption(((SFCommonProtoHeader *)task->data)->cmd));
                 result = EINVAL;
                 break;
         }
 
         if (result == 0) {
             update_lru_chain(task);
-            task->offset = task->length = 0;
+            task->recv.ptr->length = 0;
+            task->recv.ptr->offset = 0;
+            if (task->pending_send_count == 0) {
                 report_req_receipt_request(task, false);
             }
+        }
     } while (0);
 
     return result > 0 ? -1 * result : result;
@@ -410,9 +455,10 @@ static void receipt_thread_check_heartbeat(
             break;
         }
 
-        if (sf_nio_task_is_idle(channel->task)) {
+        if (channel->task->pending_send_count == 0) {
             channel->last_pkg_time = g_current_time;
             active_test_request(channel->task);
+            channel->task->pending_send_count++;
         }
     }
 }
@@ -422,18 +468,22 @@ static void receipt_thread_close_idle_channel(
 {
     IdempotencyClientChannel *channel;
     IdempotencyClientChannel *tmp;
+    char formatted_ip[FORMATTED_IP_SIZE];
 
     fc_list_for_each_entry_safe(channel, tmp, &thread_ctx->head, dlink) {
-        if (!sf_nio_task_is_idle(channel->task)) {
+        if (channel->task->pending_send_count > 0) {
             continue;
         }
 
         if (g_current_time - channel->last_report_time >
                  g_idempotency_client_cfg.channel_max_idle_time)
         {
+            if (FC_LOG_BY_LEVEL(LOG_DEBUG)) {
+                format_ip_address(channel->task->server_ip, formatted_ip);
                 logDebug("file: "__FILE__", line: %d, "
                         "close channel to server %s:%u because idle too long",
-                    __LINE__, channel->task->server_ip, channel->task->port);
+                        __LINE__, formatted_ip, channel->task->port);
+            }
             close_channel_request(channel->task);
         }
     }
@@ -464,40 +514,61 @@ static void *receipt_alloc_thread_extra_data(const int thread_index)
 {
     IdempotencyReceiptThreadContext *ctx;
 
-    ctx = receipt_thread_contexts + thread_index;
+    ctx = RECEIPT_THREAD_CONTEXTS + thread_index;
     FC_INIT_LIST_HEAD(&ctx->head);
     return ctx;
 }
 
-static int do_init()
+static int do_init(FCAddressPtrArray *address_array)
 {
+    const int task_arg_size = 0;
+    const bool double_buffers = false;
+    const bool need_shrink_task_buffer = false;
+    const bool explicit_post_recv = false;
+    int result;
     int bytes;
+    SFNetworkHandler *rdma_handler;
+    struct ibv_pd *pd;
 
     bytes = sizeof(IdempotencyReceiptThreadContext) * SF_G_WORK_THREADS;
-    receipt_thread_contexts = (IdempotencyReceiptThreadContext *)
+    RECEIPT_THREAD_CONTEXTS = (IdempotencyReceiptThreadContext *)
         fc_malloc(bytes);
-    if (receipt_thread_contexts == NULL) {
+    if (RECEIPT_THREAD_CONTEXTS == NULL) {
         return ENOMEM;
     }
-    memset(receipt_thread_contexts, 0, bytes);
+    memset(RECEIPT_THREAD_CONTEXTS, 0, bytes);
 
+    if ((rdma_handler=sf_get_rdma_network_handler(&g_sf_context)) != NULL) {
+        if ((result=sf_alloc_rdma_pd(&g_sf_context, address_array)) != 0) {
+            return result;
+        }
+
+        TASK_PADDING_SIZE = rdma_handler->get_connection_size();
+        RDMA_INIT_CONNECTION = rdma_handler->init_connection;
+        pd = rdma_handler->pd;
+    } else {
+        TASK_PADDING_SIZE = 0;
+        RDMA_INIT_CONNECTION = NULL;
+        pd = NULL;
+    }
     return sf_service_init_ex2(&g_sf_context, "idemp-receipt",
             receipt_alloc_thread_extra_data, receipt_thread_loop_callback,
-            NULL, sf_proto_set_body_length, receipt_deal_task,
+            NULL, sf_proto_set_body_length, NULL, NULL, receipt_deal_task,
             receipt_task_finish_cleanup, receipt_recv_timeout_callback,
-            1000, sizeof(SFCommonProtoHeader), 0, receipt_init_task, NULL);
+            1000, sizeof(SFCommonProtoHeader), TASK_PADDING_SIZE,
+            task_arg_size, double_buffers, need_shrink_task_buffer,
+            explicit_post_recv, receipt_init_task, pd, NULL);
 }
 
-int receipt_handler_init()
+int receipt_handler_init(FCAddressPtrArray *address_array)
 {
     int result;
 
-    if ((result=do_init()) != 0) {
+    if ((result=do_init(address_array)) != 0) {
         return result;
     }
 
     sf_enable_thread_notify(true);
-    sf_set_remove_from_ready_list(false);
     fc_sleep_ms(100);
 
     return 0;

src/idempotency/client/receipt_handler.h

@@ -24,7 +24,7 @@
 extern "C" {
 #endif
 
-int receipt_handler_init();
+int receipt_handler_init(FCAddressPtrArray *address_array);
 int receipt_handler_destroy();
 
 #ifdef __cplusplus

src/idempotency/client/rpc_wrapper.h

@@ -44,7 +44,6 @@
             &client_ctx->common_cfg.net_retry_cfg.interval_mm,    \
             &client_ctx->common_cfg.net_retry_cfg.network);       \
     \
-    while (1) {  \
     if (idempotency_enabled) {   \
         req_id = idempotency_client_channel_next_seq_id(  \
                 connection_params->channel); \
@@ -52,6 +51,7 @@
         req_id = 0;  \
     }  \
     \
+    while (1) {  \
         old_channel = connection_params != NULL ? \
              connection_params->channel : NULL;   \
         i = 0; \
@@ -79,7 +79,8 @@
                             connection_params->channel) == 0) \
                 { \
                     if ((conn_result=sf_proto_rebind_idempotency_channel( \
-                                conn, connection_params->channel->id, \
+                                conn, (conn_manager)->module_name, \
+                                connection_params->channel->id,  \
                                 connection_params->channel->key, \
                                 client_ctx->common_cfg.network_timeout)) == 0)   \
                     { \

src/idempotency/common/idempotency_types.h

@@ -0,0 +1,46 @@
+/*
+ * Copyright (c) 2020 YuQing <384681@qq.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <https://www.gnu.org/licenses/>.
+ */
+
+
+#ifndef _IDEMPOTENCY_COMMON_TYPES_H
+#define _IDEMPOTENCY_COMMON_TYPES_H
+
+#include "fastcommon/common_define.h"
+
+#define SF_IDEMPOTENCY_CHANNEL_ID_BITS    16
+#define SF_IDEMPOTENCY_REQUEST_ID_BITS    (64 - SF_IDEMPOTENCY_CHANNEL_ID_BITS)
+#define SF_IDEMPOTENCY_MAX_CHANNEL_COUNT  ((1 << SF_IDEMPOTENCY_CHANNEL_ID_BITS) - 1)
+#define SF_IDEMPOTENCY_MAX_CHANNEL_ID     SF_IDEMPOTENCY_MAX_CHANNEL_COUNT
+
+#define SF_IDEMPOTENCY_SERVER_ID_OFFSET   48
+#define SF_IDEMPOTENCY_CHANNEL_ID_OFFSET  32
+
+#define SF_IDEMPOTENCY_NEXT_REQ_ID(server_id, channel_id, seq)    \
+    (((int64_t)server_id)  << SF_IDEMPOTENCY_SERVER_ID_OFFSET)  | \
+    (((int64_t)channel_id) << SF_IDEMPOTENCY_CHANNEL_ID_OFFSET) | \
+    (int64_t)seq
+
+#define SF_IDEMPOTENCY_EXTRACT_SERVER_ID(req_id) \
+    (int)((req_id >> SF_IDEMPOTENCY_SERVER_ID_OFFSET) & 0xFFFF)
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif

src/idempotency/server/request_metadata.c

@@ -0,0 +1,250 @@
+/*
+ * Copyright (c) 2020 YuQing <384681@qq.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <https://www.gnu.org/licenses/>.
+ */
+
+#include <limits.h>
+#include <sys/stat.h>
+#include "fastcommon/shared_func.h"
+#include "fastcommon/logger.h"
+#include "fastcommon/fc_atomic.h"
+#include "sf/sf_global.h"
+#include "request_metadata.h"
+
+static struct {
+    int process_interval_ms;
+    int master_side_timeout;  //in seconds
+    struct {
+        IdempotencyRequestMetadataContext *head;
+        IdempotencyRequestMetadataContext *tail;
+    } list;
+} g_request_metadata = {1000, 300, {NULL, NULL}};
+
+
+#define CHECK_MASTER_METADATA(meta) \
+    (meta != NULL && g_current_time - (long)meta->enqueue_time > \
+     g_request_metadata.master_side_timeout)
+
+static void process_master_side(IdempotencyRequestMetadataContext *ctx)
+{
+    struct fast_mblock_chain chain;
+    struct fast_mblock_node *node;
+
+    chain.head = chain.tail = NULL;
+    PTHREAD_MUTEX_LOCK(&ctx->lock);
+    if (CHECK_MASTER_METADATA(ctx->list.head)) {
+        do {
+            node = fast_mblock_to_node_ptr(ctx->list.head);
+            if (chain.head == NULL) {
+                chain.head = node;
+            } else {
+                chain.tail->next = node;
+            }
+            chain.tail = node;
+
+            ctx->list.head = ctx->list.head->next;
+        } while (CHECK_MASTER_METADATA(ctx->list.head));
+
+        if (ctx->list.head == NULL) {
+            ctx->list.tail = NULL;
+        }
+        chain.tail->next = NULL;
+    }
+
+    if (chain.head != NULL) {
+        fast_mblock_batch_free(&ctx->allocator, &chain);
+    }
+    PTHREAD_MUTEX_UNLOCK(&ctx->lock);
+}
+
+#define CHECK_SLAVE_METADATA(meta, dv) \
+    (meta != NULL && meta->data_version <= dv)
+
+static void process_slave_side(IdempotencyRequestMetadataContext *ctx,
+        const int64_t data_version)
+{
+    struct fast_mblock_chain chain;
+    struct fast_mblock_node *node;
+
+    chain.head = chain.tail = NULL;
+    PTHREAD_MUTEX_LOCK(&ctx->lock);
+    if (CHECK_SLAVE_METADATA(ctx->list.head, data_version)) {
+        do {
+            node = fast_mblock_to_node_ptr(ctx->list.head);
+            if (chain.head == NULL) {
+                chain.head = node;
+            } else {
+                chain.tail->next = node;
+            }
+            chain.tail = node;
+
+            ctx->list.head = ctx->list.head->next;
+        } while (CHECK_SLAVE_METADATA(ctx->list.head, data_version));
+
+        if (ctx->list.head == NULL) {
+            ctx->list.tail = NULL;
+        }
+        chain.tail->next = NULL;
+    }
+
+    if (chain.head != NULL) {
+        fast_mblock_batch_free(&ctx->allocator, &chain);
+    }
+    PTHREAD_MUTEX_UNLOCK(&ctx->lock);
+}
+
+static void *thread_run(void *arg)
+{
+    IdempotencyRequestMetadataContext *ctx;
+    int64_t data_version;
+
+#ifdef OS_LINUX
+    prctl(PR_SET_NAME, "idemp-req-meta");
+#endif
+
+    ctx = g_request_metadata.list.head;
+    while (SF_G_CONTINUE_FLAG) {
+        fc_sleep_ms(g_request_metadata.process_interval_ms);
+
+        if (ctx->is_master_callback.func(ctx->is_master_callback.
+                    arg, &data_version))
+        {
+            process_master_side(ctx);
+        } else if (data_version > 0) {
+            process_slave_side(ctx, data_version);
+        }
+
+        ctx = ctx->next;
+        if (ctx == NULL) {
+            ctx = g_request_metadata.list.head;
+        }
+    }
+
+    return NULL;
+}
+
+int idempotency_request_metadata_init(IdempotencyRequestMetadataContext
+        *ctx, sf_is_master_callback is_master_callback, void *arg)
+{
+    int result;
+
+    if ((result=fast_mblock_init_ex1(&ctx->allocator, "req-metadata-info",
+                    sizeof(IdempotencyRequestMetadata), 8192, 0,
+                    NULL, NULL, false)) != 0)
+    {
+        return result;
+    }
+
+    if ((result=init_pthread_lock(&ctx->lock)) != 0) {
+        return result;
+    }
+
+    ctx->is_master_callback.func = is_master_callback;
+    ctx->is_master_callback.arg = arg;
+    ctx->list.head = ctx->list.tail = NULL;
+
+    ctx->next = NULL;
+    if (g_request_metadata.list.head == NULL) {
+        g_request_metadata.list.head = ctx;
+    } else {
+        g_request_metadata.list.tail->next = ctx;
+    }
+    g_request_metadata.list.tail = ctx;
+
+    return 0;
+}
+
+int idempotency_request_metadata_start(const int process_interval_ms,
+            const int master_side_timeout)
+{
+    pthread_t tid;
+
+    if (g_request_metadata.list.head == NULL) {
+        logError("file: "__FILE__", line: %d, "
+                "list is empty!", __LINE__);
+        return ENOENT;
+    }
+
+    if (process_interval_ms <= 0) {
+        logError("file: "__FILE__", line: %d, "
+                "invalid process interval: %d!",
+                __LINE__, process_interval_ms);
+        return EINVAL;
+    }
+
+    if (master_side_timeout <= 0) {
+        logError("file: "__FILE__", line: %d, "
+                "invalid master side timeout: %d!",
+                __LINE__, master_side_timeout);
+        return EINVAL;
+    }
+
+    g_request_metadata.process_interval_ms = process_interval_ms;
+    g_request_metadata.master_side_timeout = master_side_timeout;
+    return fc_create_thread(&tid, thread_run, NULL,
+            SF_G_THREAD_STACK_SIZE);
+}
+
+int idempotency_request_metadata_add(IdempotencyRequestMetadataContext
+        *ctx, const SFRequestMetadata *metadata, const int n)
+{
+    IdempotencyRequestMetadata *idemp_meta;
+
+    PTHREAD_MUTEX_LOCK(&ctx->lock);
+    do {
+        if ((idemp_meta=fast_mblock_alloc_object(&ctx->allocator)) == NULL) {
+            break;
+        }
+
+        idemp_meta->req_id = metadata->req_id;
+        idemp_meta->data_version = metadata->data_version;
+        idemp_meta->n = n;
+        idemp_meta->enqueue_time = g_current_time;
+        idemp_meta->next = NULL;
+
+        if (ctx->list.head == NULL) {
+            ctx->list.head = idemp_meta;
+        } else {
+            ctx->list.tail->next = idemp_meta;
+        }
+        ctx->list.tail = idemp_meta;
+    } while (0);
+    PTHREAD_MUTEX_UNLOCK(&ctx->lock);
+
+    return (idemp_meta != NULL ? 0 : ENOMEM);
+}
+
+int idempotency_request_metadata_get(IdempotencyRequestMetadataContext
+        *ctx, const int64_t req_id, int64_t *data_version, int *n)
+{
+    int result;
+    IdempotencyRequestMetadata *meta;
+
+    result = ENOENT;
+    PTHREAD_MUTEX_LOCK(&ctx->lock);
+    meta = ctx->list.head;
+    while (meta != NULL) {
+        if (req_id == meta->req_id) {
+            result = 0;
+            *data_version = meta->data_version;
+            if (n != NULL) {
+                *n = meta->n;
+            }
+            break;
+        }
+        meta = meta->next;
+    }
+    PTHREAD_MUTEX_UNLOCK(&ctx->lock);
+
+    return result;
+}

src/idempotency/server/request_metadata.h

@@ -0,0 +1,66 @@
+/*
+ * Copyright (c) 2020 YuQing <384681@qq.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <https://www.gnu.org/licenses/>.
+ */
+
+
+#ifndef _SF_IDEMPOTENCY_REQUEST_METADATA_H
+#define _SF_IDEMPOTENCY_REQUEST_METADATA_H
+
+#include "server_types.h"
+
+typedef bool (*sf_is_master_callback)(void *arg, int64_t *data_version);
+
+typedef struct idempotency_request_metadata {
+    int64_t req_id;
+    int64_t data_version;
+    int n;  //integer argument
+    uint32_t enqueue_time;
+    struct idempotency_request_metadata *next;
+} IdempotencyRequestMetadata;
+
+typedef struct idempotency_request_metadata_context {
+    struct {
+        sf_is_master_callback func;
+        void *arg;
+    } is_master_callback;
+    struct fast_mblock_man allocator;  //element: IdempotencyRequestMetadata
+    pthread_mutex_t lock;
+    struct {
+        IdempotencyRequestMetadata *head;
+        IdempotencyRequestMetadata *tail;
+    } list;
+    struct idempotency_request_metadata_context *next;
+} IdempotencyRequestMetadataContext;
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+    int idempotency_request_metadata_init(IdempotencyRequestMetadataContext
+            *ctx, sf_is_master_callback is_master_callback, void *arg);
+
+    int idempotency_request_metadata_start(const int process_interval_ms,
+            const int master_side_timeout);
+
+    int idempotency_request_metadata_add(IdempotencyRequestMetadataContext
+            *ctx, const SFRequestMetadata *metadata, const int n);
+
+    int idempotency_request_metadata_get(IdempotencyRequestMetadataContext
+            *ctx, const int64_t req_id, int64_t *data_version, int *n);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif

src/idempotency/server/server_handler.c

@@ -37,12 +37,9 @@
 #include "server_channel.h"
 #include "server_handler.h"
 
-#define SF_TASK_BODY_LENGTH(task) \
-    (task->length - sizeof(SFCommonProtoHeader))
-
 int sf_server_deal_setup_channel(struct fast_task_info *task,
-        int *task_type, IdempotencyChannel **channel,
+        int *task_type, const int server_id, IdempotencyChannel
-        SFResponseInfo *response)
+        **channel, SFResponseInfo *response)
 {
     int result;
     SFProtoSetupChannelReq *req;
@@ -52,13 +49,13 @@ int sf_server_deal_setup_channel(struct fast_task_info *task,
 
     response->header.cmd = SF_SERVICE_PROTO_SETUP_CHANNEL_RESP;
     if ((result=sf_server_expect_body_length(response,
-                    SF_TASK_BODY_LENGTH(task),
+                    SF_RECV_BODY_LENGTH(task),
                     sizeof(SFProtoSetupChannelReq))) != 0)
     {
         return result;
     }
 
-    req = (SFProtoSetupChannelReq *)(task->data + sizeof(SFCommonProtoHeader));
+    req = (SFProtoSetupChannelReq *)SF_PROTO_RECV_BODY(task);
     channel_id = buff2int(req->channel_id);
     key = buff2int(req->key);
     if (*channel != NULL) {
@@ -74,14 +71,13 @@ int sf_server_deal_setup_channel(struct fast_task_info *task,
                 "alloc channel fail, hint channel id: %d", channel_id);
         return ENOMEM;
     }
-
     *task_type = SF_SERVER_TASK_TYPE_CHANNEL_HOLDER;
 
-    resp = (SFProtoSetupChannelResp *)(task->data +
+    resp = (SFProtoSetupChannelResp *)SF_PROTO_SEND_BODY(task);
-            sizeof(SFCommonProtoHeader));
     int2buff((*channel)->id, resp->channel_id);
     int2buff((*channel)->key, resp->key);
-    int2buff(task->size, resp->buffer_size);
+    int2buff(server_id, resp->server_id);
+    int2buff(task->send.ptr->size, resp->buffer_size);
     response->header.body_len = sizeof(SFProtoSetupChannelResp);
     return 0;
 }
@@ -135,19 +131,19 @@ int sf_server_deal_report_req_receipt(struct fast_task_info *task,
     SFProtoReportReqReceiptBody *body_part;
     SFProtoReportReqReceiptBody *body_end;
 
+    response->header.cmd = SF_SERVICE_PROTO_REPORT_REQ_RECEIPT_RESP;
     if ((result=check_holder_channel(task_type, channel, response)) != 0) {
         return result;
     }
 
-    body_len = SF_TASK_BODY_LENGTH(task);
+    body_len = SF_RECV_BODY_LENGTH(task);
     if ((result=sf_server_check_min_body_length(response, body_len,
                     sizeof(SFProtoReportReqReceiptHeader))) != 0)
     {
         return result;
     }
 
-    body_header = (SFProtoReportReqReceiptHeader *)
+    body_header = (SFProtoReportReqReceiptHeader *)SF_PROTO_RECV_BODY(task);
-    (task->data + sizeof(SFCommonProtoHeader));
     count = buff2int(body_header->count);
     calc_body_len = sizeof(SFProtoReportReqReceiptHeader) +
         sizeof(SFProtoReportReqReceiptBody) * count;
@@ -169,7 +165,6 @@ int sf_server_deal_report_req_receipt(struct fast_task_info *task,
     }
 
     //logInfo("receipt count: %d, success: %d", count, success);
-    response->header.cmd = SF_SERVICE_PROTO_REPORT_REQ_RECEIPT_RESP;
     return 0;
 }
 
@@ -189,7 +184,7 @@ IdempotencyRequest *sf_server_update_prepare_and_check(
     }
 
     adheader = (SFProtoIdempotencyAdditionalHeader *)req->body;
-    request = (IdempotencyRequest *)fast_mblock_alloc_object(request_allocator);
+    request = fast_mblock_alloc_object(request_allocator);
     if (request == NULL) {
         *result = ENOMEM;
         return NULL;
@@ -220,7 +215,7 @@ int sf_server_deal_rebind_channel(struct fast_task_info *task,
     SFProtoRebindChannelReq *req;
 
     if ((result=sf_server_expect_body_length(response,
-                    SF_TASK_BODY_LENGTH(task),
+                    SF_RECV_BODY_LENGTH(task),
                     sizeof(SFProtoRebindChannelReq))) != 0)
     {
         return result;
@@ -240,7 +235,7 @@ int sf_server_deal_rebind_channel(struct fast_task_info *task,
     }
     idempotency_channel_release(*channel, false);
 
-    req = (SFProtoRebindChannelReq *)(task->data + sizeof(SFCommonProtoHeader));
+    req = (SFProtoRebindChannelReq *)SF_PROTO_RECV_BODY(task);
     channel_id = buff2int(req->channel_id);
     key = buff2int(req->key);
     *channel = idempotency_channel_find_and_hold(channel_id, key, &result);

src/idempotency/server/server_handler.h

@@ -25,8 +25,8 @@ extern "C" {
 #endif
 
 int sf_server_deal_setup_channel(struct fast_task_info *task,
-        int *task_type, IdempotencyChannel **channel,
+        int *task_type, const int server_id, IdempotencyChannel
-        SFResponseInfo *response);
+        **channel, SFResponseInfo *response);
 
 int sf_server_deal_close_channel(struct fast_task_info *task,
         int *task_type, IdempotencyChannel **channel,

src/idempotency/server/server_types.h

@@ -19,11 +19,7 @@
 
 #include "fastcommon/fast_mblock.h"
 #include "fastcommon/fast_timer.h"
-
+#include "sf/idempotency/common/idempotency_types.h"
-#define SF_IDEMPOTENCY_CHANNEL_ID_BITS    16
-#define SF_IDEMPOTENCY_REQUEST_ID_BITS    (64 - SF_IDEMPOTENCY_CHANNEL_ID_BITS)
-#define SF_IDEMPOTENCY_MAX_CHANNEL_COUNT  ((1 << SF_IDEMPOTENCY_CHANNEL_ID_BITS) - 1)
-#define SF_IDEMPOTENCY_MAX_CHANNEL_ID     SF_IDEMPOTENCY_MAX_CHANNEL_COUNT
 
 #define SF_IDEMPOTENCY_DEFAULT_REQUEST_HINT_CAPACITY      1023
 #define SF_IDEMPOTENCY_DEFAULT_CHANNEL_RESERVE_INTERVAL    600
@@ -61,6 +57,17 @@ typedef struct idempotency_channel {
     struct idempotency_channel *next;
 } IdempotencyChannel;
 
+typedef struct sf_request_metadata {
+    int64_t req_id;
+    int64_t data_version;
+} SFRequestMetadata;
+
+typedef struct sf_request_metadata_array {
+    SFRequestMetadata *elts;
+    int count;
+    int alloc;
+} SFRequestMetadataArray;
+
 #ifdef __cplusplus
 extern "C" {
 #endif

src/sf_binlog_index.c

@@ -68,10 +68,6 @@ static int parse(SFBinlogIndexContext *ctx, const string_t *lines,
     const string_t *end;
     void *bindex;
 
-    if (row_count < 1) {
-        return EINVAL;
-    }
-
     if ((result=parse_header(lines, &record_count, &ctx->
                     last_version, error_info)) != 0)
     {
@@ -140,7 +136,9 @@ static int load(SFBinlogIndexContext *ctx)
     }
 
     row_count = split_string_ex(&context, '\n', lines, row_count, true);
+    if (row_count > 0) {
         result = parse(ctx, lines, row_count);
+    }
     free(lines);
     free(context.str);
     return result;
@@ -175,7 +173,9 @@ static int save(SFBinlogIndexContext *ctx, const char *filename)
     int i;
     int result;
 
-    if ((fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0644)) < 0) {
+    if ((fd=open(filename, O_WRONLY | O_CREAT | O_TRUNC |
+                    O_CLOEXEC, 0644)) < 0)
+    {
         result = errno != 0 ? errno : EIO;
         logError("file: "__FILE__", line: %d, "
                 "open file %s fail, errno: %d, error info: %s",
@@ -186,9 +186,11 @@ static int save(SFBinlogIndexContext *ctx, const char *filename)
     result = 0;
     p = buff;
     bend = buff + sizeof(buff);
-    p += sprintf(p, "%d %"PRId64"\n",
+
-            ctx->index_array.count,
+    p += fc_itoa(ctx->index_array.count, p);
-            ctx->last_version);
+    *p++ = ' ';
+    p += fc_itoa(ctx->last_version, p);
+    *p++ = '\n';
 
     index = ctx->index_array.indexes;
     for (i=0; i<ctx->index_array.count; i++) {
@@ -227,7 +229,7 @@ int sf_binlog_index_save(SFBinlogIndexContext *ctx)
     int result;
     char tmp_filename[PATH_MAX];
 
-    snprintf(tmp_filename, sizeof(tmp_filename), "%s.tmp", ctx->filename);
+    fc_combine_two_strings(ctx->filename, "tmp", '.', tmp_filename);
     if ((result=save(ctx, tmp_filename)) != 0) {
         return result;
     }

src/sf_binlog_writer.c

@@ -13,18 +13,6 @@
  * along with this program. If not, see <https://www.gnu.org/licenses/>.
  */
 
-#include <sys/types.h>
-#include <sys/stat.h>
-#include <sys/socket.h>
-#include <netinet/in.h>
-#include <arpa/inet.h>
-#include <stdio.h>
-#include <stdlib.h>
-#include <string.h>
-#include <errno.h>
-#include <limits.h>
-#include <fcntl.h>
-#include <pthread.h>
 #include "fastcommon/logger.h"
 #include "fastcommon/sockopt.h"
 #include "fastcommon/shared_func.h"
@@ -34,6 +22,8 @@
 #include "sf_func.h"
 #include "sf_binlog_writer.h"
 
+#define ERRNO_THREAD_EXIT  -1000
+
 static inline void binlog_writer_set_next_version(SFBinlogWriterInfo *writer,
         const uint64_t next_version)
 {
@@ -44,7 +34,8 @@ static inline void binlog_writer_set_next_version(SFBinlogWriterInfo *writer,
 }
 
 #define deal_binlog_one_record(wb) \
-    sf_file_writer_deal_buffer(&wb->writer->fw, &wb->bf, wb->version.last)
+    sf_file_writer_deal_versioned_buffer(&wb->writer->fw, \
+            &wb->bf, wb->version.last)
 
 #define GET_WBUFFER_VERSION_COUNT(wb)  \
         (((wb)->version.last - (wb)->version.first) + 1)
@@ -70,9 +61,9 @@ static int deal_record_by_version(SFBinlogWriterBuffer *wb)
     if (wb->version.first < writer->version_ctx.next) {
         logError("file: "__FILE__", line: %d, subdir_name: %s, "
                 "current version: %"PRId64" is too small which "
-                "less than %"PRId64", tag: %"PRId64", buffer(%d): %.*s",
+                "less than %"PRId64", buffer(%d): %.*s",
                 __LINE__, writer->fw.cfg.subdir_name, wb->version.first,
-                writer->version_ctx.next, wb->tag, wb->bf.length,
+                writer->version_ctx.next, wb->bf.length,
                 wb->bf.length, wb->bf.buff);
         fast_mblock_free_object(&writer->thread->mblock, wb);
         return 0;
@@ -165,9 +156,6 @@ static inline int flush_writer_files(SFBinlogWriterThread *thread)
             return result;
         }
 
-        if (writer->fw.flags & SF_FILE_WRITER_FLAGS_WANT_DONE_VERSION) {
-            writer->fw.last_versions.done = writer->fw.last_versions.pending;
-        }
         writer->flush.in_queue = false;
         writer = writer->flush.next;
     }
@@ -177,9 +165,10 @@ static inline int flush_writer_files(SFBinlogWriterThread *thread)
 }
 
 static int deal_binlog_records(SFBinlogWriterThread *thread,
-        SFBinlogWriterBuffer *wb_head)
+        SFBinlogWriterBuffer *wb_head, uint32_t *last_timestamp)
 {
     int result;
+    bool skip_empty_file;
     SFBinlogWriterBuffer *wbuffer;
     SFBinlogWriterBuffer *current;
 
@@ -187,21 +176,73 @@ static int deal_binlog_records(SFBinlogWriterThread *thread,
     do {
         current = wbuffer;
         wbuffer = wbuffer->next;
+        if (wbuffer == NULL) {
+            *last_timestamp = current->timestamp;
+        }
 
         switch (current->type) {
             case SF_BINLOG_BUFFER_TYPE_CHANGE_ORDER_TYPE:
-                thread->order_by = current->version.first;
+                current->writer->order_by = current->version.first;
                 fast_mblock_free_object(&current->writer->
                         thread->mblock, current);
                 break;
-
+            case SF_BINLOG_BUFFER_TYPE_CHANGE_PASSIVE_WRITE:
+                thread->passive_write = current->version.first;
+                fast_mblock_free_object(&current->writer->
+                        thread->mblock, current);
+                break;
+            case SF_BINLOG_BUFFER_TYPE_CHANGE_CALL_FSYNC:
+                current->writer->fw.cfg.call_fsync = current->version.first;
+                fast_mblock_free_object(&current->writer->
+                        thread->mblock, current);
+                break;
+            case SF_BINLOG_BUFFER_TYPE_ROTATE_FILE:
+                flush_writer_files(thread);
+                skip_empty_file = current->version.first;
+                if (!(skip_empty_file && current->writer->fw.file.size == 0)) {
+                    if ((result=sf_file_writer_set_binlog_write_index(&current->
+                                    writer->fw, current->writer->fw.binlog.
+                                    last_index + 1)) != 0)
+                    {
+                        return result;
+                    }
+                }
+                fast_mblock_free_object(&current->writer->
+                        thread->mblock, current);
+                break;
+            case SF_BINLOG_BUFFER_TYPE_FLUSH_FILE:
+                if ((result=flush_writer_files(thread)) != 0) {
+                    return result;
+                }
+                fast_mblock_free_object(&current->writer->
+                        thread->mblock, current);
+                break;
+            case SF_BINLOG_BUFFER_TYPE_SET_WRITE_INDEX:
+                if ((result=sf_file_writer_set_binlog_write_index(&current->
+                                writer->fw, current->version.first)) != 0)
+                {
+                    return result;
+                }
+                fast_mblock_free_object(&current->writer->
+                        thread->mblock, current);
+                break;
+            case SF_BINLOG_BUFFER_TYPE_NOTIFY_EXIT:
+                flush_writer_files(thread);
+                fast_mblock_free_object(&current->writer->
+                        thread->mblock, current);
+                return ERRNO_THREAD_EXIT;
             case SF_BINLOG_BUFFER_TYPE_SET_NEXT_VERSION:
-                if (thread->order_by != SF_BINLOG_THREAD_TYPE_ORDER_BY_VERSION) {
+                if (current->writer->order_by !=
+                        SF_BINLOG_WRITER_TYPE_ORDER_BY_VERSION &&
+                        current->writer->thread->order_mode !=
+                        SF_BINLOG_THREAD_ORDER_MODE_VARY)
+                {
                     logWarning("file: "__FILE__", line: %d, "
-                            "subdir_name: %s, invalid order by: %d != %d, "
+                            "subdir_name: %s, order by: %d != %d, "
-                            "maybe some mistake happen", __LINE__,
+                            "maybe some mistake happen?", __LINE__,
-                            current->writer->fw.cfg.subdir_name, thread->order_by,
+                            current->writer->fw.cfg.subdir_name,
-                            SF_BINLOG_THREAD_TYPE_ORDER_BY_VERSION);
+                            current->writer->order_by,
+                            SF_BINLOG_WRITER_TYPE_ORDER_BY_VERSION);
                 }
 
                 if (current->writer->version_ctx.ring.waiting_count != 0) {
@@ -216,8 +257,10 @@ static int deal_binlog_records(SFBinlogWriterThread *thread,
                         __LINE__, current->writer->fw.cfg.subdir_name,
                         current->version.first);
 
-                if (current->writer->version_ctx.next !=
+                if ((current->writer->order_by ==
-                        current->version.first)
+                            SF_BINLOG_WRITER_TYPE_ORDER_BY_NONE) ||
+                        (current->writer->version_ctx.next !=
+                         current->version.first))
                 {
                     binlog_writer_set_next_version(current->writer,
                             current->version.first);
@@ -226,12 +269,13 @@ static int deal_binlog_records(SFBinlogWriterThread *thread,
                 fast_mblock_free_object(&current->writer->
                         thread->mblock, current);
                 break;
-
             default:
                 current->writer->fw.total_count++;
                 add_to_flush_writer_queue(thread, current->writer);
 
-                if (thread->order_by == SF_BINLOG_THREAD_TYPE_ORDER_BY_VERSION) {
+                if (current->writer->order_by ==
+                        SF_BINLOG_WRITER_TYPE_ORDER_BY_VERSION)
+                {
                     /* NOTE: current maybe be released in the deal function */
                     if ((result=deal_record_by_version(current)) != 0) {
                         return result;
@@ -248,19 +292,31 @@ static int deal_binlog_records(SFBinlogWriterThread *thread,
         }
     } while (wbuffer != NULL);
 
+    if (thread->passive_write) {
+        return 0;
+    } else {
         return flush_writer_files(thread);
+    }
 }
 
 void sf_binlog_writer_finish(SFBinlogWriterInfo *writer)
 {
     SFBinlogWriterBuffer *wb_head;
+    uint32_t last_timestamp;
     int count;
 
-    if (writer->fw.file.name != NULL) {
+    if (writer->fw.file.name.str != NULL) {
-        fc_queue_terminate(&writer->thread->queue);
+        while (writer->thread->running && !fc_queue_empty(
+                    &writer->thread->queue))
+        {
+            fc_sleep_ms(10);
+        }
+        if (writer->thread->running) {
+            sf_binlog_writer_notify_exit(writer);
+        }
 
         count = 0;
-        while (writer->thread->running && ++count < 300) {
+        while (writer->thread->running && ++count < 500) {
             fc_sleep_ms(10);
         }
         
@@ -273,11 +329,12 @@ void sf_binlog_writer_finish(SFBinlogWriterInfo *writer)
         wb_head = (SFBinlogWriterBuffer *)fc_queue_try_pop_all(
                 &writer->thread->queue);
         if (wb_head != NULL) {
-            deal_binlog_records(writer->thread, wb_head);
+            last_timestamp = 0;
+            deal_binlog_records(writer->thread, wb_head, &last_timestamp);
         }
 
-        free(writer->fw.file.name);
+        free(writer->fw.file.name.str);
-        writer->fw.file.name = NULL;
+        writer->fw.file.name.str = NULL;
     }
 
     if (writer->fw.file.fd >= 0) {
@@ -290,18 +347,22 @@ static void *binlog_writer_func(void *arg)
 {
     SFBinlogWriterThread *thread;
     SFBinlogWriterBuffer *wb_head;
+    uint32_t last_record_time;
+    uint32_t current_timestamp;
+    uint32_t last_timestamp;
+    int result;
 
     thread = (SFBinlogWriterThread *)arg;
 
 #ifdef OS_LINUX
     {
         char thread_name[64];
-        snprintf(thread_name, sizeof(thread_name),
+        fc_combine_two_strings(thread->name, "writer", '-', thread_name);
-                "%s-writer", thread->name);
         prctl(PR_SET_NAME, thread_name);
     }
 #endif
 
+    last_record_time = current_timestamp = last_timestamp = 0;
     thread->running = true;
     while (SF_G_CONTINUE_FLAG) {
         wb_head = (SFBinlogWriterBuffer *)fc_queue_pop_all(&thread->queue);
@@ -309,12 +370,42 @@ static void *binlog_writer_func(void *arg)
             continue;
         }
 
-        if (deal_binlog_records(thread, wb_head) != 0) {
+        if ((result=deal_binlog_records(thread, wb_head,
+                        &last_record_time)) != 0)
+        {
+            if (result != ERRNO_THREAD_EXIT) {
                 logCrit("file: "__FILE__", line: %d, "
                         "deal_binlog_records fail, "
                         "program exit!", __LINE__);
                 sf_terminate_myself();
             }
+            break;
+        }
+
+        if (fc_queue_empty(&thread->queue)) {
+            current_timestamp = 0;
+        } else {
+            current_timestamp = last_record_time;
+        }
+        if ((current_timestamp == 0 && last_timestamp != 0) ||
+                (current_timestamp > last_timestamp))
+        {
+            last_timestamp = current_timestamp;
+            FC_ATOMIC_SET(thread->flow_ctrol.last_timestamp,
+                    current_timestamp);
+
+            PTHREAD_MUTEX_LOCK(&thread->flow_ctrol.lcp.lock);
+            if (thread->flow_ctrol.waiting_count > 0) {
+                pthread_cond_broadcast(&thread->flow_ctrol.lcp.cond);
+            }
+            PTHREAD_MUTEX_UNLOCK(&thread->flow_ctrol.lcp.lock);
+        }
+
+        if (thread->write_interval_ms > 0 &&
+                last_record_time == g_current_time)
+        {
+            fc_sleep_ms(thread->write_interval_ms);
+        }
     }
 
     thread->running = false;
@@ -341,19 +432,34 @@ static int binlog_wbuffer_alloc_init(void *element, void *args)
     return 0;
 }
 
