Many files:

  Semi-synchronous replication for InnoDB type tables; before telling the client that a commit has been processed, wait that the replication thread has returned from my_net_send() where it sends the binlog to the slave; note that TCP/IP, even with the TCP_NODELAY option does not guarantee that the slave has RECEIVED the data - this is just heuristic at the moment; this is useful in failover: in almost all cases, every transaction that has returned from the commit has been sent and processed in the slave, which makes failover to the slave simpler if the master crashes; the code does not work yet as is, because MySQL should call innobase_report_binlog_offset_and_commit() in a commit; we will most probably return that call to 5.0.x, to make InnoDB Hot Backup and group commit to work again; XA code broke them temporarily in 5.0.3


sql/ha_innodb.cc:
  Semi-synchronous replication for InnoDB type tables; before telling the client that a commit has been processed, wait that the replication thread has returned from my_net_send() where it sends the binlog to the slave; note that TCP/IP, even with the TCP_NODELAY option does not guarantee that the slave has RECEIVED the data - this is just heuristic at the moment; this is useful in failover: in almost all cases, every transaction that has returned from the commit has been sent and processed in the slave, which makes failover to the slave simpler if the master crashes; the code does not work yet as is, because MySQL should call innobase_report_binlog_offset_and_commit() in a commit; we will most probably return that call to 5.0.x, to make InnoDB Hot Backup and group commit to work again; XA code broke them temporarily in 5.0.3
sql/handler.cc:
  Semi-synchronous replication for InnoDB type tables; before telling the client that a commit has been processed, wait that the replication thread has returned from my_net_send() where it sends the binlog to the slave; note that TCP/IP, even with the TCP_NODELAY option does not guarantee that the slave has RECEIVED the data - this is just heuristic at the moment; this is useful in failover: in almost all cases, every transaction that has returned from the commit has been sent and processed in the slave, which makes failover to the slave simpler if the master crashes; the code does not work yet as is, because MySQL should call innobase_report_binlog_offset_and_commit() in a commit; we will most probably return that call to 5.0.x, to make InnoDB Hot Backup and group commit to work again; XA code broke them temporarily in 5.0.3
sql/mysqld.cc:
  Semi-synchronous replication for InnoDB type tables; before telling the client that a commit has been processed, wait that the replication thread has returned from my_net_send() where it sends the binlog to the slave; note that TCP/IP, even with the TCP_NODELAY option does not guarantee that the slave has RECEIVED the data - this is just heuristic at the moment; this is useful in failover: in almost all cases, every transaction that has returned from the commit has been sent and processed in the slave, which makes failover to the slave simpler if the master crashes; the code does not work yet as is, because MySQL should call innobase_report_binlog_offset_and_commit() in a commit; we will most probably return that call to 5.0.x, to make InnoDB Hot Backup and group commit to work again; XA code broke them temporarily in 5.0.3
sql/set_var.cc:
  Semi-synchronous replication for InnoDB type tables; before telling the client that a commit has been processed, wait that the replication thread has returned from my_net_send() where it sends the binlog to the slave; note that TCP/IP, even with the TCP_NODELAY option does not guarantee that the slave has RECEIVED the data - this is just heuristic at the moment; this is useful in failover: in almost all cases, every transaction that has returned from the commit has been sent and processed in the slave, which makes failover to the slave simpler if the master crashes; the code does not work yet as is, because MySQL should call innobase_report_binlog_offset_and_commit() in a commit; we will most probably return that call to 5.0.x, to make InnoDB Hot Backup and group commit to work again; XA code broke them temporarily in 5.0.3
sql/sql_repl.cc:
  Semi-synchronous replication for InnoDB type tables; before telling the client that a commit has been processed, wait that the replication thread has returned from my_net_send() where it sends the binlog to the slave; note that TCP/IP, even with the TCP_NODELAY option does not guarantee that the slave has RECEIVED the data - this is just heuristic at the moment; this is useful in failover: in almost all cases, every transaction that has returned from the commit has been sent and processed in the slave, which makes failover to the slave simpler if the master crashes; the code does not work yet as is, because MySQL should call innobase_report_binlog_offset_and_commit() in a commit; we will most probably return that call to 5.0.x, to make InnoDB Hot Backup and group commit to work again; XA code broke them temporarily in 5.0.3
sql/ha_innodb.h:
  Semi-synchronous replication for InnoDB type tables; before telling the client that a commit has been processed, wait that the replication thread has returned from my_net_send() where it sends the binlog to the slave; note that TCP/IP, even with the TCP_NODELAY option does not guarantee that the slave has RECEIVED the data - this is just heuristic at the moment; this is useful in failover: in almost all cases, every transaction that has returned from the commit has been sent and processed in the slave, which makes failover to the slave simpler if the master crashes; the code does not work yet as is, because MySQL should call innobase_report_binlog_offset_and_commit() in a commit; we will most probably return that call to 5.0.x, to make InnoDB Hot Backup and group commit to work again; XA code broke them temporarily in 5.0.3
sql/handler.h:
  Semi-synchronous replication for InnoDB type tables; before telling the client that a commit has been processed, wait that the replication thread has returned from my_net_send() where it sends the binlog to the slave; note that TCP/IP, even with the TCP_NODELAY option does not guarantee that the slave has RECEIVED the data - this is just heuristic at the moment; this is useful in failover: in almost all cases, every transaction that has returned from the commit has been sent and processed in the slave, which makes failover to the slave simpler if the master crashes; the code does not work yet as is, because MySQL should call innobase_report_binlog_offset_and_commit() in a commit; we will most probably return that call to 5.0.x, to make InnoDB Hot Backup and group commit to work again; XA code broke them temporarily in 5.0.3