-int sf_binlog_writer_init_normal(SFBinlogWriterInfo *writer,
+static void binlog_wbuffer_destroy_func(void *element, void *args)
-        const char *data_path, const char *subdir_name,
-        const int buffer_size)
 {
-    writer->flush.in_queue = false;
+    SFBinlogWriterBuffer *wbuffer;
-    return sf_file_writer_init_normal(&writer->fw,
+    wbuffer = (SFBinlogWriterBuffer *)element;
-            data_path, subdir_name, buffer_size);
+    if (wbuffer->bf.buff != NULL) {
+        free(wbuffer->bf.buff);
+    }
 }
 
-int sf_binlog_writer_init_by_version(SFBinlogWriterInfo *writer,
+int sf_binlog_writer_init_normal_ex(SFBinlogWriterInfo *writer,
         const char *data_path, const char *subdir_name,
+        const char *file_prefix, const int max_record_size,
+        const int buffer_size, const int64_t file_rotate_size,
+        const bool call_fsync)
+{
+    memset(writer, 0, sizeof(*writer));
+    writer->order_by = SF_BINLOG_WRITER_TYPE_ORDER_BY_NONE;
+    return sf_file_writer_init(&writer->fw, data_path, subdir_name,
+            file_prefix, max_record_size, buffer_size,
+            file_rotate_size, call_fsync);
+}
+
+int sf_binlog_writer_init_by_version_ex(SFBinlogWriterInfo *writer,
+        const char *data_path, const char *subdir_name,
+        const char *file_prefix, const int max_record_size,
         const uint64_t next_version, const int buffer_size,
-        const int ring_size)
+        const int ring_size, const int64_t file_rotate_size,
+        const bool call_fsync)
 {
     int bytes;
 
@@ -367,36 +473,49 @@ int sf_binlog_writer_init_by_version(SFBinlogWriterInfo *writer,
     writer->version_ctx.ring.waiting_count = 0;
     writer->version_ctx.ring.max_waitings = 0;
     writer->version_ctx.change_count = 0;
+    writer->order_by = SF_BINLOG_WRITER_TYPE_ORDER_BY_VERSION;
 
     binlog_writer_set_next_version(writer, next_version);
-    return sf_binlog_writer_init_normal(writer,
+    writer->flush.in_queue = false;
-            data_path, subdir_name, buffer_size);
+    return sf_file_writer_init(&writer->fw, data_path, subdir_name,
+            file_prefix, max_record_size, buffer_size,
+            file_rotate_size, call_fsync);
 }
 
 int sf_binlog_writer_init_thread_ex(SFBinlogWriterThread *thread,
         const char *name, SFBinlogWriterInfo *writer, const short order_mode,
-        const short order_by, const int max_record_size,
+        const int write_interval_ms, const int max_delay,
-        const int writer_count, const bool use_fixed_buffer_size)
+        const int max_record_size, const bool use_fixed_buffer_size,
+        const bool passive_write)
 {
     const int alloc_elements_once = 1024;
+    const int64_t alloc_elements_limit = 0;
+    const int prealloc_trunk_count = 0;
+    int result;
     int element_size;
     pthread_t tid;
-    int result;
+    struct fast_mblock_object_callbacks callbacks;
 
-    snprintf(thread->name, sizeof(thread->name), "%s", name);
+    fc_safe_strcpy(thread->name, name);
     thread->order_mode = order_mode;
-    thread->order_by = order_by;
     thread->use_fixed_buffer_size = use_fixed_buffer_size;
+    thread->passive_write = passive_write;
+    thread->write_interval_ms = write_interval_ms;
+    thread->flow_ctrol.max_delay = max_delay;
     writer->fw.cfg.max_record_size = max_record_size;
     writer->thread = thread;
-
+    callbacks.init_func = binlog_wbuffer_alloc_init;
+    callbacks.args = writer;
     element_size = sizeof(SFBinlogWriterBuffer);
     if (use_fixed_buffer_size) {
         element_size += max_record_size;
+        callbacks.destroy_func = NULL;
+    } else {
+        callbacks.destroy_func = binlog_wbuffer_destroy_func;
     }
-    if ((result=fast_mblock_init_ex1(&thread->mblock, "binlog-wbuffer",
+    if ((result=fast_mblock_init_ex2(&thread->mblock, "binlog-wbuffer",
-                     element_size, alloc_elements_once, 0,
+                     element_size, alloc_elements_once, alloc_elements_limit,
-                     binlog_wbuffer_alloc_init, writer, true)) != 0)
+                     prealloc_trunk_count, &callbacks, true, NULL)) != 0)
     {
         return result;
     }
@@ -407,6 +526,12 @@ int sf_binlog_writer_init_thread_ex(SFBinlogWriterThread *thread,
         return result;
     }
 
+    thread->flow_ctrol.last_timestamp = 0;
+    thread->flow_ctrol.waiting_count = 0;
+    if ((result=init_pthread_lock_cond_pair(&thread->flow_ctrol.lcp)) != 0) {
+        return result;
+    }
+
     thread->flush_writers.head = thread->flush_writers.tail = NULL;
     return fc_create_thread(&tid, binlog_writer_func, thread,
             SF_G_THREAD_STACK_SIZE);
@@ -417,12 +542,12 @@ int sf_binlog_writer_change_order_by(SFBinlogWriterInfo *writer,
 {
     SFBinlogWriterBuffer *buffer;
 
-    if (writer->thread->order_by == order_by) {
+    if (writer->order_by == order_by) {
         return 0;
     }
 
-    if (!(order_by == SF_BINLOG_THREAD_TYPE_ORDER_BY_NONE ||
+    if (!(order_by == SF_BINLOG_WRITER_TYPE_ORDER_BY_NONE ||
-                order_by == SF_BINLOG_THREAD_TYPE_ORDER_BY_VERSION))
+                order_by == SF_BINLOG_WRITER_TYPE_ORDER_BY_VERSION))
     {
         logError("file: "__FILE__", line: %d, "
                 "invalid order by: %d!", __LINE__, order_by);
@@ -437,27 +562,147 @@ int sf_binlog_writer_change_order_by(SFBinlogWriterInfo *writer,
         return EINVAL;
     }
 
+    if (order_by == SF_BINLOG_WRITER_TYPE_ORDER_BY_VERSION) {
+        if (writer->version_ctx.ring.slots == NULL) {
+            logError("file: "__FILE__", line: %d, "
+                    "the writer is NOT versioned writer, can't "
+                    "set order by to %d!", __LINE__, order_by);
+            return EINVAL;
+        }
+    }
+
     if ((buffer=sf_binlog_writer_alloc_versioned_buffer_ex(writer, order_by,
                     order_by, SF_BINLOG_BUFFER_TYPE_CHANGE_ORDER_TYPE)) == NULL)
     {
         return ENOMEM;
     }
 
-    fc_queue_push(&writer->thread->queue, buffer);
+    sf_push_to_binlog_write_queue(writer, buffer);
     return 0;
 }
 
+static inline int sf_binlog_writer_push_directive(SFBinlogWriterInfo *writer,
+        const int buffer_type, const int64_t version)
+{
+    SFBinlogWriterBuffer *buffer;
+
+    if ((buffer=sf_binlog_writer_alloc_versioned_buffer_ex(writer,
+                    version, version, buffer_type)) == NULL)
+    {
+        return ENOMEM;
+    }
+
+    sf_push_to_binlog_write_queue(writer, buffer);
+    return 0;
+}
+
+int sf_binlog_writer_change_passive_write(SFBinlogWriterInfo *writer,
+        const bool passive_write)
+{
+    return sf_binlog_writer_push_directive(writer,
+            SF_BINLOG_BUFFER_TYPE_CHANGE_PASSIVE_WRITE,
+            passive_write);
+}
+
+int sf_binlog_writer_change_call_fsync(SFBinlogWriterInfo *writer,
+        const bool call_fsync)
+{
+    return sf_binlog_writer_push_directive(writer,
+            SF_BINLOG_BUFFER_TYPE_CHANGE_CALL_FSYNC,
+            call_fsync);
+}
+
 int sf_binlog_writer_change_next_version(SFBinlogWriterInfo *writer,
         const int64_t next_version)
 {
-    SFBinlogWriterBuffer *buffer;
+    return sf_binlog_writer_push_directive(writer,
+            SF_BINLOG_BUFFER_TYPE_SET_NEXT_VERSION,
+            next_version);
+}
 
-    if ((buffer=sf_binlog_writer_alloc_versioned_buffer_ex(writer, next_version,
+int sf_binlog_writer_change_write_index(SFBinlogWriterInfo *writer,
-                    next_version, SF_BINLOG_BUFFER_TYPE_SET_NEXT_VERSION)) == NULL)
+        const int write_index)
+{
+    return sf_binlog_writer_push_directive(writer,
+            SF_BINLOG_BUFFER_TYPE_SET_WRITE_INDEX, write_index);
+}
+
+int sf_binlog_writer_rotate_file_ex(SFBinlogWriterInfo *writer,
+        const bool skip_empty_file)
+{
+    return sf_binlog_writer_push_directive(writer,
+            SF_BINLOG_BUFFER_TYPE_ROTATE_FILE,
+            skip_empty_file ? 1 : 0);
+}
+
+int sf_binlog_writer_flush_file(SFBinlogWriterInfo *writer)
+{
+    return sf_binlog_writer_push_directive(writer,
+            SF_BINLOG_BUFFER_TYPE_FLUSH_FILE, 0);
+}
+
+int sf_binlog_writer_notify_exit(SFBinlogWriterInfo *writer)
+{
+    return sf_binlog_writer_push_directive(writer,
+            SF_BINLOG_BUFFER_TYPE_NOTIFY_EXIT, 0);
+}
+
+void sf_push_to_binlog_write_queue(SFBinlogWriterInfo *writer,
+        SFBinlogWriterBuffer *buffer)
+{
+    time_t current_time;
+    int64_t last_timestamp;
+
+    current_time = g_current_time;
+    last_timestamp = FC_ATOMIC_GET(writer->thread->flow_ctrol.last_timestamp);
+    if ((last_timestamp > 0 && current_time - last_timestamp > writer->
+                thread->flow_ctrol.max_delay) && !(writer->order_by ==
+                    SF_BINLOG_WRITER_TYPE_ORDER_BY_VERSION && buffer->
+                    version.first - writer->version_ctx.next < 128))
     {
-        return ENOMEM;
+        time_t last_log_timestamp;
+        int time_used;
+        int log_level;
+
+        PTHREAD_MUTEX_LOCK(&writer->thread->flow_ctrol.lcp.lock);
+        writer->thread->flow_ctrol.waiting_count++;
+        last_timestamp = FC_ATOMIC_GET(writer->thread->
+                flow_ctrol.last_timestamp);
+        while ((last_timestamp > 0 && current_time - last_timestamp > writer->
+                    thread->flow_ctrol.max_delay) && !(writer->order_by ==
+                        SF_BINLOG_WRITER_TYPE_ORDER_BY_VERSION && buffer->
+                        version.first - writer->version_ctx.next < 128))
+        {
+            pthread_cond_wait(&writer->thread->flow_ctrol.lcp.cond,
+                    &writer->thread->flow_ctrol.lcp.lock);
+            last_timestamp = FC_ATOMIC_GET(writer->thread->
+                    flow_ctrol.last_timestamp);
+        }
+        writer->thread->flow_ctrol.waiting_count--;
+        PTHREAD_MUTEX_UNLOCK(&writer->thread->flow_ctrol.lcp.lock);
+
+        time_used = g_current_time - current_time;
+        if (time_used > 0) {
+            last_log_timestamp = FC_ATOMIC_GET(
+                    LAST_BINLOG_WRITER_LOG_TIMESTAMP);
+            if (g_current_time != last_log_timestamp &&
+                    __sync_bool_compare_and_swap(
+                        &LAST_BINLOG_WRITER_LOG_TIMESTAMP,
+                        last_log_timestamp, g_current_time))
+            {
+                if (time_used <= writer->thread->flow_ctrol.max_delay) {
+                    log_level = LOG_DEBUG;
+                } else {
+                    log_level = LOG_WARNING;
+                }
+                log_it_ex(&g_log_context, log_level, "file: "__FILE__", line: %d, "
+                        "subdir_name: %s, max_delay: %d s, flow ctrol waiting "
+                        "time: %d s", __LINE__, writer->fw.cfg.subdir_name,
+                        writer->thread->flow_ctrol.max_delay, time_used);
+            }
+        }
     }
 
+    buffer->timestamp = g_current_time;
     fc_queue_push(&writer->thread->queue, buffer);
-    return 0;
 }

src/sf_binlog_writer.h

@@ -19,18 +19,25 @@
 #define _SF_BINLOG_WRITER_H_
 
 #include "fastcommon/fc_queue.h"
+#include "fastcommon/fc_atomic.h"
 #include "sf_types.h"
 #include "sf_file_writer.h"
 
 #define SF_BINLOG_THREAD_ORDER_MODE_FIXED       0
 #define SF_BINLOG_THREAD_ORDER_MODE_VARY        1
 
-#define SF_BINLOG_THREAD_TYPE_ORDER_BY_NONE     0
+#define SF_BINLOG_WRITER_TYPE_ORDER_BY_NONE     0
-#define SF_BINLOG_THREAD_TYPE_ORDER_BY_VERSION  1
+#define SF_BINLOG_WRITER_TYPE_ORDER_BY_VERSION  1
 
 #define SF_BINLOG_BUFFER_TYPE_WRITE_TO_FILE        0  //default type, must be 0
 #define SF_BINLOG_BUFFER_TYPE_SET_NEXT_VERSION     1
 #define SF_BINLOG_BUFFER_TYPE_CHANGE_ORDER_TYPE    2
+#define SF_BINLOG_BUFFER_TYPE_CHANGE_PASSIVE_WRITE 3
+#define SF_BINLOG_BUFFER_TYPE_CHANGE_CALL_FSYNC    4
+#define SF_BINLOG_BUFFER_TYPE_SET_WRITE_INDEX      5
+#define SF_BINLOG_BUFFER_TYPE_ROTATE_FILE          6
+#define SF_BINLOG_BUFFER_TYPE_NOTIFY_EXIT          7
+#define SF_BINLOG_BUFFER_TYPE_FLUSH_FILE           8
 
 #define SF_BINLOG_BUFFER_SET_VERSION(buffer, ver)  \
     (buffer)->version.first = (buffer)->version.last = ver
@@ -40,8 +47,8 @@ struct sf_binlog_writer_info;
 typedef struct sf_binlog_writer_buffer {
     SFVersionRange version;
     BufferInfo bf;
-    int64_t tag;
+    int type;
-    int type;    //for versioned writer
+    uint32_t timestamp;  //for flow ctrol
     struct sf_binlog_writer_info *writer;
     struct sf_binlog_writer_buffer *next;
 } SFBinlogWriterBuffer;
@@ -63,8 +70,15 @@ typedef struct binlog_writer_thread {
     char name[64];
     volatile bool running;
     bool use_fixed_buffer_size;
-    short order_mode;
+    bool passive_write;
-    short order_by;
+    char order_mode;
+    int write_interval_ms;
+    struct {
+        int max_delay;  //in seconds
+        volatile uint32_t last_timestamp;
+        int waiting_count;
+        pthread_lock_cond_pair_t lcp;
+    } flow_ctrol;
     struct {
         struct sf_binlog_writer_info *head;
         struct sf_binlog_writer_info *tail;
@@ -76,11 +90,12 @@ typedef struct sf_binlog_writer_info {
 
     struct {
         SFBinlogWriterBufferRing ring;
-        int64_t next;
+        volatile int64_t next;
         int64_t change_count;  //version change count
     } version_ctx;
     SFBinlogWriterThread *thread;
 
+    short order_by;
     struct {
         bool in_queue;
         struct sf_binlog_writer_info *next;
@@ -96,58 +111,172 @@ typedef struct sf_binlog_writer_context {
 extern "C" {
 #endif
 
-int sf_binlog_writer_init_normal(SFBinlogWriterInfo *writer,
+int sf_binlog_writer_init_normal_ex(SFBinlogWriterInfo *writer,
         const char *data_path, const char *subdir_name,
-        const int buffer_size);
+        const char *file_prefix, const int max_record_size,
+        const int buffer_size, const int64_t file_rotate_size,
+        const bool call_fsync);
 
-int sf_binlog_writer_init_by_version(SFBinlogWriterInfo *writer,
+int sf_binlog_writer_init_by_version_ex(SFBinlogWriterInfo *writer,
         const char *data_path, const char *subdir_name,
+        const char *file_prefix, const int max_record_size,
         const uint64_t next_version, const int buffer_size,
-        const int ring_size);
+        const int ring_size, const int64_t file_rotate_size,
+        const bool call_fsync);
 
 int sf_binlog_writer_init_thread_ex(SFBinlogWriterThread *thread,
         const char *name, SFBinlogWriterInfo *writer, const short order_mode,
-        const short order_by, const int max_record_size,
+        const int write_interval_ms, const int max_delay,
-        const int writer_count, const bool use_fixed_buffer_size);
+        const int max_record_size, const bool use_fixed_buffer_size,
+        const bool passive_write);
+
+#define sf_binlog_writer_init_normal(writer, data_path, \
+        subdir_name, max_record_size, buffer_size)      \
+    sf_binlog_writer_init_normal_ex(writer, data_path, subdir_name,  \
+            SF_BINLOG_FILE_PREFIX_STR, max_record_size, buffer_size, \
+            SF_BINLOG_DEFAULT_ROTATE_SIZE, true)
+
+#define sf_binlog_writer_init_by_version(writer, data_path, subdir_name, \
+        max_record_size, next_version, buffer_size, ring_size)           \
+    sf_binlog_writer_init_by_version_ex(writer, data_path, subdir_name,  \
+            SF_BINLOG_FILE_PREFIX_STR, max_record_size, next_version,    \
+            buffer_size, ring_size, SF_BINLOG_DEFAULT_ROTATE_SIZE, true)
 
 #define sf_binlog_writer_init_thread(thread, name, \
-        writer, order_by, max_record_size)   \
+        writer, write_interval_ms, max_delay, max_record_size)    \
     sf_binlog_writer_init_thread_ex(thread, name, writer,         \
-            SF_BINLOG_THREAD_ORDER_MODE_FIXED,      \
+            SF_BINLOG_THREAD_ORDER_MODE_FIXED, write_interval_ms, \
-            order_by, max_record_size, 1, true)
+            max_delay, max_record_size, true, false)
 
-static inline int sf_binlog_writer_init(SFBinlogWriterContext *context,
+static inline int sf_binlog_writer_init_ex(SFBinlogWriterContext *context,
         const char *data_path, const char *subdir_name,
-        const int buffer_size, const int max_record_size)
+        const char *file_prefix, const int buffer_size,
+        const int write_interval_ms, const int max_delay,
+        const int max_record_size, const bool call_fsync)
 {
     int result;
-    if ((result=sf_binlog_writer_init_normal(&context->writer,
+    if ((result=sf_binlog_writer_init_normal_ex(&context->writer, data_path,
-                    data_path, subdir_name, buffer_size)) != 0)
+                    subdir_name, file_prefix, max_record_size, buffer_size,
+                    SF_BINLOG_DEFAULT_ROTATE_SIZE, call_fsync)) != 0)
     {
         return result;
     }
 
     return sf_binlog_writer_init_thread(&context->thread, subdir_name,
-            &context->writer, SF_BINLOG_THREAD_TYPE_ORDER_BY_NONE,
+            &context->writer, write_interval_ms, max_delay, max_record_size);
-            max_record_size);
+}
+
+#define sf_binlog_writer_init(context, data_path, subdir_name,      \
+        buffer_size, write_interval_ms, max_delay, max_record_size) \
+    sf_binlog_writer_init_ex(context, data_path, subdir_name,       \
+            SF_BINLOG_FILE_PREFIX_STR, buffer_size, write_interval_ms,  \
+            max_delay, max_record_size, true)
+
+void sf_binlog_writer_finish(SFBinlogWriterInfo *writer);
+
+static inline void sf_binlog_writer_destroy_writer(
+        SFBinlogWriterInfo *writer)
+{
+    sf_file_writer_destroy(&writer->fw);
+    if (writer->version_ctx.ring.slots != NULL) {
+        free(writer->version_ctx.ring.slots);
+        writer->version_ctx.ring.slots = NULL;
+    }
+}
+
+static inline void sf_binlog_writer_destroy_thread(
+        SFBinlogWriterThread *thread)
+{
+    fast_mblock_destroy(&thread->mblock);
+    fc_queue_destroy(&thread->queue);
+}
+
+static inline void sf_binlog_writer_destroy(
+        SFBinlogWriterContext *context)
+{
+    sf_binlog_writer_finish(&context->writer);
+    sf_binlog_writer_destroy_writer(&context->writer);
+    sf_binlog_writer_destroy_thread(&context->thread);
 }
 
 int sf_binlog_writer_change_order_by(SFBinlogWriterInfo *writer,
         const short order_by);
 
+int sf_binlog_writer_change_passive_write(SFBinlogWriterInfo *writer,
+        const bool passive_write);
+
+int sf_binlog_writer_change_call_fsync(SFBinlogWriterInfo *writer,
+        const bool call_fsync);
+
 int sf_binlog_writer_change_next_version(SFBinlogWriterInfo *writer,
         const int64_t next_version);
 
+static inline int64_t sf_binlog_writer_get_next_version(
+        SFBinlogWriterInfo *writer)
+{
+    return writer->version_ctx.next;
+}
+
+static inline int sf_binlog_writer_get_waiting_count(
+        SFBinlogWriterInfo *writer)
+{
+    return writer->version_ctx.ring.waiting_count;
+}
+
+static inline int sf_binlog_writer_get_thread_waiting_count(
+        SFBinlogWriterThread *thread)
+{
+    int waiting_count;
+
+    PTHREAD_MUTEX_LOCK(&thread->flow_ctrol.lcp.lock);
+    waiting_count = thread->flow_ctrol.waiting_count;
+    PTHREAD_MUTEX_UNLOCK(&thread->flow_ctrol.lcp.lock);
+
+    return waiting_count;
+}
+
+int sf_binlog_writer_rotate_file_ex(SFBinlogWriterInfo *writer,
+        const bool skip_empty_file);
+
+static inline int sf_binlog_writer_rotate_file(SFBinlogWriterInfo *writer)
+{
+    const bool skip_empty_file = false;
+    return sf_binlog_writer_rotate_file_ex(writer, skip_empty_file);
+}
+
+int sf_binlog_writer_flush_file(SFBinlogWriterInfo *writer);
+
+int sf_binlog_writer_change_write_index(SFBinlogWriterInfo *writer,
+        const int write_index);
+
+int sf_binlog_writer_notify_exit(SFBinlogWriterInfo *writer);
+
 #define sf_binlog_writer_set_flags(writer, flags) \
     sf_file_writer_set_flags(&(writer)->fw, flags)
 
+#define sf_binlog_writer_set_write_done_callback(writer, callback, args) \
+    sf_file_writer_set_write_done_callback(&(writer)->fw, callback, args)
+
+#define sf_binlog_writer_get_last_version_ex(writer, log_level) \
+    sf_file_writer_get_last_version_ex(&(writer)->fw, log_level)
+
 #define sf_binlog_writer_get_last_version(writer) \
     sf_file_writer_get_last_version(&(writer)->fw)
 
-void sf_binlog_writer_finish(SFBinlogWriterInfo *writer);
+#define sf_binlog_writer_get_last_version_silence(writer) \
+    sf_file_writer_get_last_version_silence(&(writer)->fw)
+
+#define sf_binlog_get_indexes(writer, start_index, last_index) \
+    sf_file_writer_get_indexes(&(writer)->fw, start_index, last_index)
+
+#define sf_binlog_get_start_index(writer) \
+    sf_file_writer_get_start_index(&(writer)->fw)
+
+#define sf_binlog_get_last_index(writer) \
+    sf_file_writer_get_last_index(&(writer)->fw)
 
 #define sf_binlog_get_current_write_index(writer) \
-    sf_file_writer_get_current_index(&(writer)->fw)
+    sf_file_writer_get_current_write_index(&(writer)->fw)
 
 #define sf_binlog_get_current_write_position(writer, position) \
     sf_file_writer_get_current_position(&(writer)->fw, position)
@@ -155,7 +284,14 @@ void sf_binlog_writer_finish(SFBinlogWriterInfo *writer);
 static inline SFBinlogWriterBuffer *sf_binlog_writer_alloc_buffer(
         SFBinlogWriterThread *thread)
 {
-    return (SFBinlogWriterBuffer *)fast_mblock_alloc_object(&thread->mblock);
+    SFBinlogWriterBuffer *buffer;
+
+    if ((buffer=(SFBinlogWriterBuffer *)fast_mblock_alloc_object(
+                    &thread->mblock)) != NULL)
+    {
+        buffer->type = SF_BINLOG_BUFFER_TYPE_WRITE_TO_FILE;
+    }
+    return buffer;
 }
 
 #define sf_binlog_writer_alloc_one_version_buffer(writer, version) \
@@ -172,6 +308,7 @@ static inline SFBinlogWriterBuffer *sf_binlog_writer_alloc_versioned_buffer_ex(
         const int64_t last_version, const int type)
 {
     SFBinlogWriterBuffer *buffer;
+
     buffer = (SFBinlogWriterBuffer *)fast_mblock_alloc_object(
             &writer->thread->mblock);
     if (buffer != NULL) {
@@ -183,26 +320,50 @@ static inline SFBinlogWriterBuffer *sf_binlog_writer_alloc_versioned_buffer_ex(
     return buffer;
 }
 
-#define sf_binlog_writer_get_filepath(data_path, subdir_name, filename, size) \
+void sf_push_to_binlog_write_queue(SFBinlogWriterInfo *writer,
-    sf_file_writer_get_filepath(data_path, subdir_name, filename, size)
+        SFBinlogWriterBuffer *buffer);
+
+#define sf_binlog_writer_get_filepath(data_path, subdir_name, filepath, size) \
+    sf_file_writer_get_filepath(data_path, subdir_name, filepath, size)
+
+#define sf_binlog_writer_get_filename_ex(data_path, subdir_name, \
+        file_prefix, binlog_index, filename, size) \
+        sf_file_writer_get_filename_ex(data_path, subdir_name, \
+                file_prefix, binlog_index, filename, size)
 
 #define sf_binlog_writer_get_filename(data_path, \
         subdir_name, binlog_index, filename, size) \
         sf_file_writer_get_filename(data_path, subdir_name, \
                 binlog_index, filename, size)
 
-#define sf_binlog_writer_set_binlog_index(writer, binlog_index) \
+#define sf_binlog_writer_get_index_filename(data_path, \
-    sf_file_writer_set_binlog_index(&(writer)->fw, binlog_index)
+        subdir_name, filename, size) \
+        sf_file_writer_get_index_filename(data_path, \
+                subdir_name, filename, size)
 
-#define sf_push_to_binlog_thread_queue(thread, buffer) \
+#define sf_binlog_writer_get_binlog_indexes(data_path, \
-    fc_queue_push(&(thread)->queue, buffer)
+        subdir_name, start_index, last_index) \
+        sf_file_writer_get_binlog_indexes(data_path, \
+                subdir_name, start_index, last_index)
 
-static inline void sf_push_to_binlog_write_queue(SFBinlogWriterInfo *writer,
+#define sf_binlog_writer_get_binlog_start_index(data_path, \
-        SFBinlogWriterBuffer *buffer)
+        subdir_name, start_index) \
-{
+        sf_file_writer_get_binlog_start_index(data_path, \
-    buffer->type = SF_BINLOG_BUFFER_TYPE_WRITE_TO_FILE;
+                subdir_name, start_index)
-    fc_queue_push(&writer->thread->queue, buffer);
+
-}
+#define sf_binlog_writer_get_binlog_last_index(data_path, \
+        subdir_name, last_index) \
+        sf_file_writer_get_binlog_last_index(data_path, \
+                subdir_name, last_index)
+
+#define sf_binlog_set_indexes(writer, start_index, last_index) \
+    sf_file_writer_set_indexes(&(writer)->fw, start_index, last_index)
+
+#define sf_binlog_writer_set_binlog_start_index(writer, start_index) \
+    sf_file_writer_set_binlog_start_index(&(writer)->fw, start_index)
+
+#define sf_binlog_writer_set_binlog_write_index(writer, last_index) \
+    sf_file_writer_set_binlog_write_index(&(writer)->fw, last_index)
 
 #define sf_binlog_writer_get_last_lines(data_path, subdir_name, \
         current_write_index, buff, buff_size, count, length)  \

src/sf_buffered_writer.h

@@ -0,0 +1,106 @@
+/*
+ * Copyright (c) 2020 YuQing <384681@qq.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <https://www.gnu.org/licenses/>.
+ */
+
+//sf_buffered_writer.h
+
+#ifndef _SF_BUFFERED_WRITER_H_
+#define _SF_BUFFERED_WRITER_H_
+
+#include "sf_types.h"
+#include "sf_func.h"
+
+typedef struct {
+    int fd;
+    const char *filename;
+    SFBinlogBuffer buffer;
+} SFBufferedWriter;
+
+#define sf_buffered_writer_init(writer, filename) \
+    sf_buffered_writer_init_ex(writer, filename, 1024 * 1024)
+
+#define SF_BUFFERED_WRITER_LENGTH(bw)  \
+    SF_BINLOG_BUFFER_PRODUCER_DATA_LENGTH((bw).buffer)
+
+#define SF_BUFFERED_WRITER_REMAIN(bw)  \
+    SF_BINLOG_BUFFER_PRODUCER_BUFF_REMAIN((bw).buffer)
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+    static inline int sf_buffered_writer_init_ex(SFBufferedWriter *writer,
+            const char *filename, const int buffer_size)
+    {
+        int result;
+
+        writer->filename = filename;
+        writer->fd = open(filename, O_WRONLY | O_CREAT |
+                O_TRUNC | O_CLOEXEC, 0644);
+        if (writer->fd < 0) {
+            result = errno != 0 ? errno : EIO;
+            logError("file: "__FILE__", line: %d, "
+                    "open file %s fail, errno: %d, error info: %s",
+                    __LINE__, filename, result, STRERROR(result));
+            return result;
+        }
+
+        if ((result=sf_binlog_buffer_init(&writer->buffer, buffer_size)) != 0) {
+            return result;
+        }
+        return 0;
+    }
+
+    static inline int sf_buffered_writer_save(SFBufferedWriter *writer)
+    {
+        int result;
+        int length;
+
+        length = writer->buffer.data_end - writer->buffer.buff;
+        if (fc_safe_write(writer->fd, writer->buffer.buff, length) != length) {
+            result = errno != 0 ? errno : EIO;
+            logError("file: "__FILE__", line: %d, "
+                    "write to file %s fail, errno: %d, error info: %s",
+                    __LINE__, writer->filename, result, STRERROR(result));
+            return result;
+        }
+
+        writer->buffer.data_end = writer->buffer.buff;
+        return 0;
+    }
+
+    static inline int sf_buffered_writer_destroy(SFBufferedWriter *writer)
+    {
+        int result;
+
+        if (writer->fd >= 0) {
+            if (fsync(writer->fd) != 0) {
+                result = errno != 0 ? errno : EIO;
+                logError("file: "__FILE__", line: %d, "
+                        "fsync to file %s fail, errno: %d, error info: %s",
+                        __LINE__, writer->filename, result, STRERROR(result));
+                return result;
+            }
+            close(writer->fd);
+            writer->fd = -1;
+        }
+        sf_binlog_buffer_destroy(&writer->buffer);
+        return 0;
+    }
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif

src/sf_cluster_cfg.c

@@ -23,7 +23,7 @@ static int calc_cluster_config_sign(SFClusterConfig *cluster)
     FastBuffer buffer;
     int result;
 
-    if ((result=fast_buffer_init_ex(&buffer, 1024)) != 0) {
+    if ((result=fast_buffer_init1(&buffer, 1024)) != 0) {
         return result;
     }
     fc_server_to_config_string(&cluster->server_cfg, &buffer);
@@ -116,19 +116,19 @@ int sf_load_cluster_config_by_file(SFClusterConfig *cluster,
     return 0;
 }
 
-int sf_load_cluster_config_ex(SFClusterConfig *cluster, IniFullContext
+int sf_load_cluster_config_ex1(SFClusterConfig *cluster,
-        *ini_ctx, const int default_port, char *full_cluster_filename,
+        IniFullContext *ini_ctx, const char *cluster_config_item_name,
-        const int size)
+        const int default_port, char *full_cluster_filename, const int size)
 {
     const bool share_between_groups = true;
     char *cluster_config_filename;
 
     cluster_config_filename = iniGetStrValue(ini_ctx->section_name,
-            "cluster_config_filename", ini_ctx->context);
+            cluster_config_item_name, ini_ctx->context);
     if (cluster_config_filename == NULL || *cluster_config_filename == '\0') {
         logError("file: "__FILE__", line: %d, "
-                "config file: %s, item \"cluster_config_filename\" "
+                "config file: %s, item \"%s\" not exist or empty",
-                "not exist or empty", __LINE__, ini_ctx->filename);
+                __LINE__, cluster_config_item_name, ini_ctx->filename);
         return ENOENT;
     }
 

src/sf_cluster_cfg.h

@@ -23,9 +23,29 @@
 extern "C" {
 #endif
 
-int sf_load_cluster_config_ex(SFClusterConfig *cluster, IniFullContext
+int sf_load_cluster_config_ex1(SFClusterConfig *cluster,
-        *ini_ctx, const int default_port, char *full_cluster_filename,
+        IniFullContext *ini_ctx, const char *cluster_config_item_name,
-        const int size);
+        const int default_port, char *full_cluster_filename, const int size);
+
+static inline int sf_load_cluster_config_ex(SFClusterConfig *cluster,
+        IniFullContext *ini_ctx, const int default_port,
+        char *full_cluster_filename, const int size)
+{
+    const char *cluster_config_item_name = "cluster_config_filename";
+    return sf_load_cluster_config_ex1(cluster, ini_ctx,
+            cluster_config_item_name, default_port,
+            full_cluster_filename, PATH_MAX);
+}
+
+static inline int sf_load_cluster_config1(SFClusterConfig *cluster,
+        IniFullContext *ini_ctx, const char *cluster_config_item_name,
+        const int default_port)
+{
+    char full_cluster_filename[PATH_MAX];
+    return sf_load_cluster_config_ex1(cluster, ini_ctx,
+            cluster_config_item_name, default_port,
+            full_cluster_filename, PATH_MAX);
+}
 
 static inline int sf_load_cluster_config(SFClusterConfig *cluster,
         IniFullContext *ini_ctx, const int default_port)

src/sf_configs.c

@@ -25,10 +25,10 @@
 #include "fastcommon/logger.h"
 #include "sf_configs.h"
 
-#define DEFAULT_RETRY_MAX_INTERVAL_MS      5000
+#define DEFAULT_RETRY_MAX_INTERVAL_MS      3000
-#define DEFAULT_CONNECT_RETRY_TIMES          10
+#define DEFAULT_CONNECT_RETRY_TIMES         200
 #define DEFAULT_CONNECT_RETRY_INTERVAL_MS   100
-#define DEFAULT_NETWORK_RETRY_TIMES          10
+#define DEFAULT_NETWORK_RETRY_TIMES         200
 #define DEFAULT_NETWORK_RETRY_INTERVAL_MS   100
 
 int sf_load_net_retry_config(SFNetRetryConfig *net_retry_cfg,
@@ -94,13 +94,14 @@ void sf_net_retry_config_to_string(SFNetRetryConfig *net_retry_cfg,
             net_retry_cfg->network.interval_ms);
 }
 
-void sf_load_read_rule_config_ex(SFDataReadRule *rule,
+int sf_load_read_rule_config_ex(SFDataReadRule *rule,
         IniFullContext *ini_ctx, const SFDataReadRule def_rule)
 {
     char *read_rule;
+
     read_rule = iniGetStrValueEx(ini_ctx->section_name,
             "read_rule", ini_ctx->context, true);
-    if (read_rule == NULL || *read_rule == '\0') {
+    if (read_rule == NULL) {
         *rule = def_rule;
     } else if (strncasecmp(read_rule, "any", 3) == 0) {
         *rule = sf_data_read_rule_any_available;
@@ -110,8 +111,62 @@ void sf_load_read_rule_config_ex(SFDataReadRule *rule,
         *rule = sf_data_read_rule_master_only;
     } else {
         logError("file: "__FILE__", line: %d, "
-                "config file: %s, unkown read_rule: %s, set to any",
+                "config file: %s, unkown read_rule: %s",
                 __LINE__, ini_ctx->filename, read_rule);
-        *rule = sf_data_read_rule_any_available;
+        return EINVAL;
     }
+
+    return 0;
+}
+
+int sf_load_election_quorum_config_ex(SFElectionQuorum *quorum,
+        IniFullContext *ini_ctx, const SFElectionQuorum def_quorum)
+{
+    char *str;
+
+    str = iniGetStrValue(ini_ctx->section_name,
+            "quorum", ini_ctx->context);
+    if (str == NULL) {
+        *quorum = def_quorum;
+    } else if (strncasecmp(str, "auto", 4) == 0) {
+        *quorum = sf_election_quorum_auto;
+    } else if (strncasecmp(str, "any", 3) == 0) {
+        *quorum = sf_election_quorum_any;
+    } else if (strncasecmp(str, "majority", 8) == 0) {
+        *quorum = sf_election_quorum_majority;
+    } else {
+        logError("file: "__FILE__", line: %d, "
+                "config file: %s, unkown quorum: %s",
+                __LINE__, ini_ctx->filename, str);
+        return EINVAL;
+    }
+
+    return 0;
+}
+
+int sf_load_replication_quorum_config_ex(SFReplicationQuorum *quorum,
+        IniFullContext *ini_ctx, const SFReplicationQuorum def_quorum)
+{
+    char *str;
+
+    str = iniGetStrValue(ini_ctx->section_name,
+            "quorum", ini_ctx->context);
+    if (str == NULL) {
+        *quorum = def_quorum;
+    } else if (strncasecmp(str, "auto", 4) == 0) {
+        *quorum = sf_replication_quorum_auto;
+    } else if (strncasecmp(str, "any", 3) == 0) {
+        *quorum = sf_replication_quorum_any;
+    } else if (strncasecmp(str, "majority", 8) == 0) {
+        *quorum = sf_replication_quorum_majority;
+    } else if (strncasecmp(str, "smart", 5) == 0) {
+        *quorum = sf_replication_quorum_smart;
+    } else {
+        logError("file: "__FILE__", line: %d, "
+                "config file: %s, unkown quorum: %s",
+                __LINE__, ini_ctx->filename, str);
+        return EINVAL;
+    }
+
+    return 0;
 }

src/sf_configs.h

@@ -67,7 +67,7 @@ static inline int sf_calc_next_retry_interval(SFNetRetryIntervalContext *ctx)
     return ctx->interval_ms;
 }
 
-void sf_load_read_rule_config_ex(SFDataReadRule *rule,
+int sf_load_read_rule_config_ex(SFDataReadRule *rule,
         IniFullContext *ini_ctx, const SFDataReadRule def_rule);
 
 static inline const char *sf_get_read_rule_caption(
@@ -85,9 +85,113 @@ static inline const char *sf_get_read_rule_caption(
     }
 }
 
+int sf_load_election_quorum_config_ex(SFElectionQuorum *quorum,
+        IniFullContext *ini_ctx, const SFElectionQuorum def_quorum);
+
+static inline const char *sf_get_election_quorum_caption(
+        const SFElectionQuorum quorum)
+{
+    switch (quorum) {
+        case sf_election_quorum_auto:
+            return "auto";
+        case sf_election_quorum_any:
+            return "any";
+        case sf_election_quorum_majority:
+            return "majority";
+        default:
+            return "unknown";
+    }
+}
+
+static inline bool sf_election_quorum_check(const SFElectionQuorum quorum,
+        const bool vote_node_enabled, const int total_count,
+        const int active_count)
+{
+    switch (quorum) {
+        case sf_election_quorum_any:
+            return active_count > 0;
+        case sf_election_quorum_auto:
+            if (total_count % 2 == 0 && !vote_node_enabled) {
+                return active_count > 0;  //same as sf_election_quorum_any
+            }
+            //continue
+        case sf_election_quorum_majority:
+            if (active_count == total_count) {
+                return true;
+            } else {
+                return active_count > total_count / 2;
+            }
+    }
+}
+
+int sf_load_replication_quorum_config_ex(SFReplicationQuorum *quorum,
+        IniFullContext *ini_ctx, const SFReplicationQuorum def_quorum);
+
+static inline const char *sf_get_replication_quorum_caption(
+        const SFReplicationQuorum quorum)
+{
+    switch (quorum) {
+        case sf_replication_quorum_auto:
+            return "auto";
+        case sf_replication_quorum_any:
+            return "any";
+        case sf_replication_quorum_majority:
+            return "majority";
+        case sf_replication_quorum_smart:
+            return "smart";
+        default:
+            return "unknown";
+    }
+}
+
+#define SF_REPLICATION_QUORUM_MAJORITY(server_count, success_count) \
+    ((success_count == server_count) || (success_count > server_count / 2))
+
+static inline bool sf_replication_quorum_check(const SFReplicationQuorum
+        quorum, const int server_count, const int success_count)
+{
+    switch (quorum) {
+        case sf_replication_quorum_any:
+            return true;
+        case sf_replication_quorum_auto:
+            if (server_count % 2 == 0) {
+                return true;  //same as sf_replication_quorum_any
+            }
+            //continue
+        case sf_replication_quorum_smart:
+        case sf_replication_quorum_majority:
+            return SF_REPLICATION_QUORUM_MAJORITY(
+                    server_count, success_count);
+    }
+}
+
 #define sf_load_read_rule_config(rule, ini_ctx) \
     sf_load_read_rule_config_ex(rule, ini_ctx, sf_data_read_rule_master_only)
 
+#define sf_load_election_quorum_config(quorum, ini_ctx) \
+    sf_load_election_quorum_config_ex(quorum, ini_ctx, sf_election_quorum_auto)
+
+#define sf_load_replication_quorum_config(quorum, ini_ctx) \
+    sf_load_replication_quorum_config_ex(quorum, ini_ctx,  \
+            sf_replication_quorum_auto)
+
+#define SF_ELECTION_QUORUM_NEED_REQUEST_VOTE_NODE(quorum, \
+        vote_node_enabled, server_count, active_count) \
+    (active_count < server_count && vote_node_enabled && \
+     quorum != sf_election_quorum_any && server_count % 2 == 0)
+
+#define SF_ELECTION_QUORUM_NEED_CHECK_VOTE_NODE(quorum, \
+        vote_node_enabled, server_count) \
+    (vote_node_enabled && quorum != sf_election_quorum_any \
+     && server_count % 2 == 0)
+
+#define SF_REPLICATION_QUORUM_NEED_MAJORITY(quorum, server_count) \
+    (server_count > 1 && (quorum != sf_replication_quorum_any))
+
+#define SF_REPLICATION_QUORUM_NEED_DETECT(quorum, server_count)  \
+    (server_count % 2 == 0 && (quorum == sf_replication_quorum_smart || \
+                               quorum == sf_replication_quorum_auto))
+
 #define SF_NET_RETRY_FINISHED(retry_times, counter, result)  \
         !((SF_IS_RETRIABLE_ERROR(result) && ((retry_times > 0 &&  \
                     counter <= retry_times) || (retry_times < 0))))

src/sf_connection_manager.c

@@ -35,13 +35,14 @@ static int get_group_servers(SFConnectionManager *cm,
         SFCMConnGroupEntry *group);
 
 static ConnectionInfo *get_spec_connection(SFConnectionManager *cm,
-        const ConnectionInfo *target, int *err_no)
+        const ConnectionInfo *target, const bool shared, int *err_no)
 {
-    return conn_pool_get_connection(&cm->cpool, target, err_no);
+    return conn_pool_get_connection_ex(&cm->cpool,
+            target, cm->module_name, shared, err_no);
 }
 
 static ConnectionInfo *make_connection(SFConnectionManager *cm,
-        FCAddressPtrArray *addr_array, int *err_no)
+        FCAddressPtrArray *addr_array, const bool shared, int *err_no)
 {
     FCAddressInfo **current;
     FCAddressInfo **addr;
@@ -55,7 +56,7 @@ static ConnectionInfo *make_connection(SFConnectionManager *cm,
 
     current = addr_array->addrs + addr_array->index;
     if ((conn=get_spec_connection(cm, &(*current)->conn,
-                    err_no)) != NULL)
+                    shared, err_no)) != NULL)
     {
         return conn;
     }
@@ -71,7 +72,7 @@ static ConnectionInfo *make_connection(SFConnectionManager *cm,
         }
 
         if ((conn=get_spec_connection(cm, &(*addr)->conn,
-                        err_no)) != NULL)
+                        shared, err_no)) != NULL)
         {
             addr_array->index = addr - addr_array->addrs;
             return conn;
@@ -82,22 +83,22 @@ static ConnectionInfo *make_connection(SFConnectionManager *cm,
 }
 
 static ConnectionInfo *get_server_connection(SFConnectionManager *cm,
-        FCServerInfo *server, int *err_no)
+        FCServerInfo *server, const bool shared, int *err_no)
 {
     FCAddressPtrArray *addr_array;
     ConnectionInfo *conn;
 
     addr_array = &server->group_addrs[cm->server_group_index].address_array;
-    if ((conn=make_connection(cm, addr_array, err_no)) == NULL) {
+    if ((conn=make_connection(cm, addr_array, shared, err_no)) == NULL) {
         logError("file: "__FILE__", line: %d, "
-                "server id: %d, get_server_connection fail",
+                "%s server id: %d, address count: %d, get_server_connection fail",
-                __LINE__, server->id);
+                __LINE__, cm->module_name, server->id, addr_array->count);
     }
     return conn;
 }
 
 static ConnectionInfo *get_connection(SFConnectionManager *cm,
-        const int group_index, int *err_no)
+        const int group_index, const bool shared, int *err_no)
 {
     SFCMServerArray *server_array;
     ConnectionInfo *conn;
@@ -109,7 +110,7 @@ static ConnectionInfo *get_connection(SFConnectionManager *cm,
     server_hash_code = rand();
     server_index = server_hash_code % server_array->count;
     if ((conn=make_connection(cm, server_array->servers[server_index].
-                    addr_array, err_no)) != NULL)
+                    addr_array, shared, err_no)) != NULL)
     {
         return conn;
     }
@@ -121,7 +122,7 @@ static ConnectionInfo *get_connection(SFConnectionManager *cm,
             }
 
             if ((conn=make_connection(cm, server_array->servers[i].
-                            addr_array, err_no)) != NULL)
+                            addr_array, shared, err_no)) != NULL)
             {
                 return conn;
             }
@@ -129,8 +130,8 @@ static ConnectionInfo *get_connection(SFConnectionManager *cm,
     }
 
     logError("file: "__FILE__", line: %d, "
-            "data group index: %d, get_connection fail, "
+            "%s data group index: %d, get_connection fail, "
-            "configured server count: %d", __LINE__,
+            "configured server count: %d", __LINE__, cm->module_name,
             group_index, server_array->count);
     return NULL;
 }
@@ -144,33 +145,6 @@ static inline void set_connection_params(ConnectionInfo *conn,
     cparam->cm.old_alives = old_alives;
 }
 
-static inline ConnectionInfo *make_master_connection(SFConnectionManager *cm,
-        SFCMConnGroupEntry *group, int *err_no)
-{
-    SFCMServerEntry *master;
-    ConnectionInfo *conn;
-
-    master = (SFCMServerEntry *)FC_ATOMIC_GET(group->master);
-    if (master != NULL) {
-        if ((conn=make_connection(cm, master->addr_array,
-                        err_no)) != NULL)
-        {
-            if (cm->common_cfg->read_rule == sf_data_read_rule_master_only) {
-                set_connection_params(conn, master, NULL);
-            } else {
-                SFCMServerPtrArray *alives;
-                alives = (SFCMServerPtrArray *)FC_ATOMIC_GET(group->alives);
-                set_connection_params(conn, master, alives);
-            }
-            return conn;
-        }
-
-        __sync_bool_compare_and_swap(&group->master, master, NULL);
-    }
-
-    return NULL;
-}
-
 static inline int push_to_detect_queue(SFConnectionManager *cm,
         SFCMConnGroupEntry *group, SFCMServerPtrArray *alives)
 {
@@ -202,7 +176,7 @@ static inline bool alive_array_cas(SFConnectionManager *cm,
         push_to_detect_queue(cm, group, new_alives);
         fast_mblock_delay_free_object(&cm->sptr_array_allocator, old_alives,
                 (cm->common_cfg->connect_timeout + cm->common_cfg->
-                 network_timeout) * group->all.count);
+                 network_timeout) * group->all.count * 4);
         return true;
     } else {
         fast_mblock_free_object(&cm->sptr_array_allocator, new_alives);
@@ -250,14 +224,42 @@ static int remove_from_alives(SFConnectionManager *cm,
     return 0;
 }
 
+static inline ConnectionInfo *make_master_connection(SFConnectionManager *cm,
+        SFCMConnGroupEntry *group, const bool shared, int *err_no)
+{
+    SFCMServerEntry *master;
+    ConnectionInfo *conn;
+    SFCMServerPtrArray *alives;
+
+    master = (SFCMServerEntry *)FC_ATOMIC_GET(group->master);
+    if (master != NULL) {
+        if ((conn=make_connection(cm, master->addr_array,
+                        shared, err_no)) != NULL)
+        {
+            alives = (SFCMServerPtrArray *)FC_ATOMIC_GET(group->alives);
+            set_connection_params(conn, master, alives);
+            return conn;
+        } else {
+            alives = (SFCMServerPtrArray *)FC_ATOMIC_GET(group->alives);
+            if (alives != NULL) {
+                remove_from_alives(cm, group, alives, master);
+            }
+            __sync_bool_compare_and_swap(&group->master, master, NULL);
+        }
+    }
+
+    *err_no = SF_RETRIABLE_ERROR_NO_SERVER;
+    return NULL;
+}
+
 static inline ConnectionInfo *make_readable_connection(SFConnectionManager *cm,
         SFCMConnGroupEntry *group, SFCMServerPtrArray *alives,
-        const int index, int *err_no)
+        const int index, const bool shared, int *err_no)
 {
     ConnectionInfo *conn;
 
     if ((conn=make_connection(cm, alives->servers[index]->
-                    addr_array, err_no)) == NULL)
+                    addr_array, shared, err_no)) == NULL)
     {
         remove_from_alives(cm, group, alives, alives->servers[index]);
     } else {
@@ -268,72 +270,82 @@ static inline ConnectionInfo *make_readable_connection(SFConnectionManager *cm,
 }
 
 static ConnectionInfo *get_master_connection(SFConnectionManager *cm,
-        const int group_index, int *err_no)
+        const int group_index, const bool shared, int *err_no)
 {
     SFCMConnGroupEntry *group;
     ConnectionInfo *conn;
     SFNetRetryIntervalContext net_retry_ctx;
-    int i;
+    int retry_count;
 
     group = cm->groups.entries + group_index;
     sf_init_net_retry_interval_context(&net_retry_ctx,
             &cm->common_cfg->net_retry_cfg.interval_mm,
             &cm->common_cfg->net_retry_cfg.connect);
-    i = 0;
+    retry_count = 0;
     while (1) {
-        if ((conn=make_master_connection(cm, group, err_no)) != NULL) {
+        if ((conn=make_master_connection(cm, group, shared, err_no)) != NULL) {
             return conn;
         }
 
+        /*
+        logInfo("file: "__FILE__", line: %d, "
+                "retry_count: %d, interval_ms: %d, data group id: %d, "
+                "master: %p, alive count: %d, all count: %d", __LINE__,
+                retry_count, net_retry_ctx.interval_ms, group->id,
+                FC_ATOMIC_GET(group->master), ((SFCMServerPtrArray *)
+                    FC_ATOMIC_GET(group->alives))->count, group->all.count);
+         */
+
         *err_no = get_group_servers(cm, group);
         if (*err_no == 0) {
             *err_no = SF_RETRIABLE_ERROR_NO_SERVER;  //for try again
         }
         SF_NET_RETRY_CHECK_AND_SLEEP(net_retry_ctx,
                 cm->common_cfg->net_retry_cfg.
-                connect.times, ++i, *err_no);
+                connect.times, ++retry_count, *err_no);
     }
 
     logError("file: "__FILE__", line: %d, "
-            "get_master_connection fail, errno: %d",
+            "%s get_master_connection fail, group id: %d, "
-            __LINE__, *err_no);
+            "retry count: %d, " "errno: %d", __LINE__,
+            cm->module_name, group->id, retry_count, *err_no);
     return NULL;
 }
 
 static ConnectionInfo *get_readable_connection(SFConnectionManager *cm,
-        const int group_index, int *err_no)
+        const int group_index, const bool shared, int *err_no)
 {
     SFCMConnGroupEntry *group;
     SFCMServerPtrArray *alives;
     ConnectionInfo *conn;
     SFNetRetryIntervalContext net_retry_ctx;
     uint32_t index;
-    int i;
+    int retry_count;
 
     group = cm->groups.entries + group_index;
     if ((cm->common_cfg->read_rule == sf_data_read_rule_master_only) ||
             (group->all.count == 1))
     {
-        return get_master_connection(cm, group_index, err_no);
+        return get_master_connection(cm, group_index, shared, err_no);
     }
 
     sf_init_net_retry_interval_context(&net_retry_ctx,
             &cm->common_cfg->net_retry_cfg.interval_mm,
             &cm->common_cfg->net_retry_cfg.connect);
-    i = 0;
+    retry_count = 0;
     while (1) {
         alives = (SFCMServerPtrArray *)FC_ATOMIC_GET(group->alives);
         if (alives->count > 0) {
             index = rand() % alives->count;
             if ((conn=make_readable_connection(cm, group, alives,
-                            index, err_no)) != NULL)
+                            index, shared, err_no)) != NULL)
             {
                 return conn;
             }
         }
 
         if (cm->common_cfg->read_rule == sf_data_read_rule_slave_first) {
-            if ((conn=make_master_connection(cm, group, err_no)) != NULL) {
+            if ((conn=make_master_connection(cm, group, shared, err_no)) != NULL) {
                 return conn;
             }
         }
@@ -344,12 +356,12 @@ static ConnectionInfo *get_readable_connection(SFConnectionManager *cm,
         }
         SF_NET_RETRY_CHECK_AND_SLEEP(net_retry_ctx,
                 cm->common_cfg->net_retry_cfg.
-                connect.times, ++i, *err_no);
+                connect.times, ++retry_count, *err_no);
     }
 
     logError("file: "__FILE__", line: %d, "
-            "get_readable_connection fail, errno: %d",
+            "%s get_readable_connection fail, retry count: %d, errno: %d",
-            __LINE__, *err_no);
+            __LINE__, cm->module_name, retry_count, *err_no);
     return NULL;
 }
 
@@ -376,12 +388,11 @@ static void close_connection(SFConnectionManager *cm, ConnectionInfo *conn)
     if (cparam->cm.sentry != NULL) {
         server = cparam->cm.sentry;
         group = cm->groups.entries + server->group_index;
-        if (cparam->cm.old_alives == NULL) {
+        if (cparam->cm.old_alives != NULL) {
-            __sync_bool_compare_and_swap(&group->master, server, NULL);
-        } else {
             remove_from_alives(cm, group, cparam->cm.old_alives, server);
             cparam->cm.old_alives = NULL;
         }
+        __sync_bool_compare_and_swap(&group->master, server, NULL);
         cparam->cm.sentry = NULL;
     }
 
@@ -389,11 +400,12 @@ static void close_connection(SFConnectionManager *cm, ConnectionInfo *conn)
 }
 
 static ConnectionInfo *get_leader_connection(SFConnectionManager *cm,
-        FCServerInfo *server, int *err_no)
+        FCServerInfo *server, const bool shared, int *err_no)
 {
     ConnectionInfo *conn;
     SFClientServerEntry leader;
     SFNetRetryIntervalContext net_retry_ctx;
+    char formatted_ip[FORMATTED_IP_SIZE];
     int i;
     int connect_fails;
 
@@ -404,14 +416,14 @@ static ConnectionInfo *get_leader_connection(SFConnectionManager *cm,
     while (1) {
         do {
             if ((conn=get_server_connection(cm, server,
-                            err_no)) == NULL)
+                            shared, err_no)) == NULL)
             {
                 connect_fails++;
                 break;
             }
 
-            if ((*err_no=sf_proto_get_leader(conn, cm->common_cfg->
+            if ((*err_no=sf_proto_get_leader(conn, cm->module_name, cm->
-                    network_timeout, &leader)) != 0)
+                            common_cfg->network_timeout, &leader)) != 0)
             {
                 close_connection(cm, conn);
                 break;
@@ -421,9 +433,27 @@ static ConnectionInfo *get_leader_connection(SFConnectionManager *cm,
                 return conn;
             }
             release_connection(cm, conn);
-            if ((conn=get_spec_connection(cm,&leader.conn,
+
-                            err_no)) == NULL)
+            if ((conn=get_spec_connection(cm, &leader.conn,
+                            shared, err_no)) == NULL)
             {
+                if (cm->server_cfg != NULL) {
+                    FCServerInfo *ls;
+                    if ((ls=fc_server_get_by_id(cm->server_cfg,
+                                    leader.server_id)) != NULL)
+                    {
+                        if (ls->group_addrs[cm->server_group_index].
+                                address_array.count > 1)
+                        {
+                            if ((conn=get_server_connection(cm, ls,
+                                            shared, err_no)) != NULL)
+                            {
+                                return conn;
+                            }
+                        }
+                    }
+                }
+
                 break;
             }
 
@@ -439,11 +469,13 @@ static ConnectionInfo *get_leader_connection(SFConnectionManager *cm,
                 connect.times, ++i, *err_no);
     }
 
+    format_ip_address(server->group_addrs[cm->server_group_index].
+            address_array.addrs[0]->conn.ip_addr, formatted_ip);
     logWarning("file: "__FILE__", line: %d, "
-            "get_leader_connection fail, server id: %d, %s:%u, errno: %d",
+            "%s get_leader_connection fail, server id: %d, %s:%u, errno: %d",
-            __LINE__, server->id, server->group_addrs[cm->server_group_index].
+            __LINE__, cm->module_name, server->id, formatted_ip,
-            address_array.addrs[0]->conn.ip_addr, server->group_addrs[cm->
+            server->group_addrs[cm->server_group_index].address_array.
-            server_group_index].address_array.addrs[0]->conn.port, *err_no);
+            addrs[0]->conn.port, *err_no);
     return NULL;
 }
 
@@ -463,7 +495,7 @@ int sf_cm_validate_connection_callback(ConnectionInfo *conn, void *args)
     if ((result=sf_active_test(conn, &response, cm->common_cfg->
                     network_timeout)) != 0)
     {
-        sf_log_network_error(&response, conn, result);
+        sf_log_network_error(&response, conn, cm->module_name, result);
     }
 
     return result;
@@ -489,21 +521,52 @@ int sf_connection_manager_init_ex(SFConnectionManager *cm,
         const int group_count, const int server_group_index,
         const int server_count, const int max_count_per_entry,
         const int max_idle_time, fc_connection_callback_func
-        connect_done_callback, void *args, const bool bg_thread_enabled)
+        connect_done_callback, void *args, FCServerConfig *server_cfg,
+        const bool bg_thread_enabled)
 {
-    const int socket_domain = AF_INET;
+    struct {
-    int htable_init_capacity;
+        ConnectionExtraParams holder;
+        ConnectionExtraParams *ptr;
+    } extra_params;
+    FCServerGroupInfo *server_group;
+
+    int htable_capacity;
     int result;
 
-    htable_init_capacity = 4 * server_count;
+    if (server_count <= 4) {
-    if (htable_init_capacity < 256) {
+        htable_capacity = 16;
-        htable_init_capacity = 256;
+    } else if (server_count <= 16) {
+        htable_capacity = 64;
+    } else if (server_count <= 32) {
+        htable_capacity = 128;
+    } else if (server_count < 64) {
+        htable_capacity = 256;
+    } else {
+        htable_capacity = 4 * server_count;
+    }
+
+    if ((server_group=fc_server_get_group_by_index(server_cfg,
+                    server_group_index)) == NULL)
+    {
+        return ENOENT;
+    }
+
+    if (server_group->comm_type == fc_comm_type_sock) {
+        extra_params.ptr = NULL;
+    } else {
+        if ((result=conn_pool_set_rdma_extra_params(&extra_params.holder,
+                        server_cfg, server_group_index)) != 0)
+        {
+            return result;
+        }
+        extra_params.ptr = &extra_params.holder;
     }
     if ((result=conn_pool_init_ex1(&cm->cpool, common_cfg->connect_timeout,
-                    max_count_per_entry, max_idle_time, socket_domain,
+                    max_count_per_entry, max_idle_time, htable_capacity,
-                    htable_init_capacity, connect_done_callback, args,
+                    connect_done_callback, args,
                     sf_cm_validate_connection_callback, cm,
-                    sizeof(SFConnectionParameters))) != 0)
+                    sizeof(SFConnectionParameters),
+                    extra_params.ptr)) != 0)
     {
         return result;
     }
@@ -523,9 +586,11 @@ int sf_connection_manager_init_ex(SFConnectionManager *cm,
     cm->server_group_index = server_group_index;
     cm->module_name = module_name;
     cm->common_cfg = common_cfg;
+    cm->server_cfg = server_cfg;
     cm->alive_detect.bg_thread_enabled = bg_thread_enabled;
     cm->max_servers_per_group = 0;
     cm->extra = NULL;
+    cm->exclude_server_id = 0;
 
     cm->ops.get_connection = get_connection;
     cm->ops.get_server_connection = get_server_connection;
@@ -687,8 +752,8 @@ static int do_get_group_servers(SFConnectionManager *cm,
     sarray.alloc = MAX_GROUP_SERVER_COUNT;
     sarray.count = 0;
     sarray.servers = fixed_servers;
-    if ((result=sf_proto_get_group_servers(conn, cm->common_cfg->
+    if ((result=sf_proto_get_group_servers(conn, cm->module_name, cm->
-                    network_timeout, group->id, &sarray)) != 0)
+                    common_cfg->network_timeout, group->id, &sarray)) != 0)
     {
         return result;
     }
@@ -713,6 +778,7 @@ static int do_get_group_servers(SFConnectionManager *cm,
 static int get_group_servers_by_active(SFConnectionManager *cm,
         SFCMConnGroupEntry *group)
 {
+    const bool shared = true;
     SFCMServerPtrArray *alives;
     SFCMServerEntry **server;
     SFCMServerEntry **end;
@@ -728,7 +794,7 @@ static int get_group_servers_by_active(SFConnectionManager *cm,
     end = alives->servers + alives->count;
     for (server=alives->servers; server<end; server++) {
         if ((conn=make_connection(cm, (*server)->addr_array,
-                        &result)) == NULL)
+                        shared, &result)) == NULL)
         {
             continue;
         }
@@ -746,6 +812,7 @@ static int get_group_servers_by_active(SFConnectionManager *cm,
 static int get_group_servers_by_all(SFConnectionManager *cm,
         SFCMConnGroupEntry *group)
 {
+    const bool shared = true;
     SFCMServerEntry *server;
     SFCMServerEntry *end;
     ConnectionInfo *conn;
@@ -758,8 +825,12 @@ static int get_group_servers_by_all(SFConnectionManager *cm,
 
     end = group->all.servers + group->all.count;
     for (server=group->all.servers; server<end; server++) {
+        if (server->id == cm->exclude_server_id) {
+            continue;
+        }
+
         if ((conn=make_connection(cm, server->addr_array,
-                        &result)) == NULL)
+                        shared, &result)) == NULL)
         {
             continue;
         }
@@ -822,7 +893,7 @@ static void *connection_manager_thread_func(void *arg)
 #endif
 
     cm = (SFConnectionManager *)arg;
-    logInfo("file: "__FILE__", line: %d, "
+    logDebug("file: "__FILE__", line: %d, "
             "[%s] connection manager thread start",
             __LINE__, cm->module_name);
 

src/sf_connection_manager.h

@@ -26,16 +26,17 @@
 struct sf_connection_manager;
 
 typedef ConnectionInfo *(*sf_get_connection_func)(
-        struct sf_connection_manager *cm,
+        struct sf_connection_manager *cm, const int group_index,
-        const int group_index, int *err_no);
+        const bool shared, int *err_no);
 
 typedef ConnectionInfo *(*sf_get_server_connection_func)(
         struct sf_connection_manager *cm,
-        FCServerInfo *server, int *err_no);
+        FCServerInfo *server, const bool shared, int *err_no);
 
 typedef ConnectionInfo *(*sf_get_spec_connection_func)(
         struct sf_connection_manager *cm,
-        const ConnectionInfo *target, int *err_no);
+        const ConnectionInfo *target,
+        const bool shared, int *err_no);
 
 typedef void (*sf_release_connection_func)(
         struct sf_connection_manager *cm, ConnectionInfo *conn);
@@ -103,8 +104,9 @@ typedef struct sf_cm_operations {
 } SFCMOperations;
 
 typedef struct sf_connection_manager {
-    short server_group_index;
+    int exclude_server_id;  //for server side
-    short max_servers_per_group;
+    uint16_t max_servers_per_group;
+    uint8_t server_group_index;
     struct {
         bool bg_thread_enabled;
         struct common_blocked_queue queue;
@@ -115,6 +117,7 @@ typedef struct sf_connection_manager {
     ConnectionPool cpool;
     struct fast_mblock_man sptr_array_allocator; //element: SFCMServerPtrArray
     SFCMOperations ops;
+    FCServerConfig *server_cfg;
     void *extra;   //for simple connection manager
 } SFConnectionManager;
 
@@ -123,7 +126,8 @@ int sf_connection_manager_init_ex(SFConnectionManager *cm,
         const int group_count, const int server_group_index,
         const int server_count, const int max_count_per_entry,
         const int max_idle_time, fc_connection_callback_func
-        connect_done_callback, void *args, const bool bg_thread_enabled);
+        connect_done_callback, void *args, FCServerConfig *server_cfg,
+        const bool bg_thread_enabled);
 
 static inline int sf_connection_manager_init(SFConnectionManager *cm,
         const char *module_name, const SFClientCommonConfig *common_cfg,
@@ -136,7 +140,13 @@ static inline int sf_connection_manager_init(SFConnectionManager *cm,
     return sf_connection_manager_init_ex(cm, module_name,
             common_cfg, group_count, server_group_index,
             server_count, max_count_per_entry, max_idle_time,
-            connect_done_callback, args, bg_thread_enabled);
+            connect_done_callback, args, NULL, bg_thread_enabled);
+}
+
+static inline void sf_connection_manager_set_exclude_server_id(
+        SFConnectionManager *cm, const int exclude_server_id)
+{
+    cm->exclude_server_id = exclude_server_id;
 }
 
 int sf_connection_manager_add(SFConnectionManager *cm, const int group_id,

src/sf_define.h

@@ -20,8 +20,8 @@
 
 #include "fastcommon/sockopt.h"
 
-#define SF_DEFAULT_CONNECT_TIMEOUT    2
+#define SF_DEFAULT_CONNECT_TIMEOUT   10
-#define SF_DEFAULT_NETWORK_TIMEOUT   10
+#define SF_DEFAULT_NETWORK_TIMEOUT   60
 
 #define SF_DEF_THREAD_STACK_SIZE  (256 * 1024)
 #define SF_MIN_THREAD_STACK_SIZE  (64 * 1024)
@@ -45,8 +45,9 @@
 #define SF_NIO_TASK_STAGE_FETCH(task)  task->nio_stages.current
 
 #define SF_SESSION_ERROR_NOT_EXIST           9992
-#define SF_CLUSTER_ERROR_NOT_LEADER          9996
+#define SF_CLUSTER_ERROR_NOT_LEADER          9995
-#define SF_CLUSTER_ERROR_LEADER_VERSION_INCONSISTENT 9997
+#define SF_CLUSTER_ERROR_LEADER_VERSION_INCONSISTENT 9996
+#define SF_CLUSTER_ERROR_BINLOG_MISSED       9997
 #define SF_CLUSTER_ERROR_BINLOG_INCONSISTENT 9998
 #define SF_CLUSTER_ERROR_LEADER_INCONSISTENT 9999
 #define SF_CLUSTER_ERROR_MASTER_INCONSISTENT SF_CLUSTER_ERROR_LEADER_INCONSISTENT
@@ -64,9 +65,13 @@
 #define SF_ERROR_EBUSY        8816
 #define SF_ERROR_EINVAL       8822
 #define SF_ERROR_EAGAIN       8835
+#define SF_ERROR_EINPROGRESS  8836
 #define SF_ERROR_EOVERFLOW    8884
 #define SF_ERROR_EOPNOTSUPP   8895
+#define SF_ERROR_ENOLINK      8867
 #define SF_ERROR_ENODATA      8861
+#define SF_ERROR_ENOTEMPTY    8839
+#define SF_ERROR_ELOOP        8840
 
 #define SF_FORCE_CLOSE_CONNECTION_ERROR_MIN  SF_RETRIABLE_ERROR_NOT_MASTER
 #define SF_FORCE_CLOSE_CONNECTION_ERROR_MAX  SF_RETRIABLE_ERROR_MAX
@@ -91,6 +96,8 @@
 #define SF_BINLOG_SOURCE_USER        'U'  //by user call
 #define SF_BINLOG_SOURCE_REPLAY      'R'  //by binlog replay
 
+#define SF_LOG_SCHEDULE_ENTRIES_COUNT 3
+
 #ifdef __cplusplus
 extern "C" {
 #endif

src/sf_file_writer.c

@@ -34,58 +34,138 @@
 #include "sf_func.h"
 #include "sf_file_writer.h"
 
-#define BINLOG_INDEX_FILENAME  SF_BINLOG_FILE_PREFIX"_index.dat"
+#define BINLOG_INDEX_ITEM_START_INDEX_STR       "start_index"
+#define BINLOG_INDEX_ITEM_START_INDEX_LEN      \
+    (sizeof(BINLOG_INDEX_ITEM_START_INDEX_STR) - 1)
 
-#define BINLOG_INDEX_ITEM_CURRENT_WRITE     "current_write"
+#define BINLOG_INDEX_ITEM_CURRENT_WRITE_STR     "current_write"
-#define BINLOG_INDEX_ITEM_CURRENT_COMPRESS  "current_compress"
+#define BINLOG_INDEX_ITEM_CURRENT_WRITE_LEN      \
+    (sizeof(BINLOG_INDEX_ITEM_CURRENT_WRITE_STR) - 1)
 
-#define GET_BINLOG_FILENAME(writer) \
+#define BINLOG_INDEX_ITEM_CURRENT_COMPRESS_STR  "current_compress"
-    sprintf(writer->file.name, "%s/%s/%s"SF_BINLOG_FILE_EXT_FMT,  \
+#define BINLOG_INDEX_ITEM_CURRENT_COMPRESS_LEN      \
-            writer->cfg.data_path, writer->cfg.subdir_name, \
+    (sizeof(BINLOG_INDEX_ITEM_CURRENT_COMPRESS_STR) - 1)
-            SF_BINLOG_FILE_PREFIX, writer->binlog.index)
 
-static int write_to_binlog_index_file(SFFileWriterInfo *writer)
+static inline void sf_file_writer_get_binlog_filename(SFFileWriterInfo *writer)
 {
-    char full_filename[PATH_MAX];
+    sf_file_writer_get_filename_ex(
+            writer->cfg.data_path, writer->cfg.subdir_name,
+            writer->cfg.file_prefix, writer->binlog.last_index,
+            writer->file.name.str, writer->file.name.size);
+}
+
+static inline void sf_file_writer_get_index_filename_ex(
+        const char *data_path, const char *subdir_name,
+        const char *file_prefix, const int file_prefix_len,
+        char *filename, const int size)
+{
+#define INDEX_FILENAME_AFFIX_STR "_index.dat"
+#define INDEX_FILENAME_AFFIX_LEN (sizeof(INDEX_FILENAME_AFFIX_STR) - 1)
+
+    char *p;
+    int data_path_len;
+    int subdir_name_len;
+
+    data_path_len = strlen(data_path);
+    subdir_name_len = strlen(subdir_name);
+    if (data_path_len + 1 + subdir_name_len + 1 + file_prefix_len +
+            INDEX_FILENAME_AFFIX_LEN >= size)
+    {
+        *filename = '\0';
+        return;
+    }
+
+    memcpy(filename, data_path, data_path_len);
+    p = filename + data_path_len;
+    *p++ = '/';
+
+    memcpy(p, subdir_name, subdir_name_len);
+    p += subdir_name_len;
+    *p++ = '/';
+
+    memcpy(p, file_prefix, file_prefix_len);
+    p += file_prefix_len;
+
+    memcpy(p, INDEX_FILENAME_AFFIX_STR, INDEX_FILENAME_AFFIX_LEN);
+    p += INDEX_FILENAME_AFFIX_LEN;
+    *p = '\0';
+}
+
+const char *sf_file_writer_get_index_filename(const char *data_path,
+        const char *subdir_name, char *filename, const int size)
+{
+    sf_file_writer_get_index_filename_ex(data_path, subdir_name,
+            SF_BINLOG_FILE_PREFIX_STR, SF_BINLOG_FILE_PREFIX_LEN,
+            filename, size);
+    return filename;
+}
+
+int sf_file_writer_write_to_binlog_index_file_ex(const char *data_path,
+        const char *subdir_name, const char *file_prefix,
+        const int start_index, const int last_index,
+        const int compress_index)
+{
+    char filename[PATH_MAX];
     char buff[256];
+    char *p;
     int result;
     int len;
 
-    snprintf(full_filename, sizeof(full_filename), "%s/%s/%s",
+    sf_file_writer_get_index_filename_ex(data_path, subdir_name, file_prefix,
-            writer->cfg.data_path, writer->cfg.subdir_name,
+            strlen(file_prefix), filename, sizeof(filename));
-            BINLOG_INDEX_FILENAME);
+    p = buff;
+    memcpy(p, BINLOG_INDEX_ITEM_START_INDEX_STR,
+            BINLOG_INDEX_ITEM_START_INDEX_LEN);
+    p += BINLOG_INDEX_ITEM_START_INDEX_LEN;
+    *p++ = '=';
+    p += fc_itoa(start_index, p);
+    *p++ = '\n';
 
-    len = sprintf(buff, "%s=%d\n"
+    memcpy(p, BINLOG_INDEX_ITEM_CURRENT_WRITE_STR,
-            "%s=%d\n",
+            BINLOG_INDEX_ITEM_CURRENT_WRITE_LEN);
-            BINLOG_INDEX_ITEM_CURRENT_WRITE,
+    p += BINLOG_INDEX_ITEM_CURRENT_WRITE_LEN;
-            writer->binlog.index,
+    *p++ = '=';
-            BINLOG_INDEX_ITEM_CURRENT_COMPRESS,
+    p += fc_itoa(last_index, p);
-            writer->binlog.compress_index);
+    *p++ = '\n';
-    if ((result=safeWriteToFile(full_filename, buff, len)) != 0) {
+
+    memcpy(p, BINLOG_INDEX_ITEM_CURRENT_COMPRESS_STR,
+            BINLOG_INDEX_ITEM_CURRENT_COMPRESS_LEN);
+    p += BINLOG_INDEX_ITEM_CURRENT_COMPRESS_LEN;
+    *p++ = '=';
+    p += fc_itoa(compress_index, p);
+    *p++ = '\n';
+
+    len = p - buff;
+    if ((result=safeWriteToFile(filename, buff, len)) != 0) {
         logError("file: "__FILE__", line: %d, "
-            "write to file \"%s\" fail, "
+                "write to file \"%s\" fail, errno: %d, error info: %s",
-            "errno: %d, error info: %s",
+                __LINE__, filename, result, STRERROR(result));
-            __LINE__, full_filename,
-            result, STRERROR(result));
     }
 
     return result;
 }
 
-static int get_binlog_index_from_file(SFFileWriterInfo *writer)
+static inline int write_to_binlog_index_file(SFFileWriterInfo *writer)
+{
+    return sf_file_writer_write_to_binlog_index_file_ex(
+            writer->cfg.data_path, writer->cfg.subdir_name,
+            writer->cfg.file_prefix, writer->binlog.start_index,
+            writer->binlog.last_index, writer->binlog.compress_index);
+}
+
+static int get_binlog_info_from_file(const char *data_path,
+        const char *subdir_name, int *start_index,
+        int *last_index, int *compress_index)
 {
     char full_filename[PATH_MAX];
     IniContext ini_context;
     int result;
 