sql/ha_innodb.h

@@ -321,3 +321,5 @@ int innobase_rollback_by_xid(
 int innobase_xa_end(THD *thd);
 
 
+int innobase_repl_report_sent_binlog(THD *thd, char *log_file_name,
+                               my_off_t end_offset);

sql/handler.cc

@@ -2411,3 +2411,60 @@ TYPELIB *ha_known_exts(void)
   }
   return &known_extensions;
 }
+
+/*
+  Reports to table handlers up to which position we have sent the binlog
+  to a slave in replication
+
+  SYNOPSIS
+    ha_repl_report_sent_binlog()
+
+  NOTES
+    Only works for InnoDB at the moment
+
+  RETURN VALUE
+    Always 0 (= success)  
+
+  PARAMETERS
+    THD    *thd             in: thread doing the binlog communication to
+                                the slave
+    char   *log_file_name   in: binlog file name
+    my_off_t end_offset     in: the offset in the binlog file up to
+                                which we sent the contents to the slave
+*/
+
+int ha_repl_report_sent_binlog(THD *thd, char *log_file_name,
+                               my_off_t end_offset)
+{
+#ifdef HAVE_INNOBASE_DB
+  return innobase_repl_report_sent_binlog(thd,log_file_name,end_offset);
+#else
+  /* remove warnings about unused parameters */
+  thd=thd; log_file_name=log_file_name; end_offset=end_offset;
+  return 0;
+#endif
+}
+
+/*
+  Reports to table handlers that we stop replication to a specific slave
+
+  SYNOPSIS
+    ha_repl_report_replication_stop()
+
+  NOTES
+    Does nothing at the moment
+
+  RETURN VALUE
+    Always 0 (= success)  
+
+  PARAMETERS
+    THD    *thd             in: thread doing the binlog communication to
+                                the slave
+*/
+
+int ha_repl_report_replication_stop(THD *thd)
+{
+  thd = thd;
+
+  return 0;
+}

sql/handler.h

@@ -843,7 +843,7 @@ int ha_change_key_cache_param(KEY_CACHE *key_cache);
 int ha_change_key_cache(KEY_CACHE *old_key_cache, KEY_CACHE *new_key_cache);
 int ha_end_key_cache(KEY_CACHE *key_cache);
 