-    snprintf(full_filename, sizeof(full_filename), "%s/%s/%s",
+    sf_file_writer_get_index_filename_ex(data_path, subdir_name,
-            writer->cfg.data_path, writer->cfg.subdir_name,
+            SF_BINLOG_FILE_PREFIX_STR, SF_BINLOG_FILE_PREFIX_LEN,
-            BINLOG_INDEX_FILENAME);
+            full_filename, sizeof(full_filename));
     if (access(full_filename, F_OK) != 0) {
-        if (errno == ENOENT) {
+        return errno != 0 ? errno : EPERM;
-            writer->binlog.index = 0;
-            return write_to_binlog_index_file(writer);
-        }
     }
 
     if ((result=iniLoadFromFile(full_filename, &ini_context)) != 0) {
@@ -95,29 +175,71 @@ static int get_binlog_index_from_file(SFFileWriterInfo *writer)
         return result;
     }
 
-    writer->binlog.index = iniGetIntValue(NULL,
+    *start_index = iniGetIntValue(NULL,
-            BINLOG_INDEX_ITEM_CURRENT_WRITE, &ini_context, 0);
+            BINLOG_INDEX_ITEM_START_INDEX_STR,
-    writer->binlog.compress_index = iniGetIntValue(NULL,
+            &ini_context, 0);
-            BINLOG_INDEX_ITEM_CURRENT_COMPRESS, &ini_context, 0);
+    *last_index = iniGetIntValue(NULL,
+            BINLOG_INDEX_ITEM_CURRENT_WRITE_STR,
+            &ini_context, 0);
+    *compress_index = iniGetIntValue(NULL,
+            BINLOG_INDEX_ITEM_CURRENT_COMPRESS_STR,
+            &ini_context, 0);
 
     iniFreeContext(&ini_context);
     return 0;
 }
 
+int sf_file_writer_get_binlog_indexes(const char *data_path,
+        const char *subdir_name, int *start_index, int *last_index)
+{
+    int result;
+    int compress_index;
+
+    result = get_binlog_info_from_file(data_path, subdir_name,
+            start_index, last_index, &compress_index);
+    if (result == ENOENT) {
+        *start_index = *last_index = 0;
+        return 0;
+    } else {
+        return result;
+    }
+}
+
+static inline int get_binlog_index_from_file(SFFileWriterInfo *writer)
+{
+    int result;
+
+    result = get_binlog_info_from_file(writer->cfg.data_path,
+            writer->cfg.subdir_name, &writer->binlog.start_index,
+            &writer->binlog.last_index, &writer->binlog.compress_index);
+    if (result == ENOENT) {
+        writer->binlog.start_index = 0;
+        writer->binlog.last_index = 0;
+        writer->binlog.compress_index = 0;
+        if (writer->cfg.file_rotate_size > 0) {
+            return write_to_binlog_index_file(writer);
+        } else {
+            return 0;
+        }
+    } else {
+        return result;
+    }
+}
+
 static int open_writable_binlog(SFFileWriterInfo *writer)
 {
     if (writer->file.fd >= 0) {
         close(writer->file.fd);
     }
 
-    GET_BINLOG_FILENAME(writer);
+    sf_file_writer_get_binlog_filename(writer);
-    writer->file.fd = open(writer->file.name,
+    writer->file.fd = open(writer->file.name.str, O_WRONLY |
-            O_WRONLY | O_CREAT | O_APPEND, 0644);
+            O_CREAT | O_APPEND | O_CLOEXEC, 0644);
     if (writer->file.fd < 0) {
         logError("file: "__FILE__", line: %d, "
                 "open file \"%s\" fail, "
                 "errno: %d, error info: %s",
-                __LINE__, writer->file.name,
+                __LINE__, writer->file.name.str,
                 errno, STRERROR(errno));
         return errno != 0 ? errno : EACCES;
     }
@@ -127,7 +249,7 @@ static int open_writable_binlog(SFFileWriterInfo *writer)
         logError("file: "__FILE__", line: %d, "
                 "lseek file \"%s\" fail, "
                 "errno: %d, error info: %s",
-                __LINE__, writer->file.name,
+                __LINE__, writer->file.name.str,
                 errno, STRERROR(errno));
         return errno != 0 ? errno : EIO;
     }
@@ -137,23 +259,24 @@ static int open_writable_binlog(SFFileWriterInfo *writer)
 
 static int open_next_binlog(SFFileWriterInfo *writer)
 {
-    GET_BINLOG_FILENAME(writer);
+    sf_file_writer_get_binlog_filename(writer);
-    if (access(writer->file.name, F_OK) == 0) {
+    if (access(writer->file.name.str, F_OK) == 0) {
         char bak_filename[PATH_MAX];
         char date_str[32];
 
-        snprintf(bak_filename, sizeof(bak_filename), "%s.%s",
+        formatDatetime(g_current_time, "%Y%m%d%H%M%S",
-                writer->file.name, formatDatetime(g_current_time,
+                date_str, sizeof(date_str));
-                    "%Y%m%d%H%M%S", date_str, sizeof(date_str)));
+        fc_combine_two_strings(writer->file.name.str,
-        if (rename(writer->file.name, bak_filename) == 0) {
+                date_str, '.', bak_filename);
+        if (rename(writer->file.name.str, bak_filename) == 0) {
             logWarning("file: "__FILE__", line: %d, "
                     "binlog file %s exist, rename to %s",
-                    __LINE__, writer->file.name, bak_filename);
+                    __LINE__, writer->file.name.str, bak_filename);
         } else {
             logError("file: "__FILE__", line: %d, "
                     "rename binlog %s to backup %s fail, "
                     "errno: %d, error info: %s",
-                    __LINE__, writer->file.name, bak_filename,
+                    __LINE__, writer->file.name.str, bak_filename,
                     errno, STRERROR(errno));
             return errno != 0 ? errno : EPERM;
         }
@@ -172,35 +295,43 @@ static int do_write_to_file(SFFileWriterInfo *writer,
         logError("file: "__FILE__", line: %d, "
                 "write to binlog file \"%s\" fail, "
                 "errno: %d, error info: %s",
-                __LINE__, writer->file.name,
+                __LINE__, writer->file.name.str,
                 result, STRERROR(result));
         return result;
     }
 
+    if (writer->cfg.call_fsync) {
         if (fsync(writer->file.fd) != 0) {
             result = errno != 0 ? errno : EIO;
             logError("file: "__FILE__", line: %d, "
-                "fsync to binlog file \"%s\" fail, "
+                    "fsync to binlog file \"%s\" fail, errno: %d, "
-                "errno: %d, error info: %s",
+                    "error info: %s", __LINE__, writer->file.name.str,
-                __LINE__, writer->file.name,
                     result, STRERROR(result));
             return result;
         }
+    }
 
     writer->file.size += len;
+    if (writer->write_done_callback.func != NULL) {
+        writer->write_done_callback.func(writer,
+                writer->write_done_callback.args);
+    }
+
     return 0;
 }
 
-static int check_write_to_file(SFFileWriterInfo *writer,
+int sf_file_writer_direct_write(SFFileWriterInfo *writer,
         char *buff, const int len)
 {
     int result;
 
-    if (writer->file.size + len <= SF_BINLOG_FILE_MAX_SIZE) {
+    if ((writer->cfg.file_rotate_size <= 0) || (writer->file.size
+                + len <= writer->cfg.file_rotate_size))
+    {
         return do_write_to_file(writer, buff, len);
     }
 
-    writer->binlog.index++;  //binlog rotate
+    writer->binlog.last_index++;  //binlog rotate
     if ((result=write_to_binlog_index_file(writer)) == 0) {
         result = open_next_binlog(writer);
     }
@@ -208,7 +339,7 @@ static int check_write_to_file(SFFileWriterInfo *writer,
     if (result != 0) {
         logError("file: "__FILE__", line: %d, "
                 "open binlog file \"%s\" fail",
-                __LINE__, writer->file.name);
+                __LINE__, writer->file.name.str);
         return result;
     }
 
@@ -220,58 +351,96 @@ int sf_file_writer_flush(SFFileWriterInfo *writer)
     int result;
     int len;
 
-    len = SF_BINLOG_BUFFER_LENGTH(writer->binlog_buffer);
+    len = SF_BINLOG_BUFFER_PRODUCER_DATA_LENGTH(writer->binlog_buffer);
     if (len == 0) {
         return 0;
     }
 
-    result = check_write_to_file(writer, writer->binlog_buffer.buff, len);
+    if ((result=sf_file_writer_direct_write(writer, writer->
-    writer->binlog_buffer.end = writer->binlog_buffer.buff;
+                    binlog_buffer.buff, len)) == 0)
+    {
+        if (writer->flags & SF_FILE_WRITER_FLAGS_WANT_DONE_VERSION) {
+            writer->last_versions.done = writer->last_versions.pending;
+        }
+    }
+
+    writer->binlog_buffer.data_end = writer->binlog_buffer.buff;
     return result;
 }
 
-int sf_file_writer_get_current_index(SFFileWriterInfo *writer)
+int sf_file_writer_fsync(SFFileWriterInfo *writer)
 {
+    int result;
+
+    if ((result=sf_file_writer_flush(writer)) != 0) {
+        return result;
+    }
+
+    if (fsync(writer->file.fd) == 0) {
+        return 0;
+    } else {
+        result = errno != 0 ? errno : EIO;
+        logError("file: "__FILE__", line: %d, "
+                "fsync to binlog file \"%s\" fail, errno: %d, "
+                "error info: %s", __LINE__, writer->file.name.str,
+                result, STRERROR(result));
+        return result;
+    }
+}
+
+int sf_file_writer_get_indexes(SFFileWriterInfo *writer,
+        int *start_index, int *last_index)
+{
+    int result;
+
     if (writer == NULL) {   //for data recovery
+        *start_index = *last_index = 0;
         return 0;
     }
 
-    if (writer->binlog.index < 0) {
+    if (writer->binlog.last_index < 0) {
-        get_binlog_index_from_file(writer);
+        if ((result=get_binlog_index_from_file(writer)) != 0) {
+            *start_index = *last_index = -1;
+            return result;
+        }
     }
 
-    return writer->binlog.index;
+    *start_index = writer->binlog.start_index;
+    *last_index = writer->binlog.last_index;
+    return 0;
 }
 
-int sf_file_writer_deal_buffer(SFFileWriterInfo *writer,
+int sf_file_writer_deal_versioned_buffer(SFFileWriterInfo *writer,
         BufferInfo *buffer, const int64_t version)
 {
     int result;
 
     if (buffer->length >= writer->binlog_buffer.size / 4) {
-        if (SF_BINLOG_BUFFER_LENGTH(writer->binlog_buffer) > 0) {
+        if (SF_BINLOG_BUFFER_PRODUCER_DATA_LENGTH(writer->binlog_buffer) > 0) {
             if ((result=sf_file_writer_flush(writer)) != 0) {
                 return result;
             }
         }
 
-        if ((result=check_write_to_file(writer, buffer->buff,
+        if ((result=sf_file_writer_direct_write(writer, buffer->
-                        buffer->length)) == 0)
+                        buff, buffer->length)) == 0)
         {
             if (writer->flags & SF_FILE_WRITER_FLAGS_WANT_DONE_VERSION) {
                 writer->last_versions.pending = version;
+                writer->last_versions.done = version;
             }
         }
         return result;
     }
 
-    if (writer->file.size + SF_BINLOG_BUFFER_LENGTH(writer->
+    if (writer->cfg.file_rotate_size > 0 && writer->file.size +
-                binlog_buffer) + buffer->length > SF_BINLOG_FILE_MAX_SIZE)
+            SF_BINLOG_BUFFER_PRODUCER_DATA_LENGTH(writer->binlog_buffer) +
+            buffer->length > writer->cfg.file_rotate_size)
     {
         if ((result=sf_file_writer_flush(writer)) != 0) {
             return result;
         }
-    } else if (writer->binlog_buffer.size - SF_BINLOG_BUFFER_LENGTH(
+    } else if (SF_BINLOG_BUFFER_PRODUCER_BUFF_REMAIN(
                 writer->binlog_buffer) < buffer->length)
     {
         if ((result=sf_file_writer_flush(writer)) != 0) {
@@ -282,15 +451,40 @@ int sf_file_writer_deal_buffer(SFFileWriterInfo *writer,
     if (writer->flags & SF_FILE_WRITER_FLAGS_WANT_DONE_VERSION) {
         writer->last_versions.pending = version;
     }
-    memcpy(writer->binlog_buffer.end, buffer->buff, buffer->length);
+    memcpy(writer->binlog_buffer.data_end, buffer->buff, buffer->length);
-    writer->binlog_buffer.end += buffer->length;
+    writer->binlog_buffer.data_end += buffer->length;
 
     return 0;
 }
 
-int sf_file_writer_init_normal(SFFileWriterInfo *writer,
+int sf_file_writer_save_buffer_ex(SFFileWriterInfo *writer,
-        const char *data_path, const char *subdir_name,
+        const int length, const bool flush)
-        const int buffer_size)
+{
+    int result;
+
+    if (writer->cfg.file_rotate_size > 0 && writer->file.size +
+            SF_BINLOG_BUFFER_PRODUCER_DATA_LENGTH(writer->binlog_buffer) +
+            length > writer->cfg.file_rotate_size)
+    {
+        if ((result=sf_file_writer_flush(writer)) != 0) {
+            return result;
+        }
+    }
+
+    writer->binlog_buffer.data_end += length;
+    if (flush || SF_BINLOG_BUFFER_PRODUCER_BUFF_REMAIN(writer->
+                binlog_buffer) < writer->cfg.max_record_size)
+    {
+        return sf_file_writer_flush(writer);
+    } else {
+        return 0;
+    }
+}
+
+int sf_file_writer_init(SFFileWriterInfo *writer, const char *data_path,
+        const char *subdir_name, const char *file_prefix,
+        const int max_record_size, const int buffer_size,
+        const int64_t file_rotate_size, const bool call_fsync)
 {
     int result;
     int path_len;
@@ -301,15 +495,18 @@ int sf_file_writer_init_normal(SFFileWriterInfo *writer,
     writer->last_versions.pending = 0;
     writer->last_versions.done = 0;
     writer->flags = 0;
+    sf_file_writer_set_write_done_callback(writer, NULL, NULL);
     if ((result=sf_binlog_buffer_init(&writer->
                     binlog_buffer, buffer_size)) != 0)
     {
         return result;
     }
 
+    writer->cfg.max_record_size = max_record_size;
+    writer->cfg.call_fsync = call_fsync;
+    writer->cfg.file_rotate_size = file_rotate_size;
     writer->cfg.data_path = data_path;
-    path_len = snprintf(filepath, sizeof(filepath),
+    path_len = fc_combine_full_filepath(data_path, subdir_name, filepath);
-            "%s/%s", data_path, subdir_name);
     if ((result=fc_check_mkdir_ex(filepath, 0775, &create)) != 0) {
         return result;
     }
@@ -318,11 +515,11 @@ int sf_file_writer_init_normal(SFFileWriterInfo *writer,
     }
 
     writer->file.fd = -1;
-    snprintf(writer->cfg.subdir_name,
+    fc_safe_strcpy(writer->cfg.subdir_name, subdir_name);
-            sizeof(writer->cfg.subdir_name),
+    fc_safe_strcpy(writer->cfg.file_prefix, file_prefix);
-            "%s", subdir_name);
+    writer->file.name.size = path_len + 32;
-    writer->file.name = (char *)fc_malloc(path_len + 32);
+    writer->file.name.str = (char *)fc_malloc(writer->file.name.size);
-    if (writer->file.name == NULL) {
+    if (writer->file.name.str == NULL) {
         return ENOMEM;
     }
 
@@ -337,13 +534,59 @@ int sf_file_writer_init_normal(SFFileWriterInfo *writer,
     return 0;
 }
 
-int sf_file_writer_set_binlog_index(SFFileWriterInfo *writer,
+void sf_file_writer_destroy(SFFileWriterInfo *writer)
-        const int binlog_index)
+{
+    if (writer->file.fd >= 0) {
+        close(writer->file.fd);
+        writer->file.fd = -1;
+    }
+    if (writer->file.name.str != NULL) {
+        free(writer->file.name.str);
+        writer->file.name.str = NULL;
+    }
+    sf_binlog_buffer_destroy(&writer->binlog_buffer);
+}
+
+int sf_file_writer_set_indexes(SFFileWriterInfo *writer,
+        const int start_index, const int last_index)
 {
     int result;
 
-    if (writer->binlog.index != binlog_index) {
+    if (writer->binlog.start_index != start_index ||
-        writer->binlog.index = binlog_index;
+            writer->binlog.last_index != last_index)
+    {
+        writer->binlog.start_index = start_index;
+        writer->binlog.last_index = last_index;
+        if ((result=write_to_binlog_index_file(writer)) != 0) {
+            return result;
+        }
+    }
+
+    return 0;
+}
+
+int sf_file_writer_set_binlog_start_index(SFFileWriterInfo *writer,
+        const int start_index)
+{
+    int result;
+
+    if (writer->binlog.start_index != start_index) {
+        writer->binlog.start_index = start_index;
+        if ((result=write_to_binlog_index_file(writer)) != 0) {
+            return result;
+        }
+    }
+
+    return 0;
+}
+
+int sf_file_writer_set_binlog_write_index(SFFileWriterInfo *writer,
+        const int last_index)
+{
+    int result;
+
+    if (writer->binlog.last_index != last_index) {
+        writer->binlog.last_index = last_index;
         if ((result=write_to_binlog_index_file(writer)) != 0) {
             return result;
         }
@@ -357,40 +600,108 @@ int sf_file_writer_get_last_lines(const char *data_path,
         char *buff, const int buff_size, int *count, int *length)
 {
     int result;
-    int remain_count;
+    int target_count;
-    int current_count;
+    int count1;
-    int current_index;
-    int i;
     char filename[PATH_MAX];
     string_t lines;
 
-    current_index = current_write_index;
+    target_count = *count;
-    *length = 0;
-    remain_count = *count;
-    for (i=0; i<2; i++) {
-        current_count = remain_count;
     sf_file_writer_get_filename(data_path, subdir_name,
-                current_index, filename, sizeof(filename));
+            current_write_index, filename, sizeof(filename));
-        result = fc_get_last_lines(filename, buff + *length,
+    if (access(filename, F_OK) == 0) {
-                buff_size - *length, &lines, &current_count);
+        if ((result=fc_get_last_lines(filename, buff, buff_size,
-        if (result == 0) {
+                        &lines, count)) != 0)
-            memmove(buff + *length, lines.str, lines.len);
+        {
-            *length += lines.len;
+            if (result != ENOENT) {
-            remain_count -= current_count;
-            if (remain_count == 0) {
-                break;
-            }
-        } else if (result != ENOENT) {
-            *count = 0;
                 return result;
             }
-        if (current_index == 0) {
-            break;
         }
 
-        --current_index;  //try previous binlog file
+        if (*count >= target_count || current_write_index == 0) {
+            memmove(buff, lines.str, lines.len);
+            *length = lines.len;
+            return 0;
+        }
+    } else {
+        result = errno != 0 ? errno : EPERM;
+        if (result == ENOENT) {
+            *count = 0;
+            *length = 0;
+            return 0;
+        } else {
+            logError("file: "__FILE__", line: %d, "
+                    "stat file %s fail, errno: %d, error info: %s",
+                    __LINE__, filename, result, STRERROR(result));
+            *count = 0;
+            *length = 0;
+            return result;
+        }
+    }
+
+    sf_file_writer_get_filename(data_path, subdir_name,
+            current_write_index - 1, filename, sizeof(filename));
+    if (access(filename, F_OK) != 0) {
+        result = errno != 0 ? errno : EPERM;
+        if (result == ENOENT) {
+            memmove(buff, lines.str, lines.len);
+            *length = lines.len;
+            return 0;
+        } else {
+            logError("file: "__FILE__", line: %d, "
+                    "stat file %s fail, errno: %d, error info: %s",
+                    __LINE__, filename, result, STRERROR(result));
+            *count = 0;
+            *length = 0;
+            return result;
+        }
+    }
+
+    count1 = target_count - *count;
+    if ((result=fc_get_last_lines(filename, buff,
+                    buff_size, &lines, &count1)) != 0)
+    {
+        *count = 0;
+        *length = 0;
+        return result;
+    }
+
+    memmove(buff, lines.str, lines.len);
+    *length = lines.len;
+    if (*count == 0) {
+        *count = count1;
+    } else {
+        sf_file_writer_get_filename(data_path, subdir_name,
+                current_write_index, filename, sizeof(filename));
+        if ((result=fc_get_first_lines(filename, buff + (*length),
+                        buff_size - (*length), &lines, count)) != 0)
+        {
+            *count = 0;
+            *length = 0;
+            return result;
+        }
+
+        *count += count1;
+        *length += lines.len;
     }
 
-    *count -= remain_count;
     return 0;
 }
+
+int sf_file_writer_get_last_line(const char *data_path,
+        const char *subdir_name, char *buff,
+        const int buff_size, int *length)
+{
+    int result;
+    int last_index;
+    int count = 1;
+
+    if ((result=sf_file_writer_get_binlog_last_index(data_path,
+                    subdir_name, &last_index)) != 0)
+    {
+        *length = 0;
+        return result;
+    }
+
+    return sf_file_writer_get_last_lines(data_path, subdir_name,
+            last_index, buff, buff_size, &count, length);
+}

src/sf_file_writer.h

@@ -24,29 +24,42 @@
 #define SF_FILE_WRITER_FLAGS_WANT_DONE_VERSION  1
 
 #define SF_BINLOG_SUBDIR_NAME_SIZE     128
-#define SF_BINLOG_FILE_MAX_SIZE   (1024 * 1024 * 1024)  //for binlog rotating by size
+#define SF_BINLOG_FILE_PREFIX_STR_SIZE 64
-#define SF_BINLOG_FILE_PREFIX     "binlog"
+#define SF_BINLOG_DEFAULT_ROTATE_SIZE  (1024 * 1024 * 1024)
-#define SF_BINLOG_FILE_EXT_FMT    ".%06d"
+#define SF_BINLOG_NEVER_ROTATE_FILE    0
+#define SF_BINLOG_FILE_PREFIX_STR      "binlog"
+#define SF_BINLOG_FILE_PREFIX_LEN      (sizeof(SF_BINLOG_FILE_PREFIX_STR) - 1)
+#define SF_BINLOG_FILE_EXT_LEN 6
+#define SF_BINLOG_FILE_EXT_FMT ".%0"FC_MACRO_TOSTRING(SF_BINLOG_FILE_EXT_LEN)"d"
 
-#define SF_BINLOG_BUFFER_LENGTH(buffer) ((buffer).end - (buffer).buff)
+struct sf_file_writer_info;
-#define SF_BINLOG_BUFFER_REMAIN(buffer) ((buffer).end - (buffer).current)
+
+typedef void (*sf_file_write_done_callback)(
+        struct sf_file_writer_info *writer, void *args);
 
 typedef struct sf_file_writer_info {
     struct {
         const char *data_path;
         char subdir_name[SF_BINLOG_SUBDIR_NAME_SIZE];
+        char file_prefix[SF_BINLOG_FILE_PREFIX_STR_SIZE];
+        int64_t file_rotate_size;
         int max_record_size;
+        bool call_fsync;
     } cfg;
 
     struct {
-        int index;
+        int start_index;  //for read only
+        int last_index;   //for write
         int compress_index;
     } binlog;
 
     struct {
         int fd;
-        int64_t size;
+        int64_t size;  //file size
-        char *name;
+        struct {
+            char *str;
+            int size;
+        } name;
     } file;
 
     int64_t total_count;
@@ -57,73 +70,252 @@ typedef struct sf_file_writer_info {
         int64_t pending;
         volatile int64_t done;
     } last_versions;
+
+    struct {
+        sf_file_write_done_callback func;
+        void *args;
+    } write_done_callback;
+
 } SFFileWriterInfo;
 
 #ifdef __cplusplus
 extern "C" {
 #endif
 
-int sf_file_writer_init_normal(SFFileWriterInfo *writer,
+int sf_file_writer_init(SFFileWriterInfo *writer, const char *data_path,
-        const char *data_path, const char *subdir_name,
+        const char *subdir_name, const char *file_prefix,
-        const int buffer_size);
+        const int max_record_size, const int buffer_size,
+        const int64_t file_rotate_size, const bool call_fsync);
 
-int sf_file_writer_deal_buffer(SFFileWriterInfo *writer,
+void sf_file_writer_destroy(SFFileWriterInfo *writer);
+
+int sf_file_writer_direct_write(SFFileWriterInfo *writer,
+        char *buff, const int len);
+
+int sf_file_writer_deal_versioned_buffer(SFFileWriterInfo *writer,
         BufferInfo *buffer, const int64_t version);
 
+#define sf_file_writer_deal_buffer(writer, buffer) \
+    sf_file_writer_deal_versioned_buffer(writer, buffer, 0)
+
 int sf_file_writer_flush(SFFileWriterInfo *writer);
 
+int sf_file_writer_fsync(SFFileWriterInfo *writer);
+
+#define SF_FILE_WRITER_DATA_END_BUFF(writer) (writer)->binlog_buffer.data_end
+#define SF_FILE_WRITER_CURRENT_DATA_VERSION(writer) \
+    (writer)->last_versions.pending
+#define SF_FILE_WRITER_NEXT_DATA_VERSION(writer) \
+    ++((writer)->last_versions.pending)
+
+int sf_file_writer_save_buffer_ex(SFFileWriterInfo *writer,
+        const int length, const bool flush);
+
+static inline int sf_file_writer_save_buffer(
+        SFFileWriterInfo *writer, const int length)
+{
+    const bool flush = false;
+    return sf_file_writer_save_buffer_ex(writer, length, flush);
+}
+
+static inline int sf_file_writer_flush_buffer(
+        SFFileWriterInfo *writer, const int length)
+{
+    const bool flush = true;
+    return sf_file_writer_save_buffer_ex(writer, length, flush);
+}
+
 static inline void sf_file_writer_set_flags(
         SFFileWriterInfo *writer, const short flags)
 {
     writer->flags = flags;
 }
 
-static inline int64_t sf_file_writer_get_last_version(
+static inline void sf_file_writer_set_call_fsync(
-        SFFileWriterInfo *writer)
+        SFFileWriterInfo *writer, const bool call_fsync)
+{
+    writer->cfg.call_fsync = call_fsync;
+}
+
+static inline void sf_file_writer_set_write_done_callback (
+        SFFileWriterInfo *writer, sf_file_write_done_callback callback,
+        void *args)
+{
+    writer->write_done_callback.func = callback;
+    writer->write_done_callback.args = args;
+}
+
+static inline int64_t sf_file_writer_get_last_version_ex(
+        SFFileWriterInfo *writer, const int log_level)
 {
     if (writer->flags & SF_FILE_WRITER_FLAGS_WANT_DONE_VERSION) {
         return writer->last_versions.done;
     } else {
-        logError("file: "__FILE__", line: %d, "
+        if (FC_LOG_BY_LEVEL(log_level)) {
-                "should set writer flags to %d!", __LINE__,
+            log_it_ex(&g_log_context, log_level, "file: %s, line: %d, "
+                    "writer: %s, should set writer flags to %d!",
+                    __FILE__, __LINE__, writer->cfg.subdir_name,
                     SF_FILE_WRITER_FLAGS_WANT_DONE_VERSION);
+        }
+
         return -1;
     }
 }
 
-int sf_file_writer_get_current_index(SFFileWriterInfo *writer);
+#define sf_file_writer_get_last_version(writer) \
+    sf_file_writer_get_last_version_ex(writer, LOG_ERR)
+
+#define sf_file_writer_get_last_version_silence(writer) \
+    sf_file_writer_get_last_version_ex(writer, LOG_NOTHING)
+
+int sf_file_writer_get_binlog_indexes(const char *data_path,
+        const char *subdir_name, int *start_index, int *last_index);
+
+static inline int sf_file_writer_get_binlog_start_index(
+        const char *data_path, const char *subdir_name,
+        int *start_index)
+{
+    int last_index;
+    return sf_file_writer_get_binlog_indexes(data_path,
+            subdir_name, start_index, &last_index);
+}
+
+static inline int sf_file_writer_get_binlog_last_index(
+        const char *data_path, const char *subdir_name,
+        int *last_index)
+{
+    int start_index;
+    return sf_file_writer_get_binlog_indexes(data_path,
+            subdir_name, &start_index, last_index);
+}
+
+int sf_file_writer_set_indexes(SFFileWriterInfo *writer,
+        const int start_index, const int last_index);
+
+int sf_file_writer_get_indexes(SFFileWriterInfo *writer,
+        int *start_index, int *last_index);
+
+static inline int sf_file_writer_get_start_index(SFFileWriterInfo *writer)
+{
+    int start_index;
+    int last_index;
+
+    sf_file_writer_get_indexes(writer, &start_index, &last_index);
+    return start_index;
+}
+
+static inline int sf_file_writer_get_last_index(SFFileWriterInfo *writer)
+{
+    int start_index;
+    int last_index;
+
+    sf_file_writer_get_indexes(writer, &start_index, &last_index);
+    return last_index;
+}
+
+#define sf_file_writer_get_current_write_index(writer) \
+    sf_file_writer_get_last_index(writer)
 
 static inline void sf_file_writer_get_current_position(
         SFFileWriterInfo *writer, SFBinlogFilePosition *position)
 {
-    position->index = writer->binlog.index;
+    position->index = writer->binlog.last_index;
     position->offset = writer->file.size;
 }
 
 static inline const char *sf_file_writer_get_filepath(
         const char *data_path, const char *subdir_name,
+        char *filepath, const int size)
+{
+    fc_get_full_filepath_ex(data_path, strlen(data_path),
+            subdir_name, strlen(subdir_name), filepath, size);
+    return filepath;
+}
+
+static inline const char *sf_file_writer_get_filename_ex(
+        const char *data_path, const char *subdir_name,
+        const char *file_prefix, const int binlog_index,
         char *filename, const int size)
 {
-    snprintf(filename, size, "%s/%s", data_path, subdir_name);
+    char *p;
+    int data_path_len;
+    int subdir_name_len;
+    int file_prefix_len;
+
+    data_path_len = strlen(data_path);
+    subdir_name_len = strlen(subdir_name);
+    file_prefix_len = strlen(file_prefix);
+    if (data_path_len + subdir_name_len + file_prefix_len +
+            4 + SF_BINLOG_FILE_EXT_LEN >= size)
+    {
+        snprintf(filename, size, "%s/%s/%s"SF_BINLOG_FILE_EXT_FMT,
+                data_path, subdir_name, file_prefix, binlog_index);
+        return filename;
+    }
+
+    p = filename;
+    memcpy(p, data_path, data_path_len);
+    p += data_path_len;
+    *p++ = '/';
+    memcpy(p, subdir_name, subdir_name_len);
+    p += subdir_name_len;
+    *p++ = '/';
+    memcpy(p, file_prefix, file_prefix_len);
+    p += file_prefix_len;
+    *p++ = '.';
+    fc_ltostr_ex(binlog_index, p, SF_BINLOG_FILE_EXT_LEN);
+
     return filename;
 }
 
-static inline const char *sf_file_writer_get_filename(
+#define sf_file_writer_get_filename(data_path, subdir_name, \
-        const char *data_path, const char *subdir_name,
+        binlog_index, filename, size) \
-        const int binlog_index, char *filename, const int size)
+    sf_file_writer_get_filename_ex(data_path, subdir_name, \
+        SF_BINLOG_FILE_PREFIX_STR, binlog_index, filename, size)
+
+const char *sf_file_writer_get_index_filename(const char *data_path,
+        const char *subdir_name, char *filename, const int size);
+
+int sf_file_writer_set_binlog_start_index(SFFileWriterInfo *writer,
+        const int start_index);
+
+int sf_file_writer_set_binlog_write_index(SFFileWriterInfo *writer,
+        const int last_index);
+
+static inline int sf_file_writer_rotate_file(SFFileWriterInfo *writer)
 {
-    snprintf(filename, size, "%s/%s/%s"SF_BINLOG_FILE_EXT_FMT, data_path,
+    int last_index;
-            subdir_name, SF_BINLOG_FILE_PREFIX, binlog_index);
+    last_index = sf_file_writer_get_current_write_index(writer);
-    return filename;
+    return sf_file_writer_set_binlog_write_index(writer, last_index + 1);
 }
 
-int sf_file_writer_set_binlog_index(SFFileWriterInfo *writer,
-        const int binlog_index);
-
 int sf_file_writer_get_last_lines(const char *data_path,
         const char *subdir_name, const int current_write_index,
         char *buff, const int buff_size, int *count, int *length);
 
+static inline int sf_file_writer_get_last_line_ex(const char *data_path,
+        const char *subdir_name, const int current_write_index,
+        char *buff, const int buff_size, int *length)
+{
+    int count = 1;
+    return sf_file_writer_get_last_lines(data_path, subdir_name,
+            current_write_index, buff, buff_size, &count, length);
+}
+
+int sf_file_writer_get_last_line(const char *data_path,
+        const char *subdir_name, char *buff,
+        const int buff_size, int *length);
+
+int sf_file_writer_write_to_binlog_index_file_ex(const char *data_path,
+        const char *subdir_name, const char *file_prefix,
+        const int start_index, const int last_index,
+        const int compress_index);
+
+#define sf_file_writer_write_to_binlog_index_file(data_path,  \
+        subdir_name, start_index, last_index) \
+    sf_file_writer_write_to_binlog_index_file_ex(data_path, subdir_name, \
+            SF_BINLOG_FILE_PREFIX_STR, start_index, last_index, 0)
+
 #ifdef __cplusplus
 }
 #endif

src/sf_func.c

@@ -38,16 +38,17 @@ int sf_connect_to_server(const char *ip_addr, const int port, int *sock)
     if(*sock < 0) {
         return errno != 0 ? errno : ENOMEM;
     }
-    tcpsetserveropt(*sock, g_sf_global_vars.network_timeout);
+    tcpsetserveropt(*sock, g_sf_global_vars.net_buffer_cfg.network_timeout);
 
     if ((result=tcpsetnonblockopt(*sock)) != 0) {
         close(*sock);
         *sock = -1;
         return result;
     }
+    FC_SET_CLOEXEC(*sock);
 
-    if ((result=connectserverbyip_nb(*sock, ip_addr, port,
+    if ((result=connectserverbyip_nb(*sock, ip_addr, port, g_sf_global_vars.
-                    g_sf_global_vars.connect_timeout)) != 0)
+                    net_buffer_cfg.connect_timeout)) != 0)
     {
         close(*sock);
         *sock = -1;

src/sf_func.h

@@ -37,7 +37,7 @@ static inline void sf_terminate_myself_ex(const char *file,
 {
     g_sf_global_vars.continue_flag = false;
     if (kill(getpid(), SIGQUIT) == 0) { //signal myself to quit
-        logInfo("file: "__FILE__", line: %d, "
+        logWarning("file: "__FILE__", line: %d, "
                 "kill myself from caller {file: %s, line: %d, func: %s}",
                 __LINE__, file, line, func);
     } else {
@@ -56,7 +56,8 @@ static inline int sf_binlog_buffer_init(SFBinlogBuffer *buffer, const int size)
         return ENOMEM;
     }
 
-    buffer->current = buffer->end = buffer->buff;
+    buffer->current = buffer->data_end = buffer->buff;
+    buffer->buff_end = buffer->buff + size;
     buffer->size = size;
     return 0;
 }
@@ -65,7 +66,8 @@ static inline void sf_binlog_buffer_destroy(SFBinlogBuffer *buffer)
 {
     if (buffer->buff != NULL) {
         free(buffer->buff);
-        buffer->current = buffer->end = buffer->buff = NULL;
+        buffer->current = buffer->buff = NULL;
+        buffer->data_end = buffer->buff_end = NULL;
         buffer->size = 0;
     }
 }
@@ -76,6 +78,11 @@ static inline int sf_synchronize_ctx_init(SFSynchronizeContext *sctx)
     return init_pthread_lock_cond_pair(&sctx->lcp);
 }
 
+static inline void sf_synchronize_ctx_destroy(SFSynchronizeContext *sctx)
+{
+    destroy_pthread_lock_cond_pair(&sctx->lcp);
+}
+
 static inline void sf_synchronize_counter_add(
         SFSynchronizeContext *sctx, const int count)
 {
@@ -106,12 +113,40 @@ static inline void sf_synchronize_counter_notify(
 static inline void sf_synchronize_counter_wait(SFSynchronizeContext *sctx)
 {
     PTHREAD_MUTEX_LOCK(&sctx->lcp.lock);
-    while (sctx->waiting_count != 0) {
+    while (sctx->waiting_count != 0 && SF_G_CONTINUE_FLAG) {
         pthread_cond_wait(&sctx->lcp.cond, &sctx->lcp.lock);
     }
     PTHREAD_MUTEX_UNLOCK(&sctx->lcp.lock);
 }
 
+#define sf_synchronize_finished_notify_no_lock(sctx, err_no) \
+    (sctx)->finished = true;  \
+    (sctx)->result = err_no;  \
+    pthread_cond_signal(&(sctx)->lcp.cond)
+
+static inline void sf_synchronize_finished_notify(
+        SFSynchronizeContext *sctx, const int result)
+{
+    PTHREAD_MUTEX_LOCK(&sctx->lcp.lock);
+    sf_synchronize_finished_notify_no_lock(sctx, result);
+    PTHREAD_MUTEX_UNLOCK(&sctx->lcp.lock);
+}
+
+static inline int sf_synchronize_finished_wait(SFSynchronizeContext *sctx)
+{
+    int result;
+
+    PTHREAD_MUTEX_LOCK(&sctx->lcp.lock);
+    while (!sctx->finished && SF_G_CONTINUE_FLAG) {
+        pthread_cond_wait(&sctx->lcp.cond, &sctx->lcp.lock);
+    }
+    result = sctx->result;
+    sctx->finished = false;  //for next notify
+    PTHREAD_MUTEX_UNLOCK(&sctx->lcp.lock);
+
+    return result;
+}
+
 #ifdef __cplusplus
 }
 #endif

src/sf_global.h

@@ -30,32 +30,36 @@ typedef struct sf_connection_stat {
 } SFConnectionStat;
 
 typedef struct sf_global_variables {
-    int connect_timeout;
-    int network_timeout;
     struct {
         char str[MAX_PATH_SIZE];
+        int len;
         bool inited;
+        bool created;
     } base_path;
 
     volatile bool continue_flag;
     bool tcp_quick_ack;
-    int max_connections;
+    bool epoll_edge_trigger;
-    int max_pkg_size;
+
-    int min_buff_size;
+    SFNetBufferConfig net_buffer_cfg;
-    int max_buff_size;
+
     int task_buffer_extra_size;
     int thread_stack_size;
 
     time_t up_time;
-    gid_t run_by_gid;
+    struct {
-    uid_t run_by_uid;
+        bool inited;
-    char run_by_group[32];
+        gid_t gid;
-    char run_by_user[32];
+        uid_t uid;
+        char group[32];
+        char user[32];
+    } run_by;
 
     SFLogConfig error_log;
     SFConnectionStat connection_stat;
     sf_error_handler_callback error_handler;
     string_t empty;
+    volatile time_t last_binlog_writer_log_timestamp;
 } SFGlobalVariables;
 
 typedef struct sf_context_ini_config {
@@ -63,6 +67,9 @@ typedef struct sf_context_ini_config {
     int default_inner_port;
     int default_outer_port;
     int default_work_threads;
+    int max_pkg_size_min_value;
+    FCCommunicationType comm_type;
+    const char *max_pkg_size_item_name;
 } SFContextIniConfig;
 
 #ifdef __cplusplus
@@ -73,13 +80,36 @@ extern SFGlobalVariables         g_sf_global_vars;
 extern SFContext                 g_sf_context;
 
 #define SF_G_BASE_PATH_STR       g_sf_global_vars.base_path.str
+#define SF_G_BASE_PATH_LEN       g_sf_global_vars.base_path.len
 #define SF_G_BASE_PATH_INITED    g_sf_global_vars.base_path.inited
+#define SF_G_BASE_PATH_CREATED   g_sf_global_vars.base_path.created
 #define SF_G_CONTINUE_FLAG       g_sf_global_vars.continue_flag
-#define SF_G_CONNECT_TIMEOUT     g_sf_global_vars.connect_timeout
+#define SF_G_CONNECT_TIMEOUT     g_sf_global_vars.net_buffer_cfg.connect_timeout
-#define SF_G_NETWORK_TIMEOUT     g_sf_global_vars.network_timeout
+#define SF_G_NETWORK_TIMEOUT     g_sf_global_vars.net_buffer_cfg.network_timeout
-#define SF_G_MAX_CONNECTIONS     g_sf_global_vars.max_connections
+#define SF_G_MAX_CONNECTIONS     g_sf_global_vars.net_buffer_cfg.max_connections
 #define SF_G_THREAD_STACK_SIZE   g_sf_global_vars.thread_stack_size
+#define SF_G_UP_TIME             g_sf_global_vars.up_time
 
+#define SF_G_SOCK_HANDLER        (g_sf_context.handlers  \
+        [SF_IPV4_ADDRESS_FAMILY_INDEX].handlers + \
+        SF_SOCKET_NETWORK_HANDLER_INDEX)
+#define SF_G_OUTER_PORT          SF_G_SOCK_HANDLER->outer.port
+#define SF_G_INNER_PORT          SF_G_SOCK_HANDLER->inner.port
+#define SF_G_OUTER_BIND_ADDR4     g_sf_context.handlers  \
+        [SF_IPV4_ADDRESS_FAMILY_INDEX].outer_bind_addr
+#define SF_G_INNER_BIND_ADDR4     g_sf_context.handlers  \
+        [SF_IPV4_ADDRESS_FAMILY_INDEX].inner_bind_addr
+#define SF_G_OUTER_BIND_ADDR6     g_sf_context.handlers  \
+        [SF_IPV6_ADDRESS_FAMILY_INDEX].outer_bind_addr
+#define SF_G_INNER_BIND_ADDR6     g_sf_context.handlers  \
+        [SF_IPV6_ADDRESS_FAMILY_INDEX].inner_bind_addr
+
+#define SF_G_IPV4_ENABLED     (g_sf_context.handlers  \
+        [SF_IPV4_ADDRESS_FAMILY_INDEX].af == AF_INET)
+#define SF_G_IPV6_ENABLED     (g_sf_context.handlers  \
+        [SF_IPV6_ADDRESS_FAMILY_INDEX].af == AF_INET6)
+
+#define SF_G_ACCEPT_THREADS      g_sf_context.accept_threads
 #define SF_G_WORK_THREADS        g_sf_context.work_threads
 #define SF_G_ALIVE_THREAD_COUNT  g_sf_context.thread_count
 #define SF_G_THREAD_INDEX(tdata) (int)(tdata - g_sf_context.thread_data)
@@ -88,18 +118,28 @@ extern SFContext                 g_sf_context;
 
 #define SF_G_ERROR_HANDLER       g_sf_global_vars.error_handler
 #define SF_G_EMPTY_STRING        g_sf_global_vars.empty
+#define LAST_BINLOG_WRITER_LOG_TIMESTAMP  g_sf_global_vars. \
+    last_binlog_writer_log_timestamp
 
-#define SF_WORK_THREADS(sf_context)        sf_context.work_threads
+#define SF_G_EPOLL_EDGE_TRIGGER  g_sf_global_vars.epoll_edge_trigger
-#define SF_ALIVE_THREAD_COUNT(sf_context)  sf_context.thread_count
+
-#define SF_THREAD_INDEX(sf_context, tdata) (int)(tdata - sf_context.thread_data)
+#define SF_WORK_THREADS(sf_context)        (sf_context).work_threads
+#define SF_ALIVE_THREAD_COUNT(sf_context)  (sf_context).thread_count
+#define SF_THREAD_INDEX(sf_context, tdata) (int)(tdata - (sf_context).thread_data)
+
+#define SF_IPV4_ENABLED(sf_context)   ((sf_context).handlers  \
+        [SF_IPV4_ADDRESS_FAMILY_INDEX].af == AF_INET)
+#define SF_IPV6_ENABLED(sf_context)   ((sf_context).handlers  \
+        [SF_IPV6_ADDRESS_FAMILY_INDEX].af == AF_INET6)
 
 #define SF_CHOWN_RETURN_ON_ERROR(path, current_uid, current_gid) \
     do { \
-    if (!(g_sf_global_vars.run_by_gid == current_gid && \
+        if (g_sf_global_vars.run_by.inited && !(g_sf_global_vars.  \
-                g_sf_global_vars.run_by_uid == current_uid)) \
+                    run_by.gid == current_gid && g_sf_global_vars. \
+                    run_by.uid == current_uid)) \
         { \
-        if (chown(path, g_sf_global_vars.run_by_uid, \
+            if (chown(path, g_sf_global_vars.run_by.uid, \
-                    g_sf_global_vars.run_by_gid) != 0) \
+                        g_sf_global_vars.run_by.gid) != 0) \
             { \
                 logError("file: "__FILE__", line: %d, " \
                         "chown \"%s\" fail, " \
@@ -113,62 +153,115 @@ extern SFContext                 g_sf_context;
 #define SF_CHOWN_TO_RUNBY_RETURN_ON_ERROR(path) \
     SF_CHOWN_RETURN_ON_ERROR(path, geteuid(), getegid())
 
-#define SF_SET_CONTEXT_INI_CONFIG(config, filename, pIniContext, \
+
-        section_name, def_inner_port, def_outer_port, def_work_threads) \
+#define SF_FCHOWN_RETURN_ON_ERROR(fd, path, current_uid, current_gid) \
+    do { \
+        if (g_sf_global_vars.run_by.inited && !(g_sf_global_vars.  \
+                    run_by.gid == current_gid && g_sf_global_vars. \
+                    run_by.uid == current_uid)) \
+        { \
+            if (fchown(fd, g_sf_global_vars.run_by.uid, \
+                        g_sf_global_vars.run_by.gid) != 0) \
+            { \
+                logError("file: "__FILE__", line: %d, " \
+                        "fchown \"%s\" fail, " \
+                        "errno: %d, error info: %s", \
+                        __LINE__, path, errno, STRERROR(errno)); \
+                return errno != 0 ? errno : EPERM; \
+            } \
+        } \
+    } while (0)
+
+#define SF_FCHOWN_TO_RUNBY_RETURN_ON_ERROR(fd, path) \
+    SF_FCHOWN_RETURN_ON_ERROR(fd, path, geteuid(), getegid())
+
+#define SF_SET_CONTEXT_INI_CONFIG_EX(config, the_comm_type, filename, \
+        pIniContext, section_name, def_inner_port, def_outer_port,    \
+        def_work_threads, max_pkg_size_item_nm, max_pkg_size_min_val) \
     do { \
         FAST_INI_SET_FULL_CTX_EX(config.ini_ctx, filename, \
                 section_name, pIniContext);   \
+        config.comm_type = the_comm_type; \
         config.default_inner_port = def_inner_port; \
         config.default_outer_port = def_outer_port; \
         config.default_work_threads = def_work_threads; \
+        config.max_pkg_size_item_name = max_pkg_size_item_nm; \
+        config.max_pkg_size_min_value = max_pkg_size_min_val; \
     } while (0)
 
-int sf_load_global_config_ex(const char *server_name,
+#define SF_SET_CONTEXT_INI_CONFIG(config, the_comm_type,     \
-        IniFullContext *ini_ctx, const bool load_network_params,
+        filename, pIniContext, section_name, def_inner_port, \
-        const int task_buffer_extra_size);
+        def_outer_port, def_work_threads) \
+     SF_SET_CONTEXT_INI_CONFIG_EX(config, the_comm_type, filename, \
+             pIniContext, section_name, def_inner_port, def_outer_port, \
+             def_work_threads, "max_pkg_size", 0)
 
-static inline int sf_load_global_config(const char *server_name,
+int sf_load_global_config_ex(const char *log_filename_prefix,
+        IniFullContext *ini_ctx, const bool load_network_params,
+        const char *max_pkg_size_item_nm, const int fixed_buff_size,
+        const int task_buffer_extra_size, const bool need_set_run_by);
+
+static inline int sf_load_global_config(const char *log_filename_prefix,
         IniFullContext *ini_ctx)
 {
     const bool load_network_params = true;
+    const char *max_pkg_size_item_nm = "max_pkg_size";
+    const int fixed_buff_size = 0;
     const int task_buffer_extra_size = 0;
+    const bool need_set_run_by = true;
 
-    return sf_load_global_config_ex(server_name, ini_ctx,
+    return sf_load_global_config_ex(log_filename_prefix, ini_ctx,
-            load_network_params, task_buffer_extra_size);
+            load_network_params, max_pkg_size_item_nm, fixed_buff_size,
+            task_buffer_extra_size, need_set_run_by);
 }
 
-int sf_load_config_ex(const char *server_name,
+int sf_load_config_ex(const char *log_filename_prefix,
-        SFContextIniConfig *config, const int task_buffer_extra_size);
+        SFContextIniConfig *config, const int fixed_buff_size,
+        const int task_buffer_extra_size, const bool need_set_run_by);
 
-static inline int sf_load_config(const char *server_name,
+static inline int sf_load_config(const char *log_filename_prefix,
+        const FCCommunicationType comm_type,
         const char *filename, IniContext *pIniContext,
         const char *section_name, const int default_inner_port,
-        const int default_outer_port, const int task_buffer_extra_size)
+        const int default_outer_port, const int fixed_buff_size,
+        const int task_buffer_extra_size)
 {
+    const bool need_set_run_by = true;
     SFContextIniConfig config;
 
-    SF_SET_CONTEXT_INI_CONFIG(config, filename, pIniContext,
+    SF_SET_CONTEXT_INI_CONFIG(config, comm_type, filename, pIniContext,
             section_name, default_inner_port, default_outer_port,
             DEFAULT_WORK_THREADS);
-    return sf_load_config_ex(server_name, &config, task_buffer_extra_size);
+    return sf_load_config_ex(log_filename_prefix, &config, fixed_buff_size,
+            task_buffer_extra_size, need_set_run_by);
 }
 
 int sf_load_context_from_config_ex(SFContext *sf_context,
-        SFContextIniConfig *config);
+        SFContextIniConfig *config, const int fixed_buff_size,
+        const int task_buffer_extra_size);
 
 static inline int sf_load_context_from_config(SFContext *sf_context,
+        const FCCommunicationType comm_type,
         const char *filename, IniContext *pIniContext,
         const char *section_name, const int default_inner_port,
-        const int default_outer_port)
+        const int default_outer_port, const int fixed_buff_size,
+        const int task_buffer_extra_size)
 {
     SFContextIniConfig config;
 
-    SF_SET_CONTEXT_INI_CONFIG(config, filename, pIniContext,
+    SF_SET_CONTEXT_INI_CONFIG(config, comm_type, filename, pIniContext,
             section_name, default_inner_port, default_outer_port,
             DEFAULT_WORK_THREADS);
-    return sf_load_context_from_config_ex(sf_context, &config);
+    return sf_load_context_from_config_ex(sf_context, &config,
+            fixed_buff_size, task_buffer_extra_size);
 }
 
+int sf_alloc_rdma_pd(SFContext *sf_context,
+        FCAddressPtrArray *address_array);
+
+void sf_set_address_family_by_ip(SFContext *sf_context,
+        FCAddressPtrArray *address_array);
+
 int sf_load_log_config(IniFullContext *ini_ctx, LogContext *log_ctx,
         SFLogConfig *log_cfg);
 
@@ -193,7 +286,14 @@ void sf_log_config_to_string_ex(SFLogConfig *log_cfg, const char *caption,
 void sf_slow_log_config_to_string(SFSlowLogConfig *slow_log_cfg,
         const char *caption, char *output, const int size);
 
-void sf_global_config_to_string(char *output, const int size);
+void sf_global_config_to_string_ex(const char *max_pkg_size_item_nm,
+        char *output, const int size);
+
+static inline void sf_global_config_to_string(char *output, const int size)
+{
+    const char *max_pkg_size_item_nm = "max_pkg_size";
+    sf_global_config_to_string_ex(max_pkg_size_item_nm, output, size);
+}
 
 void sf_context_config_to_string(const SFContext *sf_context,
         char *output, const int size);
@@ -205,10 +305,22 @@ void sf_log_config_ex(const char *other_config);
 #define sf_log_config_to_string(log_cfg, caption, output, size)  \
     sf_log_config_to_string_ex(log_cfg, caption, NULL, output, size)
 
+int sf_get_base_path_from_conf_file(const char *config_filename);
+int sf_load_global_base_path(IniFullContext *ini_ctx);
+
+int sf_load_data_path_config_ex(IniFullContext *ini_ctx,
+        const char *item_name, const char *default_value, string_t *path);
+
+#define sf_load_data_path_config(ini_ctx, path) \
+    sf_load_data_path_config_ex(ini_ctx, "data_path", "data", path)
+
 static inline void sf_set_global_base_path(const char *base_path)
 {
-    snprintf(SF_G_BASE_PATH_STR, sizeof(SF_G_BASE_PATH_STR),
+    string_t path_string;
-            "%s", base_path);
+
+    FC_SET_STRING(path_string, (char *)base_path);
+    SF_G_BASE_PATH_LEN = normalize_path(NULL, &path_string,
+            SF_G_BASE_PATH_STR, sizeof(SF_G_BASE_PATH_STR));
     SF_G_BASE_PATH_INITED = true;
 }
 

src/sf_iov.c

@@ -0,0 +1,197 @@
+/*
+ * Copyright (c) 2020 YuQing <384681@qq.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <https://www.gnu.org/licenses/>.
+ */
+
+#include "fastcommon/logger.h"
+#include "fastcommon/sockopt.h"
+#include "sf_define.h"
+#include "sf_iov.h"
+
+int sf_iova_consume(SFDynamicIOVArray *iova, const int consume_len)
+{
+    struct iovec *iob;
+    struct iovec *end;
+    int sum_bytes;
+    int remain_len;
+    int result;
+
+    if (iova->cnt <= 0) {
+        logError("file: "__FILE__", line: %d, "
+                "invalid iov count: %d", __LINE__, iova->cnt);
+        return EINVAL;
+    }
+
+    if ((result=sf_iova_check_alloc(iova)) != 0) {
+        return result;
+    }
+
+    end = iova->iov + iova->cnt;
+    iob = iova->iov;
+    sum_bytes = iob->iov_len;
+    for (iob=iob + 1; sum_bytes <= consume_len && iob < end; iob++) {
+        sum_bytes += iob->iov_len;
+    }
+
+    if (sum_bytes < consume_len) {
+        logError("file: "__FILE__", line: %d, "
+                "iov length: %d < consume length: %d",
+                __LINE__, sum_bytes, consume_len);
+        return EOVERFLOW;
+    }
+
+    iova->cnt -= (iob - iova->iov);
+    iova->iov = iob;
+    if (iova->cnt == 0) {
+        struct iovec *last;
+
+        /* update the last iov for next slice */
+        last = iob - 1;
+        last->iov_base = (char *)last->iov_base + last->iov_len;
+        last->iov_len = 0;
+    } else {
+        /* adjust the first element */
+        remain_len = sum_bytes - consume_len;
+        if (remain_len < iob->iov_len) {
+            iob->iov_base = (char *)iob->iov_base +
+                (iob->iov_len - remain_len);
+            iob->iov_len = remain_len;
+        }
+    }
+
+    return 0;
+}
+
+static inline int iova_slice(SFDynamicIOVArray *iova, const int slice_len)
+{
+    struct iovec *iob;
+    struct iovec *end;
+    int sum_bytes;
+    int exceed_len;
+
+    sum_bytes = 0;
+    end = iova->ptr + iova->input.cnt;
+    for (iob=iova->iov; iob<end; iob++) {
+        sum_bytes += iob->iov_len;
+        if (sum_bytes > slice_len) {
+            exceed_len = sum_bytes - slice_len;
+            iob->iov_len -= exceed_len;
+            break;
+        } else if (sum_bytes == slice_len) {
+            break;
+        }
+    }
+
+    if (iob < end) {
+        iova->cnt = (iob - iova->iov) + 1;
+        return 0;
+    } else {
+        logError("file: "__FILE__", line: %d, "
+                "iov remain bytes: %d < slice length: %d",
+                __LINE__, sum_bytes, slice_len);
+        iova->cnt = 0;
+        return EOVERFLOW;
+    }
+}
+
+int sf_iova_first_slice(SFDynamicIOVArray *iova, const int slice_len)
+{
+    int result;
+
+    if ((result=sf_iova_check_alloc(iova)) != 0) {
+        return result;
+    }
+
+    return iova_slice(iova, slice_len);
+}
+
+int sf_iova_next_slice(SFDynamicIOVArray *iova,
+        const int consume_len, const int slice_len)
+{
+    struct iovec *last;
+    const struct iovec *origin;
+    int remain_len;
+    int result;
+
+    if ((result=sf_iova_consume(iova, consume_len)) != 0) {
+        return result;
+    }
+
+    last = iova->iov + iova->cnt - 1;
+    origin = iova->input.iov + (last - iova->ptr);
+    remain_len = ((char *)origin->iov_base + origin->iov_len) -
+        (char *)last->iov_base;
+    if (last->iov_len != remain_len) {
+        last->iov_len = remain_len;
+        if (iova->cnt == 0) {
+            iova->iov = last;
+        }
+    }
+
+    return iova_slice(iova, slice_len);
+}
+
+int sf_iova_memset_ex(const struct iovec *iov, const int iovcnt,
+        int c, const int offset, const int length)
+{
+    const struct iovec *iob;
+    const struct iovec *end;
+    int sum_bytes;
+    int remain_len;
+    int left_bytes;
+    char *start;
+
+    if (length == 0) {
+        return 0;
+    }
+
+    sum_bytes = 0;
+    end = iov + iovcnt;
+    for (iob=iov; iob<end; iob++) {
+        sum_bytes += iob->iov_len;
+        if (sum_bytes > offset) {
+            break;
+        }
+    }
+
+    if (iob == end) {
+        logError("file: "__FILE__", line: %d, "
+                "iov length: %d < (offset: %d + length: %d)",
+                __LINE__, sum_bytes, offset, length);
+        return EOVERFLOW;
+    }
+
+    remain_len = sum_bytes - offset;
+    start = (char *)iob->iov_base + (iob->iov_len - remain_len);
+    if (length <= remain_len) {
+        memset(start, c, length);
+        return 0;
+    }
+
+    memset(start, c, remain_len);
+    left_bytes = length - remain_len;
+    while (++iob < end) {
+        if (left_bytes <= iob->iov_len) {
+            memset(iob->iov_base, c, left_bytes);
+            return 0;
+        }
+
+        memset(iob->iov_base, c, iob->iov_len);
+        left_bytes -= iob->iov_len;
+    }
+
+    logError("file: "__FILE__", line: %d, "
+            "iov length is too short, overflow bytes: %d",
+            __LINE__, left_bytes);
+    return EOVERFLOW;
+}

src/sf_iov.h

@@ -0,0 +1,116 @@
+/*
+ * Copyright (c) 2020 YuQing <384681@qq.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <https://www.gnu.org/licenses/>.
+ */
+
+#ifndef _SF_IOV_H
+#define _SF_IOV_H
+
+#include "fastcommon/shared_func.h"
+#include "sf_types.h"
+
+#define SF_IOV_FIXED_SIZE    256
+
+typedef struct sf_dynamic_iov_array {
+    struct iovec holder[SF_IOV_FIXED_SIZE];
+    struct iovec *ptr;
+
+    struct {
+        const struct iovec *iov;
+        int cnt;
+    } input;
+
+    struct iovec *iov;
+    int cnt;
+} SFDynamicIOVArray;
+
+#define sf_iova_init(iova, _iov, _cnt)  \
+    (iova).input.iov = _iov;            \
+    (iova).iov = (struct iovec *)_iov;  \
+    (iova).cnt = (iova).input.cnt = _cnt
+
+#define sf_iova_destroy(iova) \
+    if ((iova).iov != (struct iovec *)(iova).input.iov && \
+            (iova).ptr != (iova).holder)  \
+            free((iova).ptr)
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+static inline int sf_iova_check_alloc(SFDynamicIOVArray *iova)
+{
+    if (iova->iov == (struct iovec *)iova->input.iov) {
+        if (iova->input.cnt <= SF_IOV_FIXED_SIZE) {
+            iova->ptr = iova->holder;
+        } else {
+            iova->ptr = fc_malloc(iova->input.cnt *
+                    sizeof(struct iovec));
+            if (iova->ptr == NULL) {
+                return ENOMEM;
+            }
+        }
+
+        memcpy(iova->ptr, iova->input.iov, iova->input.cnt *
+                sizeof(struct iovec));
+        iova->iov = iova->ptr;
+    }
+
+    return 0;
+}
+
+int sf_iova_consume(SFDynamicIOVArray *iova, const int consume_len);
+
+int sf_iova_first_slice(SFDynamicIOVArray *iova, const int slice_len);
+
+int sf_iova_next_slice(SFDynamicIOVArray *iova,
+        const int consume_len, const int slice_len);
+
+int sf_iova_memset_ex(const struct iovec *iov, const int iovcnt,
+        int c, const int offset, const int length);
+
+#define sf_iova_memset(iova, c, offset, length) \
+    sf_iova_memset_ex((iova).iov, (iova).cnt, c, offset, length)
+
+static inline void sf_iova_memcpy_ex(const struct iovec *iov,
+        const int iovcnt, const char *buff, const int length)
+{
+    const struct iovec *iob;
+    const struct iovec *end;
+    const char *current;
+    int remain;
+    int bytes;
+
+    current = buff;
+    remain = length;
+    end = iov + iovcnt;
+    for (iob=iov; iob<end; iob++) {
+        bytes = FC_MIN(remain, iob->iov_len);
+        memcpy(iob->iov_base, current, bytes);
+
+        remain -= bytes;
+        if (remain == 0) {
+            break;
+        }
+        current += bytes;
+    }
+}
+
+#define sf_iova_memcpy(iova, buff, length) \
+    sf_iova_memcpy_ex((iova).iov, (iova).cnt, buff, length)
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif

src/sf_nio.h

@@ -22,8 +22,13 @@
 #include <stdlib.h>
 #include <string.h>
 #include "fastcommon/fast_task_queue.h"
+#include "fastcommon/ioevent_loop.h"
 #include "sf_define.h"
 #include "sf_types.h"
+#include "sf_global.h"
+
+#define SF_CTX              (task->handler->fh->ctx)
+#define SF_NET_BUFFER_CFG   SF_CTX->net_buffer_cfg
 
 #ifdef __cplusplus
 extern "C" {
@@ -31,50 +36,63 @@ extern "C" {
 
 void sf_set_parameters_ex(SFContext *sf_context, const int header_size,
         sf_set_body_length_callback set_body_length_func,
-        sf_deal_task_func deal_func, TaskCleanUpCallback cleanup_func,
+        sf_alloc_recv_buffer_callback alloc_recv_buffer_func,
+        sf_send_done_callback send_done_callback,
+        sf_deal_task_callback deal_func, TaskCleanUpCallback cleanup_func,
         sf_recv_timeout_callback timeout_callback, sf_release_buffer_callback
         release_buffer_callback);
 
 #define sf_set_parameters(header_size, set_body_length_func, \
-        deal_func, cleanup_func, timeout_callback)   \
+        alloc_recv_buffer_func, deal_func, cleanup_func, timeout_callback) \
     sf_set_parameters_ex(&g_sf_context, header_size, \
-            set_body_length_func, deal_func, \
+            set_body_length_func, alloc_recv_buffer_func, \
-            cleanup_func, timeout_callback, NULL)
+            deal_func, cleanup_func, timeout_callback, NULL)
 
-static inline void sf_set_deal_task_func_ex(SFContext *sf_context,
+static inline void sf_set_deal_task_callback_ex(SFContext *sf_context,
-        sf_deal_task_func deal_func)
+        sf_deal_task_callback deal_func)
 {
-    sf_context->deal_task = deal_func;
+    sf_context->callbacks.deal_task = deal_func;
 }
 
-#define sf_set_deal_task_func(deal_func) \
+#define sf_set_deal_task_callback(deal_func) \
-    sf_set_deal_task_func_ex(&g_sf_context, deal_func)
+    sf_set_deal_task_callback_ex(&g_sf_context, deal_func)
 
-static inline void sf_set_remove_from_ready_list_ex(SFContext *sf_context,
+
-        const bool enabled)
+static inline void sf_set_connect_done_callback_ex(SFContext *sf_context,
+        sf_connect_done_callback done_callback)
 {
-    sf_context->remove_from_ready_list = enabled;
+    sf_context->callbacks.connect_done = done_callback;
 }
 
-#define sf_set_remove_from_ready_list(enabled) \
+#define sf_set_connect_done_callback(done_callback) \
-    sf_set_remove_from_ready_list_ex(&g_sf_context, enabled);
+    sf_set_connect_done_callback_ex(&g_sf_context, done_callback)
 
-static inline TaskCleanUpCallback sf_get_task_cleanup_func_ex(
+
+static inline TaskCleanUpCallback sf_get_task_cleanup_callback_ex(
         SFContext *sf_context)
 {
-    return sf_context->task_cleanup_func;
+    return sf_context->callbacks.task_cleanup;
 }
 
-#define sf_get_task_cleanup_func() \
+#define sf_get_task_cleanup_callback() \
-    sf_get_task_cleanup_func_ex(&g_sf_context)
+    sf_get_task_cleanup_callback_ex(&g_sf_context)
 
-#define sf_nio_task_is_idle(task) \
+#define sf_nio_task_send_done(task) \
-    (task->offset == 0 && task->length == 0)
+    (task->send.ptr->offset == 0 && task->send.ptr->length == 0)
 
-void sf_recv_notify_read(int sock, short event, void *arg);
+static inline void sf_nio_reset_task_length(struct fast_task_info *task)
+{
+    task->send.ptr->length = 0;
+    task->send.ptr->offset = 0;
+    if (task->recv.ptr != task->send.ptr) {
+        task->recv.ptr->length = 0;
+        task->recv.ptr->offset = 0;
+    }
+}
+
+void sf_socket_close_connection(struct fast_task_info *task);
+void sf_recv_notify_read(int sock, const int event, void *arg);
 int sf_send_add_event(struct fast_task_info *task);
-int sf_client_sock_write(int sock, short event, void *arg);
-int sf_client_sock_read(int sock, short event, void *arg);
 
 void sf_task_finish_clean_up(struct fast_task_info *task);
 
@@ -87,6 +105,42 @@ void sf_task_switch_thread(struct fast_task_info *task,
 
 void sf_task_detach_thread(struct fast_task_info *task);
 
+static inline int sf_set_body_length(struct fast_task_info *task)
+{
+    if (SF_CTX->callbacks.set_body_length(task) != 0) {
+        return -1;
+    }
+    if (task->recv.ptr->length < 0) {
+        logError("file: "__FILE__", line: %d, "
+                "client ip: %s, pkg length: %d < 0",
+                __LINE__, task->client_ip,
+                task->recv.ptr->length);
+        return -1;
+    }
+
+    task->recv.ptr->length += SF_CTX->header_size;
+    if (task->recv.ptr->length > SF_NET_BUFFER_CFG.max_pkg_size) {
+        logError("file: "__FILE__", line: %d, "
+                "client ip: %s, pkg length: %d > "
+                "max pkg size: %d", __LINE__,
+                task->client_ip, task->recv.ptr->length,
+                SF_NET_BUFFER_CFG.max_pkg_size);
+        return -1;
+    }
+
+    return 0;
+}
+
+int sf_socket_async_connect_server(struct fast_task_info *task);
+int sf_socket_async_connect_check(struct fast_task_info *task);
+
+ssize_t sf_socket_send_data(struct fast_task_info *task,
+        SFCommAction *action, bool *send_done);
+ssize_t sf_socket_recv_data(struct fast_task_info *task,
+        const bool call_post_recv, SFCommAction *action);
+
+int sf_rdma_busy_polling_callback(struct nio_thread_data *thread_data);
+
 static inline int sf_nio_forward_request(struct fast_task_info *task,
         const int new_thread_index)
 {
@@ -94,11 +148,18 @@ static inline int sf_nio_forward_request(struct fast_task_info *task,
     return sf_nio_notify(task, SF_NIO_STAGE_FORWARDED);
 }
 
-static inline bool sf_client_sock_in_read_stage(struct fast_task_info *task)
+static inline void sf_nio_add_to_deleted_list(struct nio_thread_data
+        *thread_data, struct fast_task_info *task)
 {
-    return (task->event.callback == (IOEventCallback)sf_client_sock_read);
+    if (task->thread_data == thread_data) {
+        ioevent_add_to_deleted_list(task);
+    } else {
+        sf_nio_notify(task, SF_NIO_STAGE_CLOSE);
+    }
 }
 
+bool sf_client_sock_in_read_stage(struct fast_task_info *task);
+
 #ifdef __cplusplus
 }
 #endif

src/sf_ordered_writer.c

@@ -35,7 +35,7 @@
 #include "sf_ordered_writer.h"
 
 #define deal_binlog_one_record(writer, wb) \
-    sf_file_writer_deal_buffer(&(writer)->fw, &wb->bf, wb->version)
+    sf_file_writer_deal_versioned_buffer(&(writer)->fw, &wb->bf, wb->version)
 
 static inline int flush_writer_files(SFOrderedWriterInfo *writer)
 {
@@ -106,7 +106,7 @@ void sf_ordered_writer_finish(SFOrderedWriterContext *ctx)
 {
     int count;
 
-    if (ctx->writer.fw.file.name != NULL) {
+    if (ctx->writer.fw.file.name.str != NULL) {
         fc_queue_terminate(&ctx->thread.queues.version);
 
         count = 0;
@@ -120,8 +120,8 @@ void sf_ordered_writer_finish(SFOrderedWriterContext *ctx)
                     __LINE__, ctx->writer.fw.cfg.subdir_name);
         }
 
-        free(ctx->writer.fw.file.name);
+        free(ctx->writer.fw.file.name.str);
-        ctx->writer.fw.file.name = NULL;
+        ctx->writer.fw.file.name.str = NULL;
     }
 
     if (ctx->writer.fw.file.fd >= 0) {
@@ -142,8 +142,7 @@ static void *binlog_writer_func(void *arg)
 #ifdef OS_LINUX
     {
         char thread_name[64];
-        snprintf(thread_name, sizeof(thread_name),
+        fc_combine_two_strings(thread->name, "writer", '-', thread_name);
-                "%s-writer", thread->name);
         prctl(PR_SET_NAME, thread_name);
     }
 #endif
@@ -179,12 +178,18 @@ static int binlog_wbuffer_alloc_init(void *element, void *args)
     return 0;
 }
 
-static int compare_buffer_version(const SFOrderedWriterBuffer *entry1,
+static int push_compare_buffer_version(const SFOrderedWriterBuffer *entry1,
         const SFOrderedWriterBuffer *entry2)
 {
     return fc_compare_int64(entry1->version, entry2->version);
 }
 
+static int pop_compare_buffer_version(const SFOrderedWriterBuffer *entry,
+        const SFOrderedWriterBuffer *less_equal, void *arg)
+{
+    return fc_compare_int64(entry->version, less_equal->version);
+}
+
 static int sf_ordered_writer_init_thread(SFOrderedWriterContext *context,
         const char *name, const int max_record_size)
 {
@@ -197,7 +202,7 @@ static int sf_ordered_writer_init_thread(SFOrderedWriterContext *context,
 
     thread = &context->thread;
     writer = &context->writer;
-    snprintf(thread->name, sizeof(thread->name), "%s", name);
+    fc_safe_strcpy(thread->name, name);
     writer->fw.cfg.max_record_size = max_record_size;
     writer->thread = thread;
 
@@ -223,9 +228,11 @@ static int sf_ordered_writer_init_thread(SFOrderedWriterContext *context,
     }
 
     if ((result=sorted_queue_init(&thread->queues.buffer, (unsigned long)
-                    (&((SFOrderedWriterBuffer *)NULL)->next),
+                    (&((SFOrderedWriterBuffer *)NULL)->dlink),
                     (int (*)(const void *, const void *))
-                    compare_buffer_version)) != 0)
+                    push_compare_buffer_version,
+                    (int (*)(const void *, const void *, void *arg))
+                    pop_compare_buffer_version, NULL)) != 0)
     {
         return result;
     }
@@ -234,13 +241,16 @@ static int sf_ordered_writer_init_thread(SFOrderedWriterContext *context,
             context, SF_G_THREAD_STACK_SIZE);
 }
 
-int sf_ordered_writer_init(SFOrderedWriterContext *context,
+int sf_ordered_writer_init_ex(SFOrderedWriterContext *context,
         const char *data_path, const char *subdir_name,
-        const int buffer_size, const int max_record_size)
+        const char *file_prefix, const int buffer_size,
+        const int max_record_size, const int64_t file_rotate_size,
+        const bool call_fsync)
 {
     int result;
-    if ((result=sf_file_writer_init_normal(&context->writer.fw,
+    if ((result=sf_file_writer_init(&context->writer.fw, data_path,
-                    data_path, subdir_name, buffer_size)) != 0)
+                    subdir_name, file_prefix, max_record_size,
+                    buffer_size, file_rotate_size, call_fsync)) != 0)
     {
         return result;
     }

src/sf_ordered_writer.h

@@ -29,7 +29,7 @@ typedef struct sf_writer_version_entry {
 typedef struct sf_ordered_writer_buffer {
     int64_t version;
     BufferInfo bf;
-    struct sf_ordered_writer_buffer *next;
+    struct fc_list_head dlink;
 } SFOrderedWriterBuffer;
 
 typedef struct sf_orderd_writer_thread {
@@ -62,9 +62,17 @@ typedef struct sf_ordered_writer_context {
 extern "C" {
 #endif
 
-int sf_ordered_writer_init(SFOrderedWriterContext *context,
+int sf_ordered_writer_init_ex(SFOrderedWriterContext *context,
         const char *data_path, const char *subdir_name,
-        const int buffer_size, const int max_record_size);
+        const char *file_prefix, const int buffer_size,
+        const int max_record_size, const int64_t file_rotate_size,
+        const bool call_fsync);
+
+#define sf_ordered_writer_init(context, data_path, \
+        subdir_name, buffer_size, max_record_size) \
+    sf_ordered_writer_init_ex(context, data_path, subdir_name,   \
+            SF_BINLOG_FILE_PREFIX_STR, buffer_size, max_record_size, \
+            SF_BINLOG_DEFAULT_ROTATE_SIZE, true)
 
 #define sf_ordered_writer_set_flags(ctx, flags) \
     sf_file_writer_set_flags(&(ctx)->writer.fw, flags)
@@ -117,10 +125,11 @@ static inline SFOrderedWriterBuffer *sf_ordered_writer_alloc_buffer(
 #define sf_ordered_writer_set_binlog_index(ctx, binlog_index) \
     sf_file_writer_set_binlog_index(&(ctx)->writer.fw, binlog_index)
 
-#define sf_push_to_binlog_thread_queue(ctx, buffer) \
+#define sf_ordered_writer_push_to_thread_queue(ctx, buffer) \
     sorted_queue_push(&(ctx)->thread.queues.buffer, buffer)
 
-static inline void sf_push_to_binlog_write_queue(SFOrderedWriterContext *ctx,
+static inline void sf_ordered_writer_push_to_queue(
+        SFOrderedWriterContext *ctx,
         SFOrderedWriterBuffer *buffer)
 {
     sorted_queue_push(&ctx->thread.queues.buffer, buffer);

src/sf_proto.c

@@ -14,8 +14,6 @@
  */
 
 
-#include <errno.h>
-#include "fastcommon/shared_func.h"
 #include "sf_util.h"
 #include "sf_nio.h"
 #include "sf_proto.h"
@@ -29,19 +27,23 @@ static int64_t log_slower_than_us = 0;
 int sf_proto_set_body_length(struct fast_task_info *task)
 {
     SFCommonProtoHeader *header;
+    char formatted_ip[FORMATTED_IP_SIZE];
 
-    header = (SFCommonProtoHeader *)task->data;
+    header = (SFCommonProtoHeader *)task->recv.ptr->data;
     if (!SF_PROTO_CHECK_MAGIC(header->magic)) {
+        format_ip_address(task->client_ip, formatted_ip);
         logError("file: "__FILE__", line: %d, "
-                "peer %s:%u, magic "SF_PROTO_MAGIC_FORMAT
+                "%s peer %s:%u, magic "SF_PROTO_MAGIC_FORMAT" is invalid, "
-                " is invalid, expect: "SF_PROTO_MAGIC_FORMAT,
+                "expect: "SF_PROTO_MAGIC_FORMAT", cmd: %d, body length: %d",
-                __LINE__, task->client_ip, task->port,
+                __LINE__, (task->handler != NULL ? task->handler->fh->ctx->
+                    name : ""), formatted_ip, task->port,
                 SF_PROTO_MAGIC_PARAMS(header->magic),
-                SF_PROTO_MAGIC_EXPECT_PARAMS);
+                SF_PROTO_MAGIC_EXPECT_PARAMS, header->cmd,
+                buff2int(header->body_len));
         return EINVAL;
     }
 
-    task->length = buff2int(header->body_len); //set body length
+    task->recv.ptr->length = buff2int(header->body_len); //set body length
     return 0;
 }
 
@@ -70,8 +72,14 @@ int sf_check_response(ConnectionInfo *conn, SFResponseInfo *response,
             response->error.length = response->header.body_len;
         }
 
-        if ((result=tcprecvdata_nb_ex(conn->sock, response->error.message,
+        if (conn->comm_type == fc_comm_type_rdma) {
-                response->error.length, network_timeout, &recv_bytes)) == 0)
+            memcpy(response->error.message, G_RDMA_CONNECTION_CALLBACKS.
+                    get_recv_buffer(conn)->buff + sizeof(SFCommonProtoHeader),
+                    response->error.length);
+            response->error.message[response->error.length] = '\0';
+        } else if ((result=tcprecvdata_nb_ex(conn->sock, response->
+                        error.message, response->error.length,
+                        network_timeout, &recv_bytes)) == 0)
         {
             response->error.message[response->error.length] = '\0';
         } else {
@@ -96,8 +104,37 @@ static inline int sf_recv_response_header(ConnectionInfo *conn,
         SFResponseInfo *response, const int network_timeout)
 {
     int result;
+    BufferInfo *buffer;
     SFCommonProtoHeader header_proto;
 
+    if (conn->comm_type == fc_comm_type_rdma) {
+        buffer = G_RDMA_CONNECTION_CALLBACKS.get_recv_buffer(conn);
+        if (buffer->length < sizeof(SFCommonProtoHeader)) {
+            response->error.length = sprintf(response->error.message,
+                    "recv pkg length: %d < header size: %d",
+                    buffer->length, (int)sizeof(SFCommonProtoHeader));
+            return EINVAL;
+        }
+
+        if ((result=sf_proto_parse_header((SFCommonProtoHeader *)
+                        buffer->buff, response)) != 0)
+        {
+            return result;
+        }
+
+        if (buffer->length != (sizeof(SFCommonProtoHeader) +
+                    response->header.body_len))
+        {
+            response->error.length = snprintf(response->error.message,
+                    sizeof(response->error.message),
+                    "recv package length: %d != calculate: %d",
+                    buffer->length, (int)(sizeof(SFCommonProtoHeader) +
+                        response->header.body_len));
+            return EINVAL;
+        }
+
+        return 0;
+    } else {
         if ((result=tcprecvdata_nb(conn->sock, &header_proto,
                         sizeof(SFCommonProtoHeader), network_timeout)) != 0)
         {
@@ -107,19 +144,8 @@ static inline int sf_recv_response_header(ConnectionInfo *conn,
                     result, STRERROR(result));
             return result;
         }
-
+        return sf_proto_parse_header(&header_proto, response);
-    if (!SF_PROTO_CHECK_MAGIC(header_proto.magic)) {
-        response->error.length = snprintf(response->error.message,
-                sizeof(response->error.message),
-                "magic "SF_PROTO_MAGIC_FORMAT" is invalid, "
-                "expect: "SF_PROTO_MAGIC_FORMAT,
-                SF_PROTO_MAGIC_PARAMS(header_proto.magic),
-                SF_PROTO_MAGIC_EXPECT_PARAMS);
-        return EINVAL;
     }
-
-    sf_proto_extract_header(&header_proto, &response->header);
-    return 0;
 }
 
 int sf_send_and_recv_response_header(ConnectionInfo *conn, char *data,
@@ -127,11 +153,9 @@ int sf_send_and_recv_response_header(ConnectionInfo *conn, char *data,
 {
     int result;
 
-    if ((result=tcpsenddata_nb(conn->sock, data, len, network_timeout)) != 0) {
+    if ((result=sf_proto_send_buf1(conn, data, len,
-        response->error.length = snprintf(response->error.message,
+                    response, network_timeout)) != 0)
-                sizeof(response->error.message),
+    {
-                "send data fail, errno: %d, error info: %s",
-                result, STRERROR(result));
         return result;
     }
 
@@ -194,7 +218,10 @@ int sf_send_and_recv_response_ex(ConnectionInfo *conn, char *send_data,
         return 0;
     }
 
-    if ((result=tcprecvdata_nb_ex(conn->sock, recv_data, response->
+    if (conn->comm_type == fc_comm_type_rdma) {
+        memcpy(recv_data, G_RDMA_CONNECTION_CALLBACKS.get_recv_buffer(conn)->
+                buff + sizeof(SFCommonProtoHeader), response->header.body_len);
+    } else if ((result=tcprecvdata_nb_ex(conn->sock, recv_data, response->
                     header.body_len, network_timeout, &recv_bytes)) != 0)
     {
         response->error.length = snprintf(response->error.message,
@@ -234,7 +261,11 @@ int sf_send_and_recv_response_ex1(ConnectionInfo *conn, char *send_data,
         return EOVERFLOW;
     }
 
-    if ((result=tcprecvdata_nb_ex(conn->sock, recv_data, response->
+    if (conn->comm_type == fc_comm_type_rdma) {
+        memcpy(recv_data, G_RDMA_CONNECTION_CALLBACKS.get_recv_buffer(conn)->
+                buff + sizeof(SFCommonProtoHeader), response->header.body_len);
+        *body_len = response->header.body_len;
+    } else if ((result=tcprecvdata_nb_ex(conn->sock, recv_data, response->
                     header.body_len, network_timeout, body_len)) != 0)
     {
         response->error.length = snprintf(response->error.message,
@@ -275,7 +306,10 @@ int sf_recv_response(ConnectionInfo *conn, SFResponseInfo *response,
         return 0;
     }
 
-    if ((result=tcprecvdata_nb_ex(conn->sock, recv_data, expect_body_len,
+    if (conn->comm_type == fc_comm_type_rdma) {
+        memcpy(recv_data, G_RDMA_CONNECTION_CALLBACKS.get_recv_buffer(conn)->
+                buff + sizeof(SFCommonProtoHeader), response->header.body_len);
+    } else if ((result=tcprecvdata_nb_ex(conn->sock, recv_data, expect_body_len,
                     network_timeout, &recv_bytes)) != 0)
     {
         response->error.length = snprintf(response->error.message,
@@ -343,7 +377,10 @@ int sf_recv_vary_response(ConnectionInfo *conn, SFResponseInfo *response,
         buffer->alloc_size = alloc_size;
     }
 
-    if ((result=tcprecvdata_nb_ex(conn->sock, buffer->buff, response->
+    if (conn->comm_type == fc_comm_type_rdma) {
+        memcpy(buffer->buff, G_RDMA_CONNECTION_CALLBACKS.get_recv_buffer(conn)->
+                buff + sizeof(SFCommonProtoHeader), response->header.body_len);
+    } else if ((result=tcprecvdata_nb_ex(conn->sock, buffer->buff, response->
                     header.body_len, network_timeout, &recv_bytes)) != 0)
     {
         response->error.length = snprintf(response->error.message,
@@ -364,13 +401,9 @@ int sf_send_and_recv_vary_response(ConnectionInfo *conn,
 {
     int result;
 
-    if ((result=tcpsenddata_nb(conn->sock, send_data,
+    if ((result=sf_proto_send_buf1(conn, send_data, send_len,
-                    send_len, network_timeout)) != 0)
+                    response, network_timeout)) != 0)
     {
-        response->error.length = snprintf(response->error.message,
-                sizeof(response->error.message),
-                "send data fail, errno: %d, error info: %s",
-                result, STRERROR(result));
         return result;
     }
 
@@ -411,6 +444,10 @@ const char *sf_get_cmd_caption(const int cmd)
             return "GET_LEADER_REQ";
         case SF_SERVICE_PROTO_GET_LEADER_RESP:
             return "GET_LEADER_RESP";
+        case SF_CLUSTER_PROTO_GET_SERVER_STATUS_REQ:
+            return "GET_SERVER_STATUS_REQ";
+        case SF_CLUSTER_PROTO_GET_SERVER_STATUS_RESP:
+            return "GET_SERVER_STATUS_RESP";
         default:
             return "UNKOWN";
     }
@@ -419,14 +456,17 @@ const char *sf_get_cmd_caption(const int cmd)
 int sf_proto_deal_ack(struct fast_task_info *task,
         SFRequestInfo *request, SFResponseInfo *response)
 {
+    char formatted_ip[FORMATTED_IP_SIZE];
+
     if (request->header.status != 0) {
         if (request->header.body_len > 0) {
             int remain_size;
             int len;
 
+            format_ip_address(task->client_ip, formatted_ip);
             response->error.length = sprintf(response->error.message,
                     "message from peer %s:%u => ",
-                    task->client_ip, task->port);
+                    formatted_ip, task->port);
             remain_size = sizeof(response->error.message) -
                 response->error.length;
             if (request->header.body_len >= remain_size) {
@@ -454,7 +494,8 @@ int sf_proto_deal_ack(struct fast_task_info *task,
 }
 
 int sf_proto_rebind_idempotency_channel(ConnectionInfo *conn,
-        const uint32_t channel_id, const int key, const int network_timeout)
+        const char *service_name, const uint32_t channel_id,
+        const int key, const int network_timeout)
 {
     char out_buff[sizeof(SFCommonProtoHeader) +
         sizeof(SFProtoRebindChannelReq)];
@@ -474,19 +515,20 @@ int sf_proto_rebind_idempotency_channel(ConnectionInfo *conn,
                     sizeof(out_buff), &response, network_timeout,
                     SF_SERVICE_PROTO_REBIND_CHANNEL_RESP)) != 0)
     {
-        sf_log_network_error(&response, conn, result);
+        sf_log_network_error(&response, conn, service_name, result);
     }
 
     return result;
 }
 
 int sf_proto_get_group_servers(ConnectionInfo *conn,
-        const int network_timeout, const int group_id,
+        const char *service_name, const int network_timeout,
-        SFGroupServerArray *sarray)
+        const int group_id, SFGroupServerArray *sarray)
 {
     char out_buff[sizeof(SFCommonProtoHeader) +
         sizeof(SFProtoGetGroupServersReq)];
     char in_buff[1024];
+    char formatted_ip[FORMATTED_IP_SIZE];
     SFCommonProtoHeader *header;
     SFProtoGetGroupServersReq *req;
     SFProtoGetGroupServersRespBodyHeader *body_header;
@@ -509,14 +551,15 @@ int sf_proto_get_group_servers(ConnectionInfo *conn,
                     SF_SERVICE_PROTO_GET_GROUP_SERVERS_RESP, in_buff,
                     sizeof(in_buff), &body_len)) != 0)
     {
-        sf_log_network_error(&response, conn, result);
+        sf_log_network_error(&response, conn, service_name, result);
         return result;
     }
 
     if (body_len < sizeof(SFProtoGetGroupServersRespBodyHeader)) {
+        format_ip_address(conn->ip_addr, formatted_ip);
         logError("file: "__FILE__", line: %d, "
-                "server %s:%d response body length: %d < %d",
+                "server %s:%u response body length: %d < %d",
-                __LINE__, conn->ip_addr, conn->port, body_len,
+                __LINE__, formatted_ip, conn->port, body_len,
                 (int)sizeof(SFProtoGetGroupServersRespBodyHeader));
         return EINVAL;
     }
@@ -524,15 +567,17 @@ int sf_proto_get_group_servers(ConnectionInfo *conn,
     body_header = (SFProtoGetGroupServersRespBodyHeader *)in_buff;
     count = buff2int(body_header->count);
     if (count <= 0) {
+        format_ip_address(conn->ip_addr, formatted_ip);
         logError("file: "__FILE__", line: %d, "
-                "server %s:%d response server count: %d <= 0",
+                "server %s:%u response server count: %d <= 0",
-                __LINE__, conn->ip_addr, conn->port, count);
+                __LINE__, formatted_ip, conn->port, count);
         return EINVAL;
     }
     if (count > sarray->alloc) {
+        format_ip_address(conn->ip_addr, formatted_ip);
         logError("file: "__FILE__", line: %d, "
-                "server %s:%d response server count: %d is too large, "
+                "server %s:%u response server count: %d is too large, "
-                "exceeds %d", __LINE__, conn->ip_addr, conn->port,
+                "exceeds %d", __LINE__, formatted_ip, conn->port,
                 count, sarray->alloc);
         return EOVERFLOW;
     }
@@ -549,9 +594,8 @@ int sf_proto_get_group_servers(ConnectionInfo *conn,
     return 0;
 }
 
-int sf_proto_get_leader(ConnectionInfo *conn,
+int sf_proto_get_leader(ConnectionInfo *conn, const char *service_name,
-        const int network_timeout,
+        const int network_timeout, SFClientServerEntry *leader)
-        SFClientServerEntry *leader)
 {
     int result;
     SFCommonProtoHeader *header;
@@ -567,12 +611,13 @@ int sf_proto_get_leader(ConnectionInfo *conn,
                     SF_SERVICE_PROTO_GET_LEADER_RESP, (char *)&server_resp,
                     sizeof(SFProtoGetServerResp))) != 0)
     {
-        sf_log_network_error(&response, conn, result);
+        sf_log_network_error(&response, conn, service_name, result);
     } else {
         leader->server_id = buff2int(server_resp.server_id);
         memcpy(leader->conn.ip_addr, server_resp.ip_addr, IP_ADDRESS_SIZE);
         *(leader->conn.ip_addr + IP_ADDRESS_SIZE - 1) = '\0';
         leader->conn.port = buff2short(server_resp.port);
+        leader->conn.comm_type = conn->comm_type;
     }
 
     return result;
@@ -585,20 +630,22 @@ void sf_proto_set_handler_context(const SFHandlerContext *ctx)
 }
 
 int sf_proto_deal_task_done(struct fast_task_info *task,
-        SFCommonTaskContext *ctx)
+        const char *service_name, SFCommonTaskContext *ctx)
 {
     SFCommonProtoHeader *proto_header;
     int status;
     int r;
     int64_t time_used;
     int log_level;
+    char formatted_ip[FORMATTED_IP_SIZE];
     char time_buff[32];
 
     if (ctx->log_level != LOG_NOTHING && ctx->response.error.length > 0) {
         log_it_ex(&g_log_context, ctx->log_level,
-                "file: "__FILE__", line: %d, "
+                "file: "__FILE__", line: %d, %s "
                 "peer %s:%u, cmd: %d (%s), req body length: %d, "
-                "resp status: %d, %s", __LINE__, task->client_ip,
+                "resp status: %d, %s", __LINE__, service_name,
+                format_ip_address(task->client_ip, formatted_ip),
                 task->port, ctx->request.header.cmd,
                 GET_CMD_CAPTION(ctx->request.header.cmd),
                 ctx->request.header.body_len, ctx->response.header.status,
@@ -610,36 +657,38 @@ int sf_proto_deal_task_done(struct fast_task_info *task,
             time_used = get_current_time_us() - ctx->req_start_time;
             log_level = GET_CMD_LOG_LEVEL(ctx->request.header.cmd);
             log_it_ex(&g_log_context, log_level, "file: "__FILE__", line: %d, "
-                    "client %s:%u, req cmd: %d (%s), req body_len: %d, "
+                    "%s client %s:%u, req cmd: %d (%s), req body_len: %d, "
-                    "resp status: %d, time used: %s us", __LINE__,
+                    "resp status: %d, time used: %s us", __LINE__, service_name,
-                    task->client_ip, task->port, ctx->request.header.cmd,
+                    format_ip_address(task->client_ip, formatted_ip),
+                    task->port, ctx->request.header.cmd,
                     GET_CMD_CAPTION(ctx->request.header.cmd),
                     ctx->request.header.body_len, ctx->response.header.status,
                     long_to_comma_str(time_used, time_buff));
         }
 
         if (ctx->response.header.status == 0) {
-            task->offset = task->length = 0;
             return sf_set_read_event(task);
         } else {
             return FC_NEGATIVE(ctx->response.header.status);
         }
     }
 
-    proto_header = (SFCommonProtoHeader *)task->data;
+    proto_header = (SFCommonProtoHeader *)task->send.ptr->data;
     if (!ctx->response_done) {
         ctx->response.header.body_len = ctx->response.error.length;
         if (ctx->response.error.length > 0) {
-            memcpy(task->data + sizeof(SFCommonProtoHeader),
+            memcpy(task->send.ptr->data + sizeof(SFCommonProtoHeader),
                     ctx->response.error.message, ctx->response.error.length);
         }
     }
 
     status = sf_unify_errno(FC_ABS(ctx->response.header.status));
     short2buff(status, proto_header->status);
+    short2buff(ctx->response.header.flags, proto_header->flags);
     proto_header->cmd = ctx->response.header.cmd;
     int2buff(ctx->response.header.body_len, proto_header->body_len);
-    task->length = sizeof(SFCommonProtoHeader) + ctx->response.header.body_len;
+    task->send.ptr->length = sizeof(SFCommonProtoHeader) +
+        ctx->response.header.body_len;
 
     r = sf_send_add_event(task);
     time_used = get_current_time_us() - ctx->req_start_time;
@@ -649,10 +698,11 @@ int sf_proto_deal_task_done(struct fast_task_info *task,
         char buff[256];
         int blen;
 
-        blen = sprintf(buff, "timed used: %s us, client %s:%u, "
+        blen = sprintf(buff, "timed used: %s us, %s client %s:%u, "
                 "req cmd: %d (%s), req body len: %d, resp cmd: %d (%s), "
                 "status: %d, resp body len: %d", long_to_comma_str(time_used,
-                    time_buff), task->client_ip, task->port, ctx->request.
+                    time_buff), service_name, format_ip_address(task->
+                        client_ip, formatted_ip), task->port, ctx->request.
                 header.cmd, GET_CMD_CAPTION(ctx->request.header.cmd),
                 ctx->request.header.body_len, ctx->response.header.cmd,
                 GET_CMD_CAPTION(ctx->response.header.cmd),
@@ -663,10 +713,11 @@ int sf_proto_deal_task_done(struct fast_task_info *task,
     if (sf_handler_ctx.callbacks.get_cmd_log_level != NULL) {
         log_level = GET_CMD_LOG_LEVEL(ctx->request.header.cmd);
         log_it_ex(&g_log_context, log_level, "file: "__FILE__", line: %d, "
-                "client %s:%u, req cmd: %d (%s), req body_len: %d, "
+                "%s client %s:%u, req cmd: %d (%s), req body_len: %d, "
                 "resp cmd: %d (%s), status: %d, resp body_len: %d, "
-                "time used: %s us", __LINE__,
+                "time used: %s us", __LINE__, service_name,
-                task->client_ip, task->port, ctx->request.header.cmd,
+                format_ip_address(task->client_ip, formatted_ip),
+                task->port, ctx->request.header.cmd,
                 GET_CMD_CAPTION(ctx->request.header.cmd),
                 ctx->request.header.body_len, ctx->response.header.cmd,
                 GET_CMD_CAPTION(ctx->response.header.cmd),

src/sf_proto.h

@@ -48,6 +48,10 @@
 #define SF_SERVICE_PROTO_REPORT_REQ_RECEIPT_REQ   125
 #define SF_SERVICE_PROTO_REPORT_REQ_RECEIPT_RESP  126
 
+#define SF_CLUSTER_PROTO_GET_SERVER_STATUS_REQ    201
+#define SF_CLUSTER_PROTO_GET_SERVER_STATUS_RESP   202
+
+
 #define SF_PROTO_MAGIC_CHAR        '@'
 #define SF_PROTO_SET_MAGIC(m)   \
     m[0] = m[1] = m[2] = m[3] = SF_PROTO_MAGIC_CHAR
@@ -64,19 +68,17 @@
 #define SF_PROTO_MAGIC_PARAMS(m) \
     m[0], m[1], m[2], m[3]
 
-#define SF_PROTO_SET_HEADER(header, _cmd, _body_len) \
+#define SF_PROTO_SET_HEADER_EX(header, _cmd, _flags, _body_len) \
     do {  \
         SF_PROTO_SET_MAGIC((header)->magic);   \
         (header)->cmd = _cmd;      \
         (header)->status[0] = (header)->status[1] = 0; \
+        short2buff(_flags, (header)->flags); \
         int2buff(_body_len, (header)->body_len); \
     } while (0)
 
-#define SF_PROTO_SET_HEADER_EX(header, _cmd, _flags, _body_len) \
+#define SF_PROTO_SET_HEADER(header, _cmd, _body_len) \
-    do {  \
+        SF_PROTO_SET_HEADER_EX(header, _cmd, 0, _body_len)
-        SF_PROTO_SET_HEADER(header, _cmd, _body_len); \
-        short2buff(_flags, (header)->flags); \
-    } while (0)
 
 #define SF_PROTO_SET_RESPONSE_HEADER(proto_header, resp_header) \
     do {  \
@@ -85,22 +87,32 @@
         int2buff((resp_header).body_len, (proto_header)->body_len);\
     } while (0)
 
-#define SF_PROTO_RESP_BODY(task)  \
+
-    (task->data + sizeof(SFCommonProtoHeader))
+#define SF_PROTO_SEND_BODY(task)  \
+    (task->send.ptr->data + sizeof(SFCommonProtoHeader))
+
+#define SF_PROTO_RECV_BODY(task) \
+    (task->recv.ptr->data + sizeof(SFCommonProtoHeader))
+
+#define SF_RECV_BODY_LENGTH(task) \
+    (task->recv.ptr->length - sizeof(SFCommonProtoHeader))
+
+#define SF_SEND_BUFF_END(task) (task->send.ptr->data + task->send.ptr->size)
+#define SF_RECV_BUFF_END(task) (task->recv.ptr->data + task->recv.ptr->size)
 
 #define SF_PROTO_UPDATE_EXTRA_BODY_SIZE \
     sizeof(SFProtoIdempotencyAdditionalHeader) + FCFS_AUTH_SESSION_ID_LEN
 
 #define SF_PROTO_QUERY_EXTRA_BODY_SIZE  FCFS_AUTH_SESSION_ID_LEN
 
-#define SF_PROTO_CLIENT_SET_REQ(client_ctx, out_buff, \
+#define SF_PROTO_CLIENT_SET_REQ_EX(client_ctx, auth_enabled, \
-        header, req, the_req_id, out_bytes) \
+        out_buff, header, req, the_req_id, out_bytes) \
     do {   \
         char *the_req_start;  \
         header = (SFCommonProtoHeader *)out_buff; \
         the_req_start = (char *)(header + 1);     \
         out_bytes = sizeof(SFCommonProtoHeader) + sizeof(*req); \
-        if (client_ctx->auth.enabled) { \
+        if (auth_enabled) { \
             out_bytes += FCFS_AUTH_SESSION_ID_LEN;   \
             memcpy(the_req_start, client_ctx->auth.ctx-> \
                     session.id, FCFS_AUTH_SESSION_ID_LEN);  \
@@ -117,6 +129,10 @@
         }  \
     } while (0)
 
+#define SF_PROTO_CLIENT_SET_REQ(client_ctx, out_buff, \
+        header, req, the_req_id, out_bytes) \
+   SF_PROTO_CLIENT_SET_REQ_EX(client_ctx, client_ctx->auth.enabled, \
+           out_buff, header, req, the_req_id, out_bytes)
 
 typedef struct sf_common_proto_header {
     unsigned char magic[4]; //magic number
@@ -172,8 +188,8 @@ typedef struct sf_proto_setup_channel_req {
 typedef struct sf_proto_setup_channel_resp {
     char channel_id[4];
     char key[4];
+    char server_id[4];
     char buffer_size[4];
-    char padding[4];
 } SFProtoSetupChannelResp;
 
 typedef struct sf_proto_rebind_channel_req {
@@ -190,6 +206,33 @@ typedef struct sf_proto_report_req_receipt_body {
     char req_id[8];
 } SFProtoReportReqReceiptBody;
 
+typedef struct {
+    unsigned char servers[SF_CLUSTER_CONFIG_SIGN_LEN];
+    unsigned char cluster[SF_CLUSTER_CONFIG_SIGN_LEN];
+} SFProtoConfigSigns;
+
+typedef struct sf_proto_get_server_status_req {
+    SFProtoConfigSigns config_signs;
+    char server_id[4];  //my server id
+    union {
+        char is_leader;
+        char is_master;
+    };
+    char auth_enabled;
+    char padding[2];
+} SFProtoGetServerStatusReq;
+
+typedef struct sf_get_server_status_request {
+    const unsigned char *servers_sign;
+    const unsigned char *cluster_sign;
+    int server_id;  //my server id
+    union {
+        bool is_leader;
+        bool is_master;
+    };
+    bool auth_enabled;
+} SFGetServerStatusRequest;
+
 typedef struct sf_group_server_info {
     int id;
     bool is_leader;
@@ -245,7 +288,17 @@ int sf_proto_set_body_length(struct fast_task_info *task);
 const char *sf_get_cmd_caption(const int cmd);
 
 int sf_proto_deal_task_done(struct fast_task_info *task,
-        SFCommonTaskContext *ctx);
+        const char *service_name, SFCommonTaskContext *ctx);
+
+static inline void sf_proto_init_task_magic(struct fast_task_info *task)
+{
+    SF_PROTO_SET_MAGIC(((SFCommonProtoHeader *)
+                task->send.ptr->data)->magic);
+    if (task->recv.ptr != task->send.ptr) {
+        SF_PROTO_SET_MAGIC(((SFCommonProtoHeader *)
+                    task->recv.ptr->data)->magic);
+    }
+}
 
 static inline void sf_proto_init_task_context(struct fast_task_info *task,
         SFCommonTaskContext *ctx)
@@ -254,58 +307,180 @@ static inline void sf_proto_init_task_context(struct fast_task_info *task,
     ctx->response.header.cmd = SF_PROTO_ACK;
     ctx->response.header.body_len = 0;
     ctx->response.header.status = 0;
+    ctx->response.header.flags = 0;
     ctx->response.error.length = 0;
     ctx->response.error.message[0] = '\0';
     ctx->log_level = LOG_ERR;
     ctx->response_done = false;
     ctx->need_response = true;
 
-    ctx->request.header.cmd = ((SFCommonProtoHeader *)task->data)->cmd;
+    ctx->request.header.cmd = ((SFCommonProtoHeader *)
-    ctx->request.header.body_len = task->length - sizeof(SFCommonProtoHeader);
+            task->recv.ptr->data)->cmd;
+    ctx->request.header.body_len = SF_RECV_BODY_LENGTH(task);
     ctx->request.header.status = buff2short(((SFCommonProtoHeader *)
-                task->data)->status);
+                task->recv.ptr->data)->status);
-    ctx->request.body = task->data + sizeof(SFCommonProtoHeader);
+    ctx->request.header.flags = buff2short(((SFCommonProtoHeader *)
+                task->recv.ptr->data)->flags);
+    if (task->recv_body != NULL) {
+        ctx->request.body = task->recv_body;
+    } else {
+        ctx->request.body = SF_PROTO_RECV_BODY(task);
+    }
 }
 
+/* task send and recv buffer operations */
+static inline int sf_set_task_send_buffer_size(
+        struct fast_task_info *task, const int expect_size)
+{
+    int result;
+    if ((result=free_queue_set_buffer_size(task, task->send.ptr,
+                    expect_size)) == 0)
+    {
+        SF_PROTO_SET_MAGIC(((SFCommonProtoHeader *)
+                    task->send.ptr->data)->magic);
+    }
+    return result;
+}
+
+static inline int sf_set_task_recv_buffer_size(
+        struct fast_task_info *task, const int expect_size)
+{
+    int result;
+    if ((result=free_queue_set_buffer_size(task, task->recv.ptr,
+                    expect_size)) == 0)
+    {
+        SF_PROTO_SET_MAGIC(((SFCommonProtoHeader *)
+                    task->recv.ptr->data)->magic);
+    }
+    return result;
+}
+
+static inline int sf_set_task_send_max_buffer_size(
+        struct fast_task_info *task)
+{
+    int result;
+    if ((result=free_queue_set_max_buffer_size(task, task->send.ptr)) == 0) {
+        SF_PROTO_SET_MAGIC(((SFCommonProtoHeader *)
+                    task->send.ptr->data)->magic);
+    }
+    return result;
+}
+
+static inline int sf_set_task_recv_max_buffer_size(
+        struct fast_task_info *task)
+{
+    int result;
+    if ((result=free_queue_set_max_buffer_size(task, task->recv.ptr)) == 0) {
+        SF_PROTO_SET_MAGIC(((SFCommonProtoHeader *)
+                    task->recv.ptr->data)->magic);
+    }
+    return result;
+}
+
+static inline int sf_realloc_task_send_buffer(
+        struct fast_task_info *task, const int expect_size)
+{
+    int result;
+    if ((result=free_queue_realloc_buffer(task, task->send.ptr,
+                    expect_size)) == 0)
+    {
+        SF_PROTO_SET_MAGIC(((SFCommonProtoHeader *)
+                    task->send.ptr->data)->magic);
+    }
+    return result;
+}
+
+static inline int sf_realloc_task_recv_buffer(
+        struct fast_task_info *task, const int expect_size)
+{
+    int result;
+    if ((result=free_queue_realloc_buffer(task, task->recv.ptr,
+                    expect_size)) == 0)
+    {
+        SF_PROTO_SET_MAGIC(((SFCommonProtoHeader *)
+                    task->recv.ptr->data)->magic);
+    }
+    return result;
+}
+
+static inline int sf_realloc_task_send_max_buffer(
+        struct fast_task_info *task)
+{
+    int result;
+    if ((result=free_queue_realloc_max_buffer(task, task->send.ptr)) == 0) {
+        SF_PROTO_SET_MAGIC(((SFCommonProtoHeader *)
+                    task->send.ptr->data)->magic);
+    }
+    return result;
+}
+
+static inline int sf_realloc_task_recv_max_buffer(
+        struct fast_task_info *task)
+{
+    int result;
+    if ((result=free_queue_realloc_max_buffer(task, task->recv.ptr)) == 0) {
+        SF_PROTO_SET_MAGIC(((SFCommonProtoHeader *)
+                    task->recv.ptr->data)->magic);
+    }
+    return result;
+}
+
+
 static inline void sf_log_network_error_ex1(SFResponseInfo *response,
-        const ConnectionInfo *conn, const int result,
+        const ConnectionInfo *conn, const char *service_name,
-        const int log_level, const char *file, const int line)
+        const int result, const int log_level,
+        const char *file, const int line)
 {
     if (response->error.length > 0) {
-        log_it_ex(&g_log_context, log_level,
+        log_it_ex(&g_log_context, log_level, "file: %s, line: %d, "
-                "file: %s, line: %d, "
+                "%s%sserver %s:%u response message: %s", file, line,
-                "server %s:%u response message: %s",
+                (service_name != NULL ? service_name : ""),
-                file, line, conn->ip_addr, conn->port,
+                (service_name != NULL ? " ": ""),
+                conn->ip_addr, conn->port,
                 response->error.message);
     } else {
-        log_it_ex(&g_log_context, log_level,
+        log_it_ex(&g_log_context, log_level, "file: %s, line: %d, "
-                "file: %s, line: %d, "
+                "communicate with %s%sserver %s:%u fail, "
-                "communicate with server %s:%u fail, "
                 "errno: %d, error info: %s", file, line,
+                (service_name != NULL ? service_name : ""),
+                (service_name != NULL ? " ": ""),
                 conn->ip_addr, conn->port,
                 result, STRERROR(result));
     }
 }
 
-#define sf_log_network_error_ex(response, conn, result, log_level) \
+#define sf_log_network_error_ex(response, conn, \
-    sf_log_network_error_ex1(response, conn, result, \
+        service_name, result, log_level) \
-            log_level, __FILE__, __LINE__)
+    sf_log_network_error_ex1(response, conn, service_name, \
+            result, log_level, __FILE__, __LINE__)
 
-#define sf_log_network_error(response, conn, result) \
+#define sf_log_network_error(response, conn, service_name, result) \
-    sf_log_network_error_ex1(response, conn, result, \
+    sf_log_network_error_ex1(response, conn, service_name, result, \
             LOG_ERR, __FILE__, __LINE__)
 
-#define sf_log_network_error_for_update(response, conn, result)  \
+#define sf_log_network_error_for_update_ex(response, conn,  \
-        sf_log_network_error_ex(response, conn, result,         \
+        service_name, result, enoent_log_level, file, line) \
-                (result == SF_RETRIABLE_ERROR_CHANNEL_INVALID) ? \
+        sf_log_network_error_ex1(response, conn, service_name, result, \
-                LOG_DEBUG : LOG_ERR)
-
-#define sf_log_network_error_for_delete(response, \
-        conn, result, enoent_log_level)  \
-        sf_log_network_error_ex(response, conn, result,          \
                 (result == SF_RETRIABLE_ERROR_CHANNEL_INVALID) ?       \
                 LOG_DEBUG : ((result == ENOENT || result == ENODATA) ? \
-                    enoent_log_level : LOG_ERR))
+                    enoent_log_level : LOG_ERR), file, line)
+
+#define sf_log_network_error_for_update(response, conn, service_name, result) \
+    sf_log_network_error_for_update_ex(response, conn, service_name, \
+            result, LOG_ERR, __FILE__, __LINE__)
+
+#define sf_log_network_error_for_delete_ex(response, conn,   \
+        service_name, result, enoent_log_level, file, line)  \
+        sf_log_network_error_ex1(response, conn, service_name, result,  \
+                (result == SF_RETRIABLE_ERROR_CHANNEL_INVALID) ? \
+                LOG_DEBUG : ((result == ENOENT || result == ENODATA) ? \
+                    enoent_log_level : LOG_ERR), file, line)
+
+#define sf_log_network_error_for_delete(response,      \
+        conn, service_name, result, enoent_log_level)  \
+        sf_log_network_error_for_delete_ex(response, conn, service_name, \
+                result, enoent_log_level, __FILE__, __LINE__)
+
 
 static inline int sf_server_expect_body_length(SFResponseInfo *response,
         const int body_length, const int expect_body_len)
@@ -387,6 +562,16 @@ int sf_recv_response(ConnectionInfo *conn, SFResponseInfo *response,
         const int network_timeout, const unsigned char expect_cmd,
         char *recv_data, const int expect_body_len);
 
+static inline int sf_recv_none_body_response(ConnectionInfo *conn,
+        SFResponseInfo *response, const int network_timeout,
+        const unsigned char expect_cmd)
+{
+    char *recv_data = NULL;
+    const int expect_body_len = 0;
+    return sf_recv_response(conn, response, network_timeout,
+            expect_cmd, recv_data, expect_body_len);
+}
+
 int sf_recv_vary_response(ConnectionInfo *conn, SFResponseInfo *response,
         const int network_timeout, const unsigned char expect_cmd,
         SFProtoRecvBuffer *buffer, const int min_body_len);
@@ -421,6 +606,56 @@ static inline void sf_free_recv_buffer(SFProtoRecvBuffer *buffer)
     }
 }
 
+static inline int sf_proto_send_buf1(ConnectionInfo *conn, char *data,
+        const int len, SFResponseInfo *response, const int network_timeout)
+{
+    int result;
+
+    if (conn->comm_type == fc_comm_type_rdma) {
+        result = G_RDMA_CONNECTION_CALLBACKS.request_by_buf1(
+                conn, data, len, network_timeout * 1000);
+    } else {
+        result = tcpsenddata_nb(conn->sock, data, len, network_timeout);
+    }
+    if (result != 0) {
+        response->error.length = snprintf(response->error.message,
+                sizeof(response->error.message),
+                "send data fail, errno: %d, error info: %s",
+                result, STRERROR(result));
+    }
+
+    return result;
+}
+
+static inline int sf_proto_send_buf2(ConnectionInfo *conn, char *buff1,
+        const int length1, char *buff2, const int length2,
+        SFResponseInfo *response, const int network_timeout)
+{
+    int result;
+
+    if (conn->comm_type == fc_comm_type_rdma) {
+        result = G_RDMA_CONNECTION_CALLBACKS.request_by_buf2(
+                conn, buff1, length1, buff2, length2,
+                network_timeout * 1000);
+    } else {
+        if ((result=tcpsenddata_nb(conn->sock, buff1, length1,
+                        network_timeout)) == 0)
+        {
+            result = tcpsenddata_nb(conn->sock, buff2, length2,
+                    network_timeout);
+        }
+    }
+
+    if (result != 0) {
+        response->error.length = snprintf(response->error.message,
+                sizeof(response->error.message),
+                "send data fail, errno: %d, error info: %s",
+                result, STRERROR(result));
+    }
+
+    return result;
+}
+
 int sf_send_and_recv_response_header(ConnectionInfo *conn, char *data,
         const int len, SFResponseInfo *response, const int network_timeout);
 
@@ -481,16 +716,27 @@ int sf_send_and_recv_vary_response(ConnectionInfo *conn,
         const int network_timeout, const unsigned char expect_cmd,
         SFProtoRecvBuffer *buffer, const int min_body_len);
 
-static inline void sf_proto_extract_header(const SFCommonProtoHeader
+static inline int sf_proto_parse_header(const SFCommonProtoHeader
-        *header_proto, SFHeaderInfo *header_info)
+        *header_proto, SFResponseInfo *response)
 {
-    header_info->cmd = header_proto->cmd;
+    if (!SF_PROTO_CHECK_MAGIC(header_proto->magic)) {
-    header_info->body_len = buff2int(header_proto->body_len);
+        response->error.length = snprintf(response->error.message,
-    header_info->flags = buff2short(header_proto->flags);
+                sizeof(response->error.message),
-    header_info->status = buff2short(header_proto->status);
+                "magic "SF_PROTO_MAGIC_FORMAT" is invalid, "
-    if (header_info->status > 255) {
+                "expect: "SF_PROTO_MAGIC_FORMAT,
-        header_info->status = sf_localize_errno(header_info->status);
+                SF_PROTO_MAGIC_PARAMS(header_proto->magic),
+                SF_PROTO_MAGIC_EXPECT_PARAMS);
+        return EINVAL;
     }
+
+    response->header.cmd = header_proto->cmd;
+    response->header.body_len = buff2int(header_proto->body_len);
+    response->header.flags = buff2short(header_proto->flags);
+    response->header.status = buff2short(header_proto->status);
+    if (response->header.status > 255) {
+        response->header.status = sf_localize_errno(response->header.status);
+    }
+    return 0;
 }
 
 static inline void sf_proto_pack_limit(const SFListLimitInfo
@@ -529,16 +775,33 @@ int sf_proto_deal_ack(struct fast_task_info *task,
         SFRequestInfo *request, SFResponseInfo *response);
 
 int sf_proto_rebind_idempotency_channel(ConnectionInfo *conn,
-        const uint32_t channel_id, const int key, const int network_timeout);
+        const char *service_name, const uint32_t channel_id,
+        const int key, const int network_timeout);
 
 int sf_proto_get_group_servers(ConnectionInfo *conn,
-        const int network_timeout, const int group_id,
+        const char *service_name, const int network_timeout,
-        SFGroupServerArray *sarray);
+        const int group_id, SFGroupServerArray *sarray);
 
-int sf_proto_get_leader(ConnectionInfo *conn,
+int sf_proto_get_leader(ConnectionInfo *conn, const char *service_name,
-        const int network_timeout,
+        const int network_timeout, SFClientServerEntry *leader);
-        SFClientServerEntry *leader);
 
+static inline void sf_proto_get_server_status_pack(
+        const SFGetServerStatusRequest *r,
+        SFProtoGetServerStatusReq *req)
+{
+    int2buff(r->server_id, req->server_id);
+    req->is_leader = (r->is_leader ? 1 : 0);
+    req->auth_enabled = (r->auth_enabled ? 1 : 0);
+    memcpy(req->config_signs.servers, r->servers_sign,
+            SF_CLUSTER_CONFIG_SIGN_LEN);
+    if (r->cluster_sign != NULL) {
+        memcpy(req->config_signs.cluster, r->cluster_sign,
+                SF_CLUSTER_CONFIG_SIGN_LEN);
+    } else {
+        memset(req->config_signs.cluster, 0,
+                SF_CLUSTER_CONFIG_SIGN_LEN);
+    }
+}
 
 #define SF_CLIENT_RELEASE_CONNECTION(cm, conn, result) \
     do { \

src/sf_serializer.h

@@ -21,6 +21,7 @@
 #include "fastcommon/common_define.h"
 #include "fastcommon/shared_func.h"
 #include "fastcommon/fast_buffer.h"
+#include "fastcommon/uniq_skiplist.h"
 #include "fastcommon/hash.h"
 
 #define SF_SERIALIZER_VALUE_TYPE_COUNT  12
@@ -438,6 +439,44 @@ static inline int sf_serializer_pack_id_name_array(FastBuffer *buffer,
     return 0;
 }
 
+static inline int sf_serializer_pack_id_name_skiplist(
+        FastBuffer *buffer, const unsigned char fid,
+        UniqSkiplist *sl)
+{
+    int result;
+    int length;
+    SFSerializerPackFieldArray *obj;
+    const id_name_pair_t *pair;
+    UniqSkiplistIterator it;
+    char *p;
+
+    length = sizeof(SFSerializerPackFieldArray);
+    uniq_skiplist_iterator(sl, &it);
+    while ((pair=uniq_skiplist_next(&it)) != NULL) {
+        length += sizeof(int64_t) + pair->name.len +
+            sizeof(SFSerializerPackStringValue);
+    }
+
+    if ((result=fast_buffer_check_inc_size(buffer, length)) != 0) {
+        return result;
+    }
+
+    obj = (SFSerializerPackFieldArray *)(buffer->data + buffer->length);
+    obj->field.id = fid;
+    obj->field.type = sf_serializer_value_type_id_name_array;
+    int2buff(uniq_skiplist_count(sl), obj->value.count);
+
+    p = obj->value.ptr;
+    uniq_skiplist_iterator(sl, &it);
+    while ((pair=uniq_skiplist_next(&it)) != NULL) {
+        long2buff(pair->id, p);
+        p += sizeof(int64_t);
+        SF_SERIALIZER_PACK_STRING_AND_MOVE_PTR(p, &pair->name);
+    }
+    buffer->length += length;
+    return 0;
+}
+
 static inline int sf_serializer_pack_map(FastBuffer *buffer,
         const unsigned char fid, const key_value_pair_t *kv_pairs,
         const int count)

src/sf_service.c

@@ -25,6 +25,7 @@
 #include <string.h>
 #include <errno.h>
 #include <fcntl.h>
+#include <ifaddrs.h>
 #include "fastcommon/logger.h"
 #include "fastcommon/sockopt.h"
 #include "fastcommon/shared_func.h"
@@ -32,9 +33,8 @@
 #include "fastcommon/sched_thread.h"
 #include "fastcommon/ioevent_loop.h"
 #include "fastcommon/fc_memory.h"
-#include "sf_nio.h"
+#include "sf_proto.h"
 #include "sf_util.h"
-#include "sf_global.h"
 #include "sf_service.h"
 
 #if defined(OS_LINUX)
@@ -57,62 +57,46 @@ struct worker_thread_context {
     struct nio_thread_data *thread_data;
 };
 
-struct accept_thread_context {
-    SFContext *sf_context;
-    int server_sock;
-};
-
-
-int sf_init_task(struct fast_task_info *task)
-{
-    task->connect_timeout = SF_G_CONNECT_TIMEOUT; //for client side
-    task->network_timeout = SF_G_NETWORK_TIMEOUT;
-    return 0;
-}
-
 static void *worker_thread_entrance(void *arg);
 
-static int sf_init_free_queues(const int task_arg_size,
+static int sf_init_free_queue(SFContext *sf_context, const char *name,
-        TaskInitCallback init_callback)
+        const bool double_buffers, const bool need_shrink_task_buffer,
+        const int task_padding_size, const int task_arg_size,
+        TaskInitCallback init_callback, void *init_arg)
 {
-#define ALLOC_CONNECTIONS_ONCE 1024
-
-    static bool sf_inited = false;
     int result;
+    int buffer_size;
     int m;
-    int init_connections;
+    int max_m;
     int alloc_conn_once;
 
-    if (sf_inited) {
-        return 0;
-    }
-
-    sf_inited = true;
     if ((result=set_rand_seed()) != 0) {
         logCrit("file: "__FILE__", line: %d, "
                 "set_rand_seed fail, program exit!", __LINE__);
         return result;
     }
 
-    m = g_sf_global_vars.min_buff_size / (64 * 1024);
+    if (strcmp(name, "cluster") == 0 || strcmp(name, "replica") == 0) {
+        buffer_size = FC_MAX(4 * 1024 * 1024, sf_context->
+                net_buffer_cfg.max_buff_size);
+        max_m = 64;
+    } else {
+        buffer_size = sf_context->net_buffer_cfg.min_buff_size;
+        max_m = 16;
+    }
+    m = buffer_size / (64 * 1024);
     if (m == 0) {
         m = 1;
-    } else if (m > 16) {
+    } else if (m > max_m) {
-        m = 16;
+        m = max_m;
-    }
-    alloc_conn_once = ALLOC_CONNECTIONS_ONCE / m;
-    init_connections = g_sf_global_vars.max_connections < alloc_conn_once ?
-        g_sf_global_vars.max_connections : alloc_conn_once;
-    if ((result=free_queue_init_ex2(g_sf_global_vars.max_connections,
-                    init_connections, alloc_conn_once, g_sf_global_vars.
-                    min_buff_size, g_sf_global_vars.max_buff_size,
-                    task_arg_size, init_callback != NULL ?
-                    init_callback : sf_init_task)) != 0)
-    {
-        return result;
     }
+    alloc_conn_once = 256 / m;
 
-    return 0;
+    return free_queue_init_ex2(&sf_context->free_queue, name, double_buffers,
+            need_shrink_task_buffer, sf_context->net_buffer_cfg.max_connections,
+            alloc_conn_once, sf_context->net_buffer_cfg.min_buff_size,
+            sf_context->net_buffer_cfg.max_buff_size, task_padding_size,
+            task_arg_size, init_callback, init_arg);
 }
 
 int sf_service_init_ex2(SFContext *sf_context, const char *name,
@@ -121,14 +105,21 @@ int sf_service_init_ex2(SFContext *sf_context, const char *name,
         ThreadLoopCallback thread_loop_callback,
         sf_accept_done_callback accept_done_callback,
         sf_set_body_length_callback set_body_length_func,
-        sf_deal_task_func deal_func, TaskCleanUpCallback task_cleanup_func,
+        sf_alloc_recv_buffer_callback alloc_recv_buffer_func,
+        sf_send_done_callback send_done_callback,
+        sf_deal_task_callback deal_func, TaskCleanUpCallback task_cleanup_func,
         sf_recv_timeout_callback timeout_callback, const int net_timeout_ms,
-        const int proto_header_size, const int task_arg_size,
+        const int proto_header_size, const int task_padding_size,
-        TaskInitCallback init_callback, sf_release_buffer_callback
+        const int task_arg_size, const bool double_buffers,
-        release_buffer_callback)
+        const bool need_shrink_task_buffer, const bool explicit_post_recv,
+        TaskInitCallback init_callback, void *init_arg,
+        sf_release_buffer_callback release_buffer_callback)
 {
     int result;
     int bytes;
+    int extra_events;
+    int max_entries;
+    int i;
     struct worker_thread_context *thread_contexts;
     struct worker_thread_context *thread_ctx;
     struct nio_thread_data *thread_data;
@@ -136,15 +127,26 @@ int sf_service_init_ex2(SFContext *sf_context, const char *name,
     pthread_t tid;
     pthread_attr_t thread_attr;
 
-    snprintf(sf_context->name, sizeof(sf_context->name), "%s", name);
+    fc_safe_strcpy(sf_context->name, name);
-    sf_context->realloc_task_buffer = g_sf_global_vars.
+    sf_context->connect_need_log = true;
-                    min_buff_size < g_sf_global_vars.max_buff_size;
+    sf_context->realloc_task_buffer = sf_context->net_buffer_cfg.
-    sf_context->accept_done_func = accept_done_callback;
+        min_buff_size < sf_context->net_buffer_cfg.max_buff_size;
+    sf_context->callbacks.accept_done = accept_done_callback;
     sf_set_parameters_ex(sf_context, proto_header_size,
-            set_body_length_func, deal_func, task_cleanup_func,
+            set_body_length_func, alloc_recv_buffer_func,
+            send_done_callback, deal_func, task_cleanup_func,
             timeout_callback, release_buffer_callback);
+    if (explicit_post_recv) {
+        for (i=0; i<SF_ADDRESS_FAMILY_COUNT; i++) {
+            sf_context->handlers[i].handlers[SF_RDMACM_NETWORK_HANDLER_INDEX].
+                explicit_post_recv = true;
+        }
+    }
 
-    if ((result=sf_init_free_queues(task_arg_size, init_callback)) != 0) {
+    if ((result=sf_init_free_queue(sf_context, name, double_buffers,
+                    need_shrink_task_buffer, task_padding_size,
+                    task_arg_size, init_callback, init_arg)) != 0)
+    {
         return result;
     }
 
@@ -169,11 +171,71 @@ int sf_service_init_ex2(SFContext *sf_context, const char *name,
         return ENOMEM;
     }
 
+    if (SF_G_EPOLL_EDGE_TRIGGER) {
+#ifdef OS_LINUX
+#if IOEVENT_USE_EPOLL
+        extra_events = EPOLLET;
+#else
+        extra_events = 0;
+#endif
+#elif defined(OS_FREEBSD)
+        extra_events = EV_CLEAR;
+#else
+        extra_events = 0;
+#endif
+    } else {
+        extra_events = 0;
+    }
+
+    max_entries = (sf_context->net_buffer_cfg.max_connections +
+            sf_context->work_threads - 1) / sf_context->work_threads;
+    if (strcmp(sf_context->name, "cluster") == 0 ||
+            strcmp(sf_context->name, "replica") == 0)
+    {
+        if (max_entries < 1024) {
+            max_entries += 8;
+        } else {
+            max_entries = 1024;
+        }
+    } else {
+        if (max_entries < 4 * 1024) {
+            max_entries = max_entries * 2;
+        } else if (max_entries < 8 * 1024) {
+            max_entries = (max_entries * 3) / 2;
+        } else if (max_entries < 16 * 1024) {
+            max_entries = (max_entries * 5) / 4;
+        } else if (max_entries < 32 * 1024) {
+            max_entries = (max_entries * 6) / 5;
+        } else if (max_entries < 64 * 1024) {
+            max_entries = (max_entries * 11) / 10;
+        } else if (max_entries < 128 * 1024) {
+            max_entries = (max_entries * 21) / 20;
+        }
+
+#if IOEVENT_USE_URING
+        if (sf_context->use_io_uring) {
+            if (max_entries > 32 * 1024) {
+                max_entries = 32 * 1024;
+            }
+        }
+#endif
+    }
+
+    g_current_time = time(NULL);
     sf_context->thread_count = 0;
     data_end = sf_context->thread_data + sf_context->work_threads;
     for (thread_data=sf_context->thread_data,thread_ctx=thread_contexts;
             thread_data<data_end; thread_data++,thread_ctx++)
     {
+        thread_data->timeout_ms = net_timeout_ms;
+        FC_INIT_LIST_HEAD(&thread_data->polling_queue);
+        if (sf_context->smart_polling.enabled) {
+            thread_data->busy_polling_callback =
+                sf_rdma_busy_polling_callback;
+        } else {
+            thread_data->busy_polling_callback = NULL;
+        }
+
         thread_data->thread_loop_callback = thread_loop_callback;
         if (alloc_thread_extra_data_callback != NULL) {
             thread_data->arg = alloc_thread_extra_data_callback(
@@ -183,32 +245,60 @@ int sf_service_init_ex2(SFContext *sf_context, const char *name,
             thread_data->arg = NULL;
         }
 
-        if (ioevent_init(&thread_data->ev_puller,
+        if ((result=ioevent_init(&thread_data->ev_puller, sf_context->
-            g_sf_global_vars.max_connections + 2, net_timeout_ms, 0) != 0)
+                        name, sf_context->use_io_uring, max_entries,
+                        net_timeout_ms, extra_events)) != 0)
         {
-            result  = errno != 0 ? errno : ENOMEM;
+            char prompt[256];
+#if IOEVENT_USE_URING
+            if (sf_context->use_io_uring) {
+                if (result == EPERM) {
+                    strcpy(prompt, " make sure kernel."
+                            "io_uring_disabled set to 0");
+                } else if (result == EINVAL) {
+                    sprintf(prompt, " maybe max_connections: %d is too large"
+                            " or [%s]'s work_threads: %d is too small",
+                            sf_context->net_buffer_cfg.max_connections,
+                            sf_context->name, sf_context->work_threads);
+                } else {
+                    *prompt = '\0';
+                }
+            } else {
+#endif
+                *prompt = '\0';
+#if IOEVENT_USE_URING
+            }
+#endif
+
             logError("file: "__FILE__", line: %d, "
-                "ioevent_init fail, "
+                "ioevent_init fail, errno: %d, error info: %s.%s"
-                "errno: %d, error info: %s",
+                , __LINE__, result, strerror(result), prompt);
-                __LINE__, result, strerror(result));
             return result;
         }
 
-        result = fast_timer_init(&thread_data->timer,
+#if IOEVENT_USE_URING
-                2 * g_sf_global_vars.network_timeout, g_current_time);
+        if (sf_context->use_io_uring && send_done_callback != NULL) {
+            ioevent_set_send_zc_done_notify(&thread_data->ev_puller, true);
+        }
+#endif
+
+        result = fast_timer_init(&thread_data->timer, 2 * sf_context->
+                net_buffer_cfg.network_timeout, g_current_time);
         if (result != 0) {
             logError("file: "__FILE__", line: %d, "
-                "fast_timer_init fail, "
+                    "fast_timer_init fail, errno: %d, error info: %s",
-                "errno: %d, error info: %s",
                     __LINE__, result, strerror(result));
             return result;
         }
 
-        if ((result=init_pthread_lock(&thread_data->waiting_queue.lock)) != 0) {
+        if ((result=init_pthread_lock(&thread_data->
+                        waiting_queue.lock)) != 0)
+        {
             return result;
         }
 #if defined(OS_LINUX)
-        FC_NOTIFY_READ_FD(thread_data) = eventfd(0, EFD_NONBLOCK);
+        FC_NOTIFY_READ_FD(thread_data) = eventfd(0,
+                EFD_NONBLOCK | EFD_CLOEXEC);
         if (FC_NOTIFY_READ_FD(thread_data) < 0) {
             result = errno != 0 ? errno : EPERM;
             logError("file: "__FILE__", line: %d, "
@@ -232,6 +322,8 @@ int sf_service_init_ex2(SFContext *sf_context, const char *name,
         {
             break;
         }
+        FC_SET_CLOEXEC(FC_NOTIFY_READ_FD(thread_data));
+        FC_SET_CLOEXEC(FC_NOTIFY_WRITE_FD(thread_data));
 #endif
 
         thread_ctx->sf_context = sf_context;
@@ -256,7 +348,7 @@ int sf_service_destroy_ex(SFContext *sf_context)
 {
     struct nio_thread_data *data_end, *thread_data;
 
-    free_queue_destroy();
+    free_queue_destroy(&sf_context->free_queue);
     data_end = sf_context->thread_data + sf_context->work_threads;
     for (thread_data=sf_context->thread_data; thread_data<data_end;
             thread_data++)
@@ -309,7 +401,7 @@ static void *worker_thread_entrance(void *arg)
 
     ioevent_loop(thread_ctx->thread_data,
             sf_recv_notify_read,
-            thread_ctx->sf_context->task_cleanup_func,
+            thread_ctx->sf_context->callbacks.task_cleanup,
             &g_sf_global_vars.continue_flag);
     ioevent_destroy(&thread_ctx->thread_data->ev_puller);
 
@@ -324,15 +416,20 @@ static void *worker_thread_entrance(void *arg)
     return NULL;
 }
 
-static int _socket_server(const char *bind_addr, int port, int *sock)
+int sf_socket_create_server(SFListener *listener,
+        int af, const char *bind_addr)
 {
     int result;
-    *sock = socketServer(bind_addr, port, &result);
+
-    if (*sock < 0) {
+    listener->sock = socketServer2(af, bind_addr,
+            listener->port, &result);
+    if (listener->sock < 0) {
         return result;
     }
 
-    if ((result=tcpsetserveropt(*sock, g_sf_global_vars.network_timeout)) != 0) {
+    if ((result=tcpsetserveropt(listener->sock, listener->handler->
+                    fh->ctx->net_buffer_cfg.network_timeout)) != 0)
+    {
         return result;
     }
 
@@ -342,55 +439,113 @@ static int _socket_server(const char *bind_addr, int port, int *sock)
 int sf_socket_server_ex(SFContext *sf_context)
 {
     int result;
+    int i;
+    bool dual_ports;
     const char *bind_addr;
+    SFAddressFamilyHandler *fh;
+    SFNetworkHandler *handler;
+    SFNetworkHandler *end;
 
-    sf_context->inner_sock = sf_context->outer_sock = -1;
+    for (i=0; i<SF_ADDRESS_FAMILY_COUNT; i++) {
-    if (sf_context->outer_port == sf_context->inner_port) {
+        fh = sf_context->handlers + i;
-        if (*sf_context->outer_bind_addr == '\0' ||
+        if (fh->af == AF_UNSPEC) {
-                *sf_context->inner_bind_addr == '\0') {
+            continue;
+        }
+
+        end = fh->handlers + SF_NETWORK_HANDLER_COUNT;
+        for (handler=fh->handlers; handler<end; handler++) {
+            if (!handler->enabled) {
+                continue;
+            }
+
+            handler->inner.enabled = false;
+            handler->outer.enabled = false;
+            if (handler->outer.port == handler->inner.port) {
+                if (*fh->outer_bind_addr == '\0' ||
+                        *fh->inner_bind_addr == '\0')
+                {
                     bind_addr = "";
-            return _socket_server(bind_addr, sf_context->outer_port,
+                    if ((result=handler->create_server(&handler->
-                    &sf_context->outer_sock);
+                                    outer, fh->af, bind_addr)) != 0)
-        } else if (strcmp(sf_context->outer_bind_addr,
+                    {
-                    sf_context->inner_bind_addr) == 0) {
+                        return result;
-            bind_addr = sf_context->outer_bind_addr;
+                    }
+                    handler->outer.enabled = true;
+                    dual_ports = false;
+                } else if (strcmp(fh->outer_bind_addr,
+                            fh->inner_bind_addr) == 0)
+                {
+                    bind_addr = fh->outer_bind_addr;
                     if (is_private_ip(bind_addr)) {
-                return _socket_server(bind_addr, sf_context->
+                        if ((result=handler->create_server(&handler->
-                        inner_port, &sf_context->inner_sock);
+                                        inner, fh->af, bind_addr)) != 0)
+                        {
+                            return result;
+                        }
+                        handler->inner.enabled = true;
                     } else {
-                return _socket_server(bind_addr, sf_context->
+                        if ((result=handler->create_server(&handler->
-                        outer_port, &sf_context->outer_sock);
+                                        outer, fh->af, bind_addr)) != 0)
+                        {
+                            return result;
                         }
+                        handler->outer.enabled = true;
                     }
+                    dual_ports = false;
+                } else {
+                    dual_ports = true;
+                }
+            } else {
+                dual_ports = true;
             }
 
-    if ((result=_socket_server(sf_context->outer_bind_addr,
+            if (dual_ports) {
-                    sf_context->outer_port, &sf_context->outer_sock)) != 0)
+                if ((result=handler->create_server(&handler->outer,
+                                fh->af, fh->outer_bind_addr)) != 0)
                 {
                     return result;
                 }
 
-    if ((result=_socket_server(sf_context->inner_bind_addr,
+                if ((result=handler->create_server(&handler->inner,
-                    sf_context->inner_port, &sf_context->inner_sock)) != 0)
+                                fh->af, fh->inner_bind_addr)) != 0)
                 {
                     return result;
                 }
+                handler->inner.enabled = true;
+                handler->outer.enabled = true;
+            }
+
+            /*
+               logInfo("%p [%d] inner {port: %d, enabled: %d}, "
+               "outer {port: %d, enabled: %d}", sf_context,
+               (int)(handler-sf_context->handlers),
+               handler->inner.port, handler->inner.enabled,
+               handler->outer.port, handler->outer.enabled);
+             */
+        }
+    }
 
     return 0;
 }
 
-static void accept_run(struct accept_thread_context *accept_context)
+void sf_socket_close_server(SFListener *listener)
+{
+    if (listener->sock >= 0) {
+        close(listener->sock);
+        listener->sock = -1;
+    }
+}
+
+struct fast_task_info *sf_socket_accept_connection(SFListener *listener)
 {
     int incomesock;
     int port;
-    struct sockaddr_in inaddr;
     socklen_t sockaddr_len;
     struct fast_task_info *task;
 
-    while (g_sf_global_vars.continue_flag) {
+    sockaddr_len = sizeof(listener->inaddr);
-        sockaddr_len = sizeof(inaddr);
+    incomesock = accept(listener->sock, (struct sockaddr *)
-        incomesock = accept(accept_context->server_sock,
+            &listener->inaddr, &sockaddr_len);
-                (struct sockaddr*)&inaddr, &sockaddr_len);
     if (incomesock < 0) { //error
         if (!(errno == EINTR || errno == EAGAIN)) {
             logError("file: "__FILE__", line: %d, "
@@ -398,57 +553,101 @@ static void accept_run(struct accept_thread_context *accept_context)
                     __LINE__, errno, strerror(errno));
         }
 
-            continue;
+        return NULL;
     }
 
     if (tcpsetnonblockopt(incomesock) != 0) {
         close(incomesock);
-            continue;
+        return NULL;
     }
+    FC_SET_CLOEXEC(incomesock);
 
-        if ((task=sf_alloc_init_task(accept_context->
+    if ((task=sf_alloc_init_server_task(listener->handler,
-                        sf_context, incomesock)) == NULL)
+                    incomesock)) == NULL)
     {
         close(incomesock);
-            continue;
+        return NULL;
     }
 
     getPeerIpAddPort(incomesock, task->client_ip,
             sizeof(task->client_ip), &port);
     task->port = port;
-        task->thread_data = accept_context->sf_context->thread_data +
+    return task;
-            incomesock % accept_context->sf_context->work_threads;
+}
-        if (accept_context->sf_context->accept_done_func != NULL) {
+
-            accept_context->sf_context->accept_done_func(task,
+void sf_socket_close_ex(SFContext *sf_context)
-                    accept_context->server_sock ==
+{
-                    accept_context->sf_context->inner_sock);
+    int i;
+    SFNetworkHandler *handler;
+    SFNetworkHandler *end;
+
+    for (i=0; i<SF_ADDRESS_FAMILY_COUNT; i++) {
+        if (sf_context->handlers[i].af == AF_UNSPEC) {
+            continue;
+        }
+
+        end = sf_context->handlers[i].handlers + SF_NETWORK_HANDLER_COUNT;
+        for (handler=sf_context->handlers[i].handlers; handler<end; handler++) {
+            if (!handler->enabled) {
+                continue;
+            }
+
+            if (handler->outer.enabled) {
+                handler->close_server(&handler->outer);
+            }
+            if (handler->inner.enabled) {
+                handler->close_server(&handler->inner);
+            }
+        }
+    }
+}
+
+static void accept_run(SFListener *listener)
+{
+    struct fast_task_info *task;
+
+    while (g_sf_global_vars.continue_flag) {
+        if ((task=listener->handler->accept_connection(listener)) == NULL) {
+            continue;
+        }
+
+        task->thread_data = listener->handler->fh->ctx->thread_data +
+            task->event.fd % listener->handler->fh->ctx->work_threads;
+        if (listener->handler->fh->ctx->callbacks.accept_done != NULL) {
+            if (listener->handler->fh->ctx->callbacks.accept_done(task,
+                        listener->inaddr.sin_addr.s_addr,
+                        listener->is_inner) != 0)
+            {
+                listener->handler->close_connection(task);
+                sf_release_task(task);
+                continue;
+            }
         }
 
         if (sf_nio_notify(task, SF_NIO_STAGE_INIT) != 0) {
-            close(incomesock);
+            listener->handler->close_connection(task);
             sf_release_task(task);
         }
     }
 }
 
-static void *accept_thread_entrance(struct accept_thread_context
+static void *accept_thread_entrance(SFListener *listener)
-        *accept_context)
 {
 #ifdef OS_LINUX
     {
         char thread_name[32];
-        snprintf(thread_name, sizeof(thread_name), "%s-listen",
+        snprintf(thread_name, sizeof(thread_name), "%s-%s-listen",
-                accept_context->sf_context->name);
+                listener->handler->comm_type == fc_comm_type_sock ?
+                "sock" : "rdma", listener->handler->fh->ctx->name);
         prctl(PR_SET_NAME, thread_name);
     }
 #endif
 
-    accept_run(accept_context);
+    accept_run(listener);
     return NULL;
 }
 
-void _accept_loop(struct accept_thread_context *accept_context,
+int _accept_loop(SFListener *listener, const int accept_threads)
-        const int accept_threads)
 {
     pthread_t tid;
     pthread_attr_t thread_attr;
@@ -456,7 +655,7 @@ void _accept_loop(struct accept_thread_context *accept_context,
     int i;
 
     if (accept_threads <= 0) {
-       return;
+        return 0;
     }
 
     if ((result=init_pthread_attr(&thread_attr, g_sf_global_vars.
@@ -464,68 +663,83 @@ void _accept_loop(struct accept_thread_context *accept_context,
     {
         logWarning("file: "__FILE__", line: %d, "
                 "init_pthread_attr fail!", __LINE__);
+        return result;
     }
-    else {
+
     for (i=0; i<accept_threads; i++) {
         if ((result=pthread_create(&tid, &thread_attr,
                         (void * (*)(void *))accept_thread_entrance,
-                            accept_context)) != 0)
+                        listener)) != 0)
         {
             logError("file: "__FILE__", line: %d, "
                     "create thread failed, startup threads: %d, "
                     "errno: %d, error info: %s",
                     __LINE__, i, result, strerror(result));
-                break;
+            return result;
         }
     }
 
     pthread_attr_destroy(&thread_attr);
-    }
+    return 0;
 }
 
-void sf_accept_loop_ex(SFContext *sf_context, const bool block)
+int sf_accept_loop_ex(SFContext *sf_context, const bool blocked)
 {
-    struct accept_thread_context *accept_contexts;
+    int i;
-    int count;
+    SFNetworkHandler *handler;
-    int bytes;
+    SFNetworkHandler *hend;
+    SFListener *listeners[SF_ADDRESS_FAMILY_COUNT *
+        SF_NETWORK_HANDLER_COUNT * 2];
+    SFListener **listener;
+    SFListener **last;
+    SFListener **lend;
 
-    if (sf_context->outer_sock >= 0) {
+    listener = listeners;
-        count = 2;
+    for (i=0; i<SF_ADDRESS_FAMILY_COUNT; i++) {
+        if (sf_context->handlers[i].af == AF_UNSPEC) {
+            continue;
+        }
+
+        hend = sf_context->handlers[i].handlers + SF_NETWORK_HANDLER_COUNT;
+        for (handler=sf_context->handlers[i].handlers;
+                handler<hend; handler++)
+        {
+            if (!handler->enabled) {
+                continue;
+            }
+
+            if (handler->inner.enabled) {
+                *listener++ = &handler->inner;
+            }
+            if (handler->outer.enabled) {
+                *listener++ = &handler->outer;
+            }
+        }
+    }
+
+    if (listener == listeners) {
+        logError("file: "__FILE__", line: %d, "
+                "no listener!", __LINE__);
+        return ENOENT;
+    }
+
+    last = listener - 1;
+    if (blocked) {
+        lend = listener - 1;
     } else {
-        count = 1;
+        lend = listener;
     }
 
-    bytes = sizeof(struct accept_thread_context) * count;
+    for (listener=listeners; listener<lend; listener++) {
-    accept_contexts = (struct accept_thread_context *)fc_malloc(bytes);
+        _accept_loop(*listener, sf_context->accept_threads);
-    if (accept_contexts == NULL) {
-        return;
     }
 
-    accept_contexts[0].sf_context = sf_context;
+    if (blocked) {
-    accept_contexts[0].server_sock = sf_context->inner_sock;
+        _accept_loop(*last, sf_context->accept_threads - 1);
-
+        accept_run(*last);
-    if (sf_context->outer_sock >= 0) {
-        accept_contexts[1].sf_context = sf_context;
-        accept_contexts[1].server_sock = sf_context->outer_sock;
-
-        if (sf_context->inner_sock >= 0) {
-            _accept_loop(accept_contexts, sf_context->accept_threads);
     }
 
-        if (block) {
+    return 0;
-            _accept_loop(accept_contexts + 1, sf_context->accept_threads - 1);
-            accept_run(accept_contexts + 1);
-        } else {
-            _accept_loop(accept_contexts + 1, sf_context->accept_threads);
-        }
-    } else {
-        if (block) {
-            _accept_loop(accept_contexts, sf_context->accept_threads - 1);
-            accept_run(accept_contexts);
-        } else {
-            _accept_loop(accept_contexts, sf_context->accept_threads);
-        }
-    }
 }
 
 #if defined(DEBUG_FLAG)
@@ -639,15 +853,13 @@ int sf_setup_signal_handler()
     return 0;
 }
 
-#define LOG_SCHEDULE_ENTRIES_COUNT 3
-
 int sf_startup_schedule(pthread_t *schedule_tid)
 {
     ScheduleArray scheduleArray;
-    ScheduleEntry scheduleEntries[LOG_SCHEDULE_ENTRIES_COUNT];
+    ScheduleEntry scheduleEntries[SF_LOG_SCHEDULE_ENTRIES_COUNT];
 
     scheduleArray.entries = scheduleEntries;
-    sf_setup_schedule(&g_log_context, &g_sf_global_vars.error_log,
+    sf_logger_setup_schedule(&g_log_context, &g_sf_global_vars.error_log,
             &scheduleArray);
     return sched_start(&scheduleArray, schedule_tid,
             g_sf_global_vars.thread_stack_size, (bool * volatile)
@@ -658,7 +870,7 @@ int sf_add_slow_log_schedule(SFSlowLogContext *slowlog_ctx)
 {
     int result;
     ScheduleArray scheduleArray;
-    ScheduleEntry scheduleEntries[LOG_SCHEDULE_ENTRIES_COUNT];
+    ScheduleEntry scheduleEntries[SF_LOG_SCHEDULE_ENTRIES_COUNT];
 
     if (!slowlog_ctx->cfg.enabled) {
         return 0;
@@ -671,8 +883,8 @@ int sf_add_slow_log_schedule(SFSlowLogContext *slowlog_ctx)
     }
 
     scheduleArray.entries = scheduleEntries;
-    sf_setup_schedule(&slowlog_ctx->ctx, &slowlog_ctx->cfg.log_cfg,
+    sf_logger_setup_schedule(&slowlog_ctx->ctx, &slowlog_ctx->
-            &scheduleArray);
+            cfg.log_cfg, &scheduleArray);
     return sched_add_entries(&scheduleArray);
 }
 
@@ -683,6 +895,12 @@ void sf_set_current_time()
     srand(g_sf_global_vars.up_time);
 }
 
+int sf_global_init(const char *log_filename_prefix)
+{
+    sf_set_current_time();
+    return log_set_prefix(SF_G_BASE_PATH_STR, log_filename_prefix);
+}
+
 void sf_enable_thread_notify_ex(SFContext *sf_context, const bool enabled)
 {
     struct nio_thread_data *thread_data;

src/sf_service.h

@@ -25,6 +25,9 @@
 #include "fastcommon/ioevent.h"
 #include "fastcommon/fast_task_queue.h"
 #include "sf_types.h"
+#include "sf_proto.h"
+#include "sf_global.h"
+#include "sf_nio.h"
 
 typedef void* (*sf_alloc_thread_extra_data_callback)(const int thread_index);
 typedef void (*sf_sig_quit_handler)(int sig);
@@ -39,29 +42,35 @@ int sf_service_init_ex2(SFContext *sf_context, const char *name,
         ThreadLoopCallback thread_loop_callback,
         sf_accept_done_callback accept_done_callback,
         sf_set_body_length_callback set_body_length_func,
-        sf_deal_task_func deal_func, TaskCleanUpCallback task_cleanup_func,
+        sf_alloc_recv_buffer_callback alloc_recv_buffer_func,
+        sf_send_done_callback send_done_callback,
+        sf_deal_task_callback deal_func, TaskCleanUpCallback task_cleanup_func,
         sf_recv_timeout_callback timeout_callback, const int net_timeout_ms,
-        const int proto_header_size, const int task_arg_size,
+        const int proto_header_size, const int task_padding_size,
-        TaskInitCallback init_callback, sf_release_buffer_callback
+        const int task_arg_size, const bool double_buffers,
-        release_buffer_callback);
+        const bool need_shrink_task_buffer, const bool explicit_post_recv,
+        TaskInitCallback init_callback, void *init_arg,
+        sf_release_buffer_callback release_buffer_callback);
 
 #define sf_service_init_ex(sf_context, name, alloc_thread_extra_data_callback,\
         thread_loop_callback, accept_done_callback, set_body_length_func,   \
-        deal_func, task_cleanup_func, timeout_callback, net_timeout_ms,   \
+        send_done_callback, deal_func, task_cleanup_func, timeout_callback, \
-        proto_header_size, task_arg_size) \
+        net_timeout_ms, proto_header_size, task_arg_size) \
     sf_service_init_ex2(sf_context, name, alloc_thread_extra_data_callback,   \
         thread_loop_callback, accept_done_callback, set_body_length_func,     \
-        deal_func, task_cleanup_func, timeout_callback, net_timeout_ms,     \
+        NULL, send_done_callback, deal_func, task_cleanup_func, \
-        proto_header_size, task_arg_size, NULL, NULL)
+        timeout_callback, net_timeout_ms, proto_header_size, \
+        0, task_arg_size, false, true, false, NULL, NULL, NULL)
 
 #define sf_service_init(name, alloc_thread_extra_data_callback, \
         thread_loop_callback, accept_done_callback, set_body_length_func,   \
-        deal_func, task_cleanup_func, timeout_callback, net_timeout_ms,   \
+        send_done_callback, deal_func, task_cleanup_func, timeout_callback, \
-        proto_header_size, task_arg_size) \
+        net_timeout_ms, proto_header_size, task_arg_size) \
     sf_service_init_ex2(&g_sf_context, name, alloc_thread_extra_data_callback, \
-        thread_loop_callback, accept_done_callback, set_body_length_func, \
+        thread_loop_callback, accept_done_callback, set_body_length_func, NULL,\
-        deal_func, task_cleanup_func, timeout_callback, net_timeout_ms,   \
+        send_done_callback, deal_func, task_cleanup_func, timeout_callback, \
-        proto_header_size, task_arg_size, NULL, NULL)
+        net_timeout_ms, proto_header_size, 0, task_arg_size, false, true,   \
+        false, NULL, NULL, NULL)
 
 int sf_service_destroy_ex(SFContext *sf_context);
 
@@ -73,17 +82,43 @@ void sf_service_set_thread_loop_callback_ex(SFContext *sf_context,
 #define sf_service_set_thread_loop_callback(thread_loop_callback) \
     sf_service_set_thread_loop_callback_ex(&g_sf_context, thread_loop_callback)
 
+static inline void sf_service_set_smart_polling_ex(SFContext *sf_context,
+        const FCSmartPollingConfig *smart_polling)
+{
+    sf_context->smart_polling = *smart_polling;
+}
+#define sf_service_set_smart_polling(smart_polling) \
+    sf_service_set_smart_polling_ex(&g_sf_context, smart_polling)
+
+static inline void sf_service_set_connect_need_log_ex(
+        SFContext *sf_context, const bool need_log)
+{
+    sf_context->connect_need_log = need_log;
+}
+#define sf_service_set_connect_need_log(need_log) \
+    sf_service_set_connect_need_log_ex(&g_sf_context, need_log)
+
+
 int sf_setup_signal_handler();
 
 int sf_startup_schedule(pthread_t *schedule_tid);
 int sf_add_slow_log_schedule(SFSlowLogContext *slowlog_ctx);
 
 void sf_set_current_time();
+int sf_global_init(const char *log_filename_prefix);
+
+int sf_socket_create_server(SFListener *listener,
+        int af, const char *bind_addr);
+void sf_socket_close_server(SFListener *listener);
+struct fast_task_info *sf_socket_accept_connection(SFListener *listener);
 
 int sf_socket_server_ex(SFContext *sf_context);
 #define sf_socket_server() sf_socket_server_ex(&g_sf_context)
 
-void sf_accept_loop_ex(SFContext *sf_context, const bool block);
+void sf_socket_close_ex(SFContext *sf_context);
+#define sf_socket_close() sf_socket_close_ex(&g_sf_context)
+
+int sf_accept_loop_ex(SFContext *sf_context, const bool blocked);
 
 #define sf_accept_loop()  sf_accept_loop_ex(&g_sf_context, true)
 
@@ -115,14 +150,13 @@ void sf_notify_all_threads_ex(SFContext *sf_context);
 
 void sf_set_sig_quit_handler(sf_sig_quit_handler quit_handler);
 
-int sf_init_task(struct fast_task_info *task);
+static inline struct fast_task_info *sf_alloc_init_task_ex(
-
+        SFNetworkHandler *handler, const int fd,
-static inline struct fast_task_info *sf_alloc_init_task(
+        const int reffer_count)
-        SFContext *sf_context, const int sock)
 {
     struct fast_task_info *task;
 
-    task = free_queue_pop();
+    task = free_queue_pop(&handler->fh->ctx->free_queue);
     if (task == NULL) {
         logError("file: "__FILE__", line: %d, "
                 "malloc task buff failed, you should "
@@ -130,19 +164,57 @@ static inline struct fast_task_info *sf_alloc_init_task(
                 __LINE__);
         return NULL;
     }
-    __sync_add_and_fetch(&task->reffer_count, 1);
+
+    if (task->shrinked) {
+        task->shrinked = false;
+        sf_proto_init_task_magic(task);
+    }
+
+    __sync_add_and_fetch(&task->reffer_count, reffer_count);
     __sync_bool_compare_and_swap(&task->canceled, 1, 0);
-    task->ctx = sf_context;
+    task->handler = handler;
-    task->event.fd = sock;
+    task->event.fd = fd;
+    return task;
+}
+
+#define sf_hold_task_ex(task, inc_count)  fc_hold_task_ex(task, inc_count)
+#define sf_hold_task(task)  fc_hold_task(task)
+
+#define sf_alloc_init_task(handler, fd) sf_alloc_init_task_ex(handler, fd, 1)
+
+static inline struct fast_task_info *sf_alloc_init_server_task(
+        SFNetworkHandler *handler, const int fd)
+{
+    const int reffer_count = 1;
+    struct fast_task_info *task;
+
+    if ((task=sf_alloc_init_task_ex(handler, fd, reffer_count)) != NULL) {
+#if IOEVENT_USE_URING
+        FC_URING_IS_CLIENT(task) = false;
+#endif
+    }
 
     return task;
 }
 
-#define sf_hold_task(task) __sync_add_and_fetch(&task->reffer_count, 1)
+static inline struct fast_task_info *sf_alloc_init_client_task(
+        SFNetworkHandler *handler)
+{
+    const int fd = -1;
+    const int reffer_count = 1;
+    struct fast_task_info *task;
+
+    if ((task=sf_alloc_init_task_ex(handler, fd, reffer_count)) != NULL) {
+#if IOEVENT_USE_URING
+        FC_URING_IS_CLIENT(task) = true;
+#endif
+    }
+
+    return task;
+}
 
 static inline void sf_release_task(struct fast_task_info *task)
 {
-    //int reffer_count;
     if (__sync_sub_and_fetch(&task->reffer_count, 1) == 0) {
         /*
         int free_count = free_queue_count();
@@ -152,13 +224,100 @@ static inline void sf_release_task(struct fast_task_info *task)
                 "used: %d, freed: %d", __LINE__, task,
                 alloc_count, alloc_count - free_count, free_count);
                 */
+
+#if IOEVENT_USE_URING
+        if (SF_CTX->use_io_uring) {
+            task->handler->close_connection(task);
+            __sync_fetch_and_sub(&g_sf_global_vars.
+                    connection_stat.current_count, 1);
+        }
+#endif
+
         free_queue_push(task);
-    } else {
+    }
-        /*
+}
-        logInfo("file: "__FILE__", line: %d, "
+
-                "release task %p, current reffer: %d",
+static inline SFNetworkHandler *sf_get_first_network_handler_ex(
-                __LINE__, task, reffer_count);
+        SFContext *sf_context)
-                */
+{
+    int i;
+    SFNetworkHandler *handler;
+    SFNetworkHandler *end;
+
+    for (i=0; i<SF_ADDRESS_FAMILY_COUNT; i++) {
+        if (sf_context->handlers[i].af == AF_UNSPEC) {
+            continue;
+        }
+
+        end = sf_context->handlers[i].handlers + SF_NETWORK_HANDLER_COUNT;
+        for (handler=sf_context->handlers[i].handlers; handler<end; handler++) {
+            if (handler->enabled) {
+                return handler;
+            }
+        }
+    }
+
+    return NULL;
+}
+#define sf_get_first_network_handler() \
+    sf_get_first_network_handler_ex(&g_sf_context)
+
+
+static inline SFNetworkHandler *sf_get_rdma_network_handler(
+        SFContext *sf_context)
+{
+    int i;
+    SFNetworkHandler *handler;
+
+    for (i=0; i<SF_ADDRESS_FAMILY_COUNT; i++) {
+        if (sf_context->handlers[i].af != AF_UNSPEC) {
+            handler = sf_context->handlers[i].handlers +
+                SF_RDMACM_NETWORK_HANDLER_INDEX;
+            if (handler->enabled) {
+                return handler;
+            }
+        }
+    }
+
+    return NULL;
+}
+
+static inline SFNetworkHandler *sf_get_rdma_network_handler2(
+        SFContext *sf_context1, SFContext *sf_context2)
+{
+    SFNetworkHandler *handler;
+
+    if ((handler=sf_get_rdma_network_handler(sf_context1)) != NULL) {
+        return handler;
+    }
+    return sf_get_rdma_network_handler(sf_context2);
+}
+
+static inline SFNetworkHandler *sf_get_rdma_network_handler3(
+        SFContext *sf_context1, SFContext *sf_context2,
+        SFContext *sf_context3)
+{
+    SFNetworkHandler *handler;
+
+    if ((handler=sf_get_rdma_network_handler(sf_context1)) != NULL) {
+        return handler;
+    }
+    if ((handler=sf_get_rdma_network_handler(sf_context2)) != NULL) {
+        return handler;
+    }
+    return sf_get_rdma_network_handler(sf_context3);
+}
+
+static inline bool sf_get_double_buffers_flag(FCServerGroupInfo *server_group)
+{
+    if (server_group->comm_type == fc_comm_type_sock) {
+#if IOEVENT_USE_URING
+        return true;
+#else
+        return false;
+#endif
+    } else {  //RDMA
+        return true;
     }
 }
 

src/sf_sharding_htable.c

@@ -119,6 +119,7 @@ int sf_sharding_htable_init_ex(SFHtableShardingContext *sharding_ctx,
         const SFShardingHtableKeyType key_type,
         sf_sharding_htable_insert_callback insert_callback,
         sf_sharding_htable_find_callback find_callback,
+        sf_sharding_htable_delete_callback delete_callback,
         sf_sharding_htable_accept_reclaim_callback reclaim_callback,
         const int sharding_count, const int64_t htable_capacity,
         const int allocator_count, const int element_size,
@@ -132,6 +133,7 @@ int sf_sharding_htable_init_ex(SFHtableShardingContext *sharding_ctx,
     if (element_limit <= 0) {
         element_limit = 1000 * 1000;
     }
+
     if ((result=init_allocators(sharding_ctx, allocator_count,
                     element_size, element_limit)) != 0)
     {
@@ -149,7 +151,10 @@ int sf_sharding_htable_init_ex(SFHtableShardingContext *sharding_ctx,
     sharding_ctx->key_type = key_type;
     sharding_ctx->insert_callback = insert_callback;
     sharding_ctx->find_callback = find_callback;
+    sharding_ctx->delete_callback = delete_callback;
     sharding_ctx->accept_reclaim_callback = reclaim_callback;
+
+    sharding_ctx->sharding_reclaim.enabled = (delete_callback == NULL);
     sharding_ctx->sharding_reclaim.elt_water_mark =
         per_elt_limit * low_water_mark_ratio;
     sharding_ctx->sharding_reclaim.min_ttl_ms = min_ttl_ms;
@@ -280,7 +285,7 @@ static SFShardingHashEntry *hash_entry_reclaim(SFHtableSharding *sharding)
     }
 
     if (reclaim_count > 0) {
-        logInfo("sharding index: %d, element_count: %"PRId64", "
+        logDebug("sharding index: %d, element_count: %"PRId64", "
                 "reclaim_ttl_ms: %"PRId64" ms, reclaim_count: %"PRId64", "
                 "reclaim_limit: %"PRId64, (int)(sharding - sharding->ctx->
                     sharding_array.entries), sharding->element_count,
@@ -291,14 +296,16 @@ static SFShardingHashEntry *hash_entry_reclaim(SFHtableSharding *sharding)
 }
 
 static inline SFShardingHashEntry *htable_entry_alloc(
+        SFHtableShardingContext *sharding_ctx,
         SFHtableSharding *sharding)
 {
     SFShardingHashEntry *entry;
     int64_t current_time_ms;
     int64_t last_reclaim_time_ms;
 
-    if (sharding->element_count > sharding->ctx->
+    if (sharding_ctx->sharding_reclaim.enabled &&
-            sharding_reclaim.elt_water_mark)
+            (sharding->element_count > sharding->ctx->
+             sharding_reclaim.elt_water_mark))
     {
         current_time_ms = 1000LL * (int64_t)get_current_time();
         last_reclaim_time_ms = FC_ATOMIC_GET(sharding->last_reclaim_time_ms);
@@ -325,7 +332,6 @@ static inline SFShardingHashEntry *htable_entry_alloc(
 #define SET_SHARDING_AND_BUCKET(sharding_ctx, key) \
     SFHtableSharding *sharding; \
     struct fc_list_head *bucket;   \
-    SFShardingHashEntry *entry; \
     uint64_t hash_code;    \
     \
     hash_code = sf_sharding_htable_key_ids_one == sharding_ctx-> \
@@ -340,6 +346,7 @@ void *sf_sharding_htable_find(SFHtableShardingContext
         *sharding_ctx, const SFTwoIdsHashKey *key, void *arg)
 {
     void *data;
+    SFShardingHashEntry *entry;
     SET_SHARDING_AND_BUCKET(sharding_ctx, key);
 
     PTHREAD_MUTEX_LOCK(&sharding->lock);
@@ -354,9 +361,43 @@ void *sf_sharding_htable_find(SFHtableShardingContext
     return data;
 }
 
+int sf_sharding_htable_delete(SFHtableShardingContext
+        *sharding_ctx, const SFTwoIdsHashKey *key, void *arg)
+{
+    int result;
+    SFShardingHashEntry *entry;
+
+    if (sharding_ctx->delete_callback != NULL) {
+        SET_SHARDING_AND_BUCKET(sharding_ctx, key);
+        PTHREAD_MUTEX_LOCK(&sharding->lock);
+        entry = htable_find(sharding_ctx, key, bucket);
+        if (entry != NULL) {
+            if (sharding_ctx->delete_callback(entry, arg)) {
+                fc_list_del_init(&entry->dlinks.htable);
+                if (sharding_ctx->sharding_reclaim.enabled) {
+                    fc_list_del_init(&entry->dlinks.lru);
+                }
+                fast_mblock_free_object(sharding->allocator, entry);
+                sharding->element_count--;
+            }
+            result = 0;
+        } else {
+            result = ENOENT;
+        }
+        PTHREAD_MUTEX_UNLOCK(&sharding->lock);
+    } else {
+        logError("file: "__FILE__", line: %d, "
+                "delete callback is NULL!", __LINE__);
+        result = EINVAL;
+    }
+
+    return result;
+}
+
 int sf_sharding_htable_insert(SFHtableShardingContext
         *sharding_ctx, const SFTwoIdsHashKey *key, void *arg)
 {
+    SFShardingHashEntry *entry;
     bool new_create;
     int result;
     SET_SHARDING_AND_BUCKET(sharding_ctx, key);
@@ -364,7 +405,7 @@ int sf_sharding_htable_insert(SFHtableShardingContext
     PTHREAD_MUTEX_LOCK(&sharding->lock);
     do {
         if ((entry=htable_find(sharding_ctx, key, bucket)) == NULL) {
-            if ((entry=htable_entry_alloc(sharding)) == NULL) {
+            if ((entry=htable_entry_alloc(sharding_ctx, sharding)) == NULL) {
                 result = ENOMEM;
                 break;
             }
@@ -372,11 +413,15 @@ int sf_sharding_htable_insert(SFHtableShardingContext
             new_create = true;
             entry->key = *key;
             htable_insert(sharding_ctx, entry, bucket);
+            if (sharding_ctx->sharding_reclaim.enabled) {
                 fc_list_add_tail(&entry->dlinks.lru, &sharding->lru);
+            }
         } else {
             new_create = false;
+            if (sharding_ctx->sharding_reclaim.enabled) {
                 fc_list_move_tail(&entry->dlinks.lru, &sharding->lru);
             }
+        }
 
         entry->last_update_time_ms = 1000LL * (int64_t)get_current_time();
         result = sharding_ctx->insert_callback(

src/sf_sharding_htable.h

@@ -18,7 +18,6 @@
 
 #include <limits.h>
 #include <sys/types.h>
-#include <sys/stat.h>
 #include "fastcommon/common_define.h"
 #include "fastcommon/fc_list.h"
 #include "fastcommon/pthread_func.h"
@@ -37,6 +36,9 @@ typedef int (*sf_sharding_htable_insert_callback)
 typedef void *(*sf_sharding_htable_find_callback)
     (struct sf_sharding_hash_entry *entry, void *arg);
 
+typedef bool (*sf_sharding_htable_delete_callback)
+    (struct sf_sharding_hash_entry *entry, void *arg);
+
 typedef bool (*sf_sharding_htable_accept_reclaim_callback)
     (struct sf_sharding_hash_entry *entry);
 
@@ -91,6 +93,7 @@ typedef struct sf_htable_sharding_context {
         int64_t max_ttl_ms;
         double elt_ttl_ms;
         int elt_water_mark;  //trigger reclaim when elements exceeds water mark
+        bool enabled;
     } sharding_reclaim;
 
     struct {
@@ -101,6 +104,7 @@ typedef struct sf_htable_sharding_context {
     SFShardingHtableKeyType key_type;  //id count in the hash entry
     sf_sharding_htable_insert_callback insert_callback;
     sf_sharding_htable_find_callback find_callback;
+    sf_sharding_htable_delete_callback delete_callback;
     sf_sharding_htable_accept_reclaim_callback accept_reclaim_callback;
     SFHtableShardingArray sharding_array;
 } SFHtableShardingContext;
@@ -113,6 +117,7 @@ extern "C" {
             const SFShardingHtableKeyType key_type,
             sf_sharding_htable_insert_callback insert_callback,
             sf_sharding_htable_find_callback find_callback,
+            sf_sharding_htable_delete_callback delete_callback,
             sf_sharding_htable_accept_reclaim_callback reclaim_callback,
             const int sharding_count, const int64_t htable_capacity,
             const int allocator_count, const int element_size,
@@ -123,6 +128,7 @@ extern "C" {
             *sharding_ctx, const SFShardingHtableKeyType key_type,
             sf_sharding_htable_insert_callback insert_callback,
             sf_sharding_htable_find_callback find_callback,
+            sf_sharding_htable_delete_callback delete_callback,
             sf_sharding_htable_accept_reclaim_callback reclaim_callback,
             const int sharding_count, const int64_t htable_capacity,
             const int allocator_count, const int element_size,
@@ -131,10 +137,10 @@ extern "C" {
     {
         const double low_water_mark_ratio = 0.10;
         return sf_sharding_htable_init_ex(sharding_ctx, key_type,
-                insert_callback, find_callback, reclaim_callback,
+                insert_callback, find_callback, delete_callback,
-                sharding_count, htable_capacity, allocator_count,
+                reclaim_callback, sharding_count, htable_capacity,
-                element_size, element_limit, min_ttl_ms, max_ttl_ms,
+                allocator_count, element_size, element_limit,
-                low_water_mark_ratio);
+                min_ttl_ms, max_ttl_ms, low_water_mark_ratio);
     }
 
     int sf_sharding_htable_insert(SFHtableShardingContext
@@ -143,6 +149,9 @@ extern "C" {
     void *sf_sharding_htable_find(SFHtableShardingContext
             *sharding_ctx, const SFTwoIdsHashKey *key, void *arg);
 
+    int sf_sharding_htable_delete(SFHtableShardingContext
+            *sharding_ctx, const SFTwoIdsHashKey *key, void *arg);
+
 #ifdef __cplusplus
 }
 #endif

src/sf_shared_mbuffer.c

@@ -0,0 +1,94 @@
+/*
+ * Copyright (c) 2020 YuQing <384681@qq.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the Lesser GNU General Public License, version 3
+ * or later ("LGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the Lesser GNU General Public License
+ * along with this program. If not, see <https://www.gnu.org/licenses/>.
+ */
+
+#include "sf_shared_mbuffer.h"
+
+static int sf_shared_mbuffer_alloc_init(void *element, void *args)
+{
+    SFSharedMBuffer *buffer;
+
+    buffer = (SFSharedMBuffer *)((char *)element +
+            sizeof(struct fast_allocator_wrapper));
+    buffer->ctx = (SFSharedMBufferContext *)args;
+    return 0;
+}
+
+int sf_shared_mbuffer_init_ex(SFSharedMBufferContext *context,
+        const char *name_prefix, const int buff_extra_size,
+        const int min_buff_size, const int max_buff_size,
+        const int min_alloc_once, const int64_t memory_limit,
+        const bool need_lock)
+{
+    const double expect_usage_ratio = 0.75;
+    const int reclaim_interval = 1;
+    struct fast_region_info regions[32];
+    struct fast_mblock_object_callbacks object_callbacks;
+    int count;
+    int start;
+    int end;
+    int alloc_once;
+    int buff_size;
+    int i;
+
+    alloc_once = (4 * 1024 * 1024) / max_buff_size;
+    if (alloc_once == 0) {
+        alloc_once = min_alloc_once;
+    } else {
+        i = min_alloc_once;
+        while (i < alloc_once) {
+            i *= 2;
+        }
+        alloc_once = i;
+    }
+
+    count = 1;
+    buff_size = min_buff_size;
+    while (buff_size < max_buff_size) {
+        buff_size *= 2;
+        ++count;
+        alloc_once *= 2;
+    }
+
+    buff_size = min_buff_size;
+    start = 0;
+    end = buff_extra_size + buff_size;
+    FAST_ALLOCATOR_INIT_REGION(regions[0], start, end,
+            end - start, alloc_once);
+
+    //logInfo("[1] start: %d, end: %d, alloc_once: %d", start, end, alloc_once);
+
+    start = end;
+    for (i=1; i<count; i++) {
+        buff_size *= 2;
+        alloc_once /= 2;
+        end = buff_extra_size + buff_size;
+        FAST_ALLOCATOR_INIT_REGION(regions[i], start, end,
+                end - start, alloc_once);
+        //logInfo("[%d] start: %d, end: %d, alloc_once: %d", i + 1, start, end, alloc_once);
+        start = end;
+    }
+
+    object_callbacks.init_func = sf_shared_mbuffer_alloc_init;
+    object_callbacks.destroy_func = NULL;
+    object_callbacks.args = context;
+    return fast_allocator_init_ex(&context->allocator, name_prefix,
+            sizeof(SFSharedMBuffer), &object_callbacks, regions, count,
+            memory_limit, expect_usage_ratio, reclaim_interval, need_lock);
+}
+
+void sf_shared_mbuffer_destroy(SFSharedMBufferContext *context)
+{
+    fast_allocator_destroy(&context->allocator);
+}

src/sf_shared_mbuffer.h

@@ -0,0 +1,114 @@
+/*
+ * Copyright (c) 2020 YuQing <384681@qq.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the Lesser GNU General Public License, version 3
+ * or later ("LGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the Lesser GNU General Public License
+ * along with this program. If not, see <https://www.gnu.org/licenses/>.
+ */
+
+#ifndef _SF_SHARED_MBUFFER_H__
+#define _SF_SHARED_MBUFFER_H__
+
+#include "fastcommon/fc_list.h"
+#include "fastcommon/fast_task_queue.h"
+#include "fastcommon/shared_func.h"
+#include "fastcommon/logger.h"
+#include "fastcommon/fast_allocator.h"
+
+typedef struct sf_shared_mbuffer_context {
+    struct fast_allocator_context allocator;
+} SFSharedMBufferContext;
+
+typedef struct sf_shared_mbuffer {
+    int length;
+    volatile int reffer_count;
+    SFSharedMBufferContext *ctx;
+    char buff[0];  //must be last
+} SFSharedMBuffer;
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#define sf_shared_mbuffer_init(context, name_prefix, buff_extra_size, \
+        min_buff_size, max_buff_size, min_alloc_once, memory_limit)   \
+    sf_shared_mbuffer_init_ex(context, name_prefix, buff_extra_size,  \
+        min_buff_size, max_buff_size, min_alloc_once, memory_limit, true)
+
+int sf_shared_mbuffer_init_ex(SFSharedMBufferContext *context,
+        const char *name_prefix, const int buff_extra_size,
+        const int min_buff_size, const int max_buff_size,
+        const int min_alloc_once, const int64_t memory_limit,
+        const bool need_lock);
+
+void sf_shared_mbuffer_destroy(SFSharedMBufferContext *context);
+
+#define sf_shared_mbuffer_alloc(context, buffer_size)  \
+    sf_shared_mbuffer_alloc_ex(context, buffer_size, 1)
+
+static inline SFSharedMBuffer *sf_shared_mbuffer_alloc_ex(
+        SFSharedMBufferContext *context, const int buffer_size,
+        const int init_reffer_count)
+{
+    SFSharedMBuffer *buffer;
+    int sleep_ms;
+
+    sleep_ms = 5;
+    while ((buffer=fast_allocator_alloc(&context->allocator,
+                    buffer_size)) == NULL)
+    {
+        if (sleep_ms < 100) {
+            sleep_ms *= 2;
+        }
+        fc_sleep_ms(sleep_ms);
+    }
+
+    if (init_reffer_count > 0) {
+        __sync_add_and_fetch(&buffer->reffer_count, init_reffer_count);
+    }
+
+    /*
+    logInfo("file: "__FILE__", line: %d, "
+            "alloc shared buffer: %p, buff: %p, reffer_count: %d",
+            __LINE__, buffer, buffer->buff, __sync_add_and_fetch(&buffer->reffer_count, 0));
+            */
+
+    return buffer;
+}
+
+static inline void sf_shared_mbuffer_hold(SFSharedMBuffer *buffer)
+{
+    __sync_add_and_fetch(&buffer->reffer_count, 1);
+}
+
+static inline void sf_shared_mbuffer_release(SFSharedMBuffer *buffer)
+{
+    if (__sync_sub_and_fetch(&buffer->reffer_count, 1) == 0) {
+        /*
+        logInfo("file: "__FILE__", line: %d, "
+                "free shared buffer: %p", __LINE__, buffer);
+                */
+        fast_allocator_free(&buffer->ctx->allocator, buffer);
+    }
+}
+
+static inline void sf_release_task_shared_mbuffer(struct fast_task_info *task)
+{
+    SFSharedMBuffer *mbuffer;
+    mbuffer = fc_list_entry(task->recv_body, SFSharedMBuffer, buff);
+    sf_shared_mbuffer_release(mbuffer);
+    task->recv_body = NULL;
+}
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif

src/sf_types.h

@@ -34,41 +34,170 @@
 #define SF_SERVER_TASK_TYPE_CHANNEL_HOLDER     101   //for request idempotency
 #define SF_SERVER_TASK_TYPE_CHANNEL_USER       102   //for request idempotency
 
-typedef void (*sf_accept_done_callback)(struct fast_task_info *task,
+#define SF_ADDRESS_FAMILY_COUNT           2
-        const bool bInnerPort);
+#define SF_IPV4_ADDRESS_FAMILY_INDEX      0
+#define SF_IPV6_ADDRESS_FAMILY_INDEX      1
+
+#define SF_NETWORK_HANDLER_COUNT          2
+#define SF_SOCKET_NETWORK_HANDLER_INDEX   0
+#define SF_RDMACM_NETWORK_HANDLER_INDEX   1
+
+#define SF_BINLOG_BUFFER_PRODUCER_DATA_LENGTH(bf) ((bf).data_end - (bf).buff)
+#define SF_BINLOG_BUFFER_PRODUCER_BUFF_REMAIN(bf) ((bf).buff_end - (bf).data_end)
+#define SF_BINLOG_BUFFER_CONSUMER_DATA_LENGTH(bf) ((bf).current  - (bf).buff)
+#define SF_BINLOG_BUFFER_CONSUMER_DATA_REMAIN(bf) ((bf).data_end - (bf).current)
+
+typedef int (*sf_accept_done_callback)(struct fast_task_info *task,
+        const in_addr_64_t client_addr, const bool bInnerPort);
 typedef int (*sf_set_body_length_callback)(struct fast_task_info *task);
-typedef int (*sf_deal_task_func)(struct fast_task_info *task, const int stage);
+typedef char *(*sf_alloc_recv_buffer_callback)(struct fast_task_info *task,
+        const int buff_size, bool *new_alloc);
+typedef int (*sf_deal_task_callback)(struct fast_task_info *task, const int stage);
 typedef int (*sf_recv_timeout_callback)(struct fast_task_info *task);
+typedef int (*sf_send_done_callback)(struct fast_task_info *task,
+        const int length, int *next_stage);
+typedef void (*sf_connect_done_callback)(struct fast_task_info *task,
+        const int err_no);
 
 /* calback for release iovec buffer */
 typedef void (*sf_release_buffer_callback)(struct fast_task_info *task);
 
 typedef int (*sf_error_handler_callback)(const int errnum);
 
+typedef enum {
+    sf_comm_action_continue = 'c',
+    sf_comm_action_break = 'b',
+    sf_comm_action_finish = 'f'
+} SFCommAction;
+
+typedef enum {
+    sf_address_family_auto = 0,
+    sf_address_family_ipv4 = 1,
+    sf_address_family_ipv6 = 2,
+    sf_address_family_both = 3
+} SFAddressFamily;
+
+struct ibv_pd;
+struct sf_listener;
+
+typedef int (*sf_get_connection_size_callback)();
+typedef int (*sf_init_connection_callback)(
+        struct fast_task_info *task, void *arg);
+#define sf_alloc_pd_callback fc_alloc_pd_callback
+
+typedef int (*sf_create_server_callback)(struct sf_listener
+        *listener, int af, const char *bind_addr);
+typedef void (*sf_close_server_callback)(struct sf_listener *listener);
+typedef struct fast_task_info * (*sf_accept_connection_callback)(
+        struct sf_listener *listener);
+typedef int (*sf_async_connect_server_callback)(struct fast_task_info *task);
+typedef int (*sf_async_connect_check_callback)(struct fast_task_info *task);
+typedef void (*sf_close_connection_callback)(struct fast_task_info *task);
+
+typedef ssize_t (*sf_send_data_callback)(struct fast_task_info *task,
+        SFCommAction *action, bool *send_done);
+typedef ssize_t (*sf_recv_data_callback)(struct fast_task_info *task,
+        const bool call_post_recv, SFCommAction *action);
+typedef int (*sf_post_recv_callback)(struct fast_task_info *task);
+
+struct sf_network_handler;
+typedef struct sf_listener {
+    struct sf_network_handler *handler;
+    int port;
+    bool enabled;
+    bool is_inner;
+    union {
+        int sock;  //for socket
+        void *id;  //for rdma_cm
+    };
+    struct sockaddr_in inaddr;  //for accept
+} SFListener;
+
+struct sf_context;
+struct sf_address_family_handler;
+
+typedef struct sf_network_handler {
+    bool enabled;
+    bool explicit_post_recv;
+    FCCommunicationType comm_type;
+    struct sf_address_family_handler *fh;
+    struct ibv_pd *pd;
+
+    SFListener inner;
+    SFListener outer;
+
+    /* for server side */
+    sf_get_connection_size_callback get_connection_size;
+    sf_init_connection_callback init_connection;
+    sf_alloc_pd_callback alloc_pd;
+    sf_create_server_callback create_server;
+    sf_close_server_callback close_server;
+    sf_accept_connection_callback accept_connection;
+
+    /* for client side */
+    sf_async_connect_server_callback async_connect_server;
+    sf_async_connect_check_callback  async_connect_check;
+
+    /* server and client both */
+    sf_close_connection_callback close_connection;
+
+    sf_send_data_callback send_data;
+    sf_recv_data_callback recv_data;
+    sf_post_recv_callback post_recv;  //for rdma
+} SFNetworkHandler;
+
+typedef struct sf_nio_callbacks {
+    TaskCleanUpCallback task_cleanup;
+    sf_deal_task_callback deal_task;
+    sf_set_body_length_callback set_body_length;
+    sf_alloc_recv_buffer_callback alloc_recv_buffer;
+    sf_accept_done_callback accept_done;
+    sf_connect_done_callback connect_done;
+    sf_send_done_callback send_done;
+    sf_recv_timeout_callback task_timeout;
+    sf_release_buffer_callback release_buffer;
+} SFNIOCallbacks;
+
+typedef struct sf_address_family_handler {
+    int af;   //AF_UNSPEC for disabled
+    SFNetworkHandler handlers[SF_NETWORK_HANDLER_COUNT];
+    char inner_bind_addr[IP_ADDRESS_SIZE];
+    char outer_bind_addr[IP_ADDRESS_SIZE];
+    struct sf_context *ctx;
+} SFAddressFamilyHandler;
+
+typedef struct sf_net_buffer_config {
+    int connect_timeout;
+    int network_timeout;
+    int max_connections;
+    int max_pkg_size;
+    int min_buff_size;
+    int max_buff_size;
+} SFNetBufferConfig;
+
 typedef struct sf_context {
     char name[64];
     struct nio_thread_data *thread_data;
     volatile int thread_count;
-    int outer_sock;
-    int inner_sock;
 
-    int outer_port;
+    bool is_client;    //since v1.2.5
-    int inner_port;
+    bool use_io_uring; //since v1.2.9
+    bool use_send_zc;  //since v1.2.9
+    SFAddressFamily address_family;
+    SFAddressFamilyHandler handlers[SF_ADDRESS_FAMILY_COUNT];
+
+    SFNetBufferConfig net_buffer_cfg;
+
     int accept_threads;
     int work_threads;
 
-    char inner_bind_addr[IP_ADDRESS_SIZE];
-    char outer_bind_addr[IP_ADDRESS_SIZE];
-
     int header_size;
-    bool remove_from_ready_list;
     bool realloc_task_buffer;
-    sf_deal_task_func deal_task;
+    bool connect_need_log;  //for client connect
-    sf_set_body_length_callback set_body_length;
+    FCSmartPollingConfig smart_polling;
-    sf_accept_done_callback accept_done_func;
+
-    TaskCleanUpCallback task_cleanup_func;
+    SFNIOCallbacks callbacks;
-    sf_recv_timeout_callback timeout_callback;
+    struct fast_task_queue free_queue;
-    sf_release_buffer_callback release_buffer_callback;
 } SFContext;
 
 typedef struct {
@@ -110,7 +239,8 @@ typedef struct sf_binlog_file_position {
 typedef struct server_binlog_buffer {
     char *buff;       //the buffer pointer
     char *current;    //for the consumer
-    char *end;     //data end ptr
+    char *data_end;   //data end ptr
+    char *buff_end;   //buffer end ptr
     int size;         //the buffer size (capacity)
 } SFBinlogBuffer;
 
@@ -237,11 +367,41 @@ typedef struct sf_cluster_config {
 
 typedef struct sf_synchronize_context {
     pthread_lock_cond_pair_t lcp;
+    int result;
     union {
         bool finished;
-        int result;
+        bool ready;
         int waiting_count;
     };
 } SFSynchronizeContext;
 
+typedef enum sf_election_quorum {
+    sf_election_quorum_auto,
+    sf_election_quorum_any,
+    sf_election_quorum_majority
+} SFElectionQuorum;
+
+typedef enum sf_replication_quorum {
+    sf_replication_quorum_auto,
+    sf_replication_quorum_any,
+    sf_replication_quorum_majority,
+    sf_replication_quorum_smart
+} SFReplicationQuorum;
+
+typedef struct sf_block_key {
+    int64_t oid;    //object id
+    int64_t offset; //aligned by block size
+    uint64_t hash_code;
+} SFBlockKey;
+
+typedef struct sf_slice_size {
+    int offset;  //offset within the block
+    int length;  //slice length
+} SFSliceSize;
+
+typedef struct sf_block_slice_key_info {
+    SFBlockKey block;
+    SFSliceSize slice;
+} SFBlockSliceKeyInfo;
+
 #endif

src/sf_util.c

@@ -281,7 +281,7 @@ int sf_logger_init(LogContext *pContext, const char *filename_prefix)
     return 0;
 }
 
-ScheduleEntry *sf_logger_set_schedule_entry(struct log_context *pContext,
+ScheduleEntry *sf_logger_set_schedule_entries(struct log_context *pContext,
         SFLogConfig *log_cfg, ScheduleEntry *pScheduleEntry)
 {
     INIT_SCHEDULE_ENTRY(*pScheduleEntry, sched_generate_next_id(),
@@ -311,6 +311,8 @@ ScheduleEntry *sf_logger_set_schedule_entry(struct log_context *pContext,
 const char *sf_strerror(const int errnum)
 {
     switch (errnum) {
+        case SF_CLUSTER_ERROR_BINLOG_MISSED:
+            return "binlog missed";
         case SF_CLUSTER_ERROR_BINLOG_INCONSISTENT:
             return "binlog inconsistent";
         case SF_CLUSTER_ERROR_LEADER_INCONSISTENT:
@@ -329,6 +331,8 @@ const char *sf_strerror(const int errnum)
             return STRERROR(EINVAL);
         case SF_ERROR_EAGAIN:
             return STRERROR(EAGAIN);
+        case SF_ERROR_EINPROGRESS:
+            return STRERROR(EINPROGRESS);
         case SF_ERROR_EOVERFLOW:
             return STRERROR(EOVERFLOW);
         case SF_ERROR_ENODATA:

src/sf_util.h

@@ -96,14 +96,14 @@ void sf_parse_cmd_option_bool(int argc, char *argv[],
 
 int sf_logger_init(LogContext *pContext, const char *filename_prefix);
 
-ScheduleEntry *sf_logger_set_schedule_entry(struct log_context *pContext,
+ScheduleEntry *sf_logger_set_schedule_entries(struct log_context *pContext,
         SFLogConfig *log_cfg, ScheduleEntry *pScheduleEntry);
 
-static inline void sf_setup_schedule(struct log_context *pContext,
+static inline void sf_logger_setup_schedule(struct log_context *pContext,
         SFLogConfig *log_cfg, ScheduleArray *scheduleArray)
 {
     ScheduleEntry *scheduleEntry;
-    scheduleEntry = sf_logger_set_schedule_entry(pContext,
+    scheduleEntry = sf_logger_set_schedule_entries(pContext,
             log_cfg, scheduleArray->entries);
     scheduleArray->count = scheduleEntry - scheduleArray->entries;
 }
@@ -117,12 +117,20 @@ static inline int sf_unify_errno(const int errnum)
             return SF_ERROR_EINVAL;
         case EAGAIN:
             return SF_ERROR_EAGAIN;
+        case EINPROGRESS:
+            return SF_ERROR_EINPROGRESS;
         case EOVERFLOW:
             return SF_ERROR_EOVERFLOW;
         case EOPNOTSUPP:
             return SF_ERROR_EOPNOTSUPP;
         case ENODATA:
             return SF_ERROR_ENODATA;
+        case ENOLINK:
+            return SF_ERROR_ENOLINK;
+        case ENOTEMPTY:
+            return SF_ERROR_ENOTEMPTY;
+        case ELOOP:
+            return SF_ERROR_ELOOP;
         default:
             return errnum;
     }
@@ -141,6 +149,8 @@ static inline int sf_localize_errno(int errnum)
             return EINVAL;
         case SF_ERROR_EAGAIN:
             return EAGAIN;
+        case SF_ERROR_EINPROGRESS:
+            return EINPROGRESS;
         case SF_ERROR_EOVERFLOW:
             return EOVERFLOW;
         case SF_ERROR_EOPNOTSUPP:
@@ -149,6 +159,12 @@ static inline int sf_localize_errno(int errnum)
             return ENODATA;
         case SF_SESSION_ERROR_NOT_EXIST:
             return EPERM;
+        case SF_ERROR_ENOLINK:
+            return ENOLINK;
+        case SF_ERROR_ENOTEMPTY:
+            return ENOTEMPTY;
+        case SF_ERROR_ELOOP:
+            return ELOOP;
         default:
             return errnum;
     }