-/* weird stuff */
+/* report to InnoDB that control passes to the client */
 int ha_release_temporary_latches(THD *thd);
 
 /* transactions: interface to handlerton functions */
@@ -875,3 +875,7 @@ void trans_register_ha(THD *thd, bool all, handlerton *ht);
 #define trans_need_2pc(thd, all)                   ((total_ha_2pc > 1) && \
         !((all ? &thd->transaction.all : &thd->transaction.stmt)->no_2pc))
 
+/* semi-synchronous replication */
+int ha_repl_report_sent_binlog(THD *thd, char *log_file_name,
+                               my_off_t end_offset);
+int ha_repl_report_replication_stop(THD *thd);

sql/mysqld.cc

@@ -5495,7 +5495,6 @@ The minimum value for this variable is 4096.",
   {"sync-frm", OPT_SYNC_FRM, "Sync .frm to disk on create. Enabled by default.",
    (gptr*) &opt_sync_frm, (gptr*) &opt_sync_frm, 0, GET_BOOL, NO_ARG, 1, 0,
    0, 0, 0, 0},
-#ifdef DOES_NOTHING_YET
   {"sync-replication", OPT_SYNC_REPLICATION,
    "Enable synchronous replication.",
    (gptr*) &global_system_variables.sync_replication,
@@ -5511,7 +5510,6 @@ The minimum value for this variable is 4096.",
    (gptr*) &global_system_variables.sync_replication_timeout,
    (gptr*) &global_system_variables.sync_replication_timeout,
    0, GET_ULONG, REQUIRED_ARG, 10, 0, ~0L, 0, 1, 0},
-#endif
   {"table_cache", OPT_TABLE_CACHE,
    "The number of open tables for all threads.", (gptr*) &table_cache_size,
    (gptr*) &table_cache_size, 0, GET_ULONG, REQUIRED_ARG, 64, 1, 512*1024L,

sql/set_var.cc

@@ -957,11 +957,9 @@ struct show_var_st init_vars[]= {
   {"sql_warnings",            (char*) &sys_sql_warnings,            SHOW_BOOL},
 #ifdef HAVE_REPLICATION
   {sys_sync_binlog_period.name,(char*) &sys_sync_binlog_period,     SHOW_SYS},
-#ifdef DOES_NOTHING_YET
   {sys_sync_replication.name, (char*) &sys_sync_replication,        SHOW_SYS},
   {sys_sync_replication_slave_id.name, (char*) &sys_sync_replication_slave_id,SHOW_SYS},
   {sys_sync_replication_timeout.name, (char*) &sys_sync_replication_timeout,SHOW_SYS},
-#endif
 #endif
   {sys_sync_frm.name,         (char*) &sys_sync_frm,               SHOW_SYS},
 #ifdef HAVE_TZNAME

sql/sql_repl.cc

@@ -385,6 +385,11 @@ impossible position";
     goto err;
   }
 
+  printf("Binlog file name %s\n", log_file_name);
+
+  if (thd->variables.sync_replication)
+    ha_repl_report_sent_binlog(thd, log_file_name, pos);
+
   /*
     We need to start a packet with something other than 255
     to distinguish it from error
@@ -470,6 +475,10 @@ impossible position";
            my_errno= ER_UNKNOWN_ERROR;
            goto err;
          }
+
+         if (thd->variables.sync_replication)
+           ha_repl_report_sent_binlog(thd, log_file_name, my_b_tell(&log));
+
          /*
            No need to save this event. We are only doing simple reads
            (no real parsing of the events) so we don't need it. And so
@@ -527,6 +536,13 @@ impossible position";
 	my_errno= ER_UNKNOWN_ERROR;
 	goto err;
       }
+
+      printf("Dump loop: %s: Current log position %lu\n", log_file_name,
+                  (ulong)my_b_tell(&log));
+
+      if (thd->variables.sync_replication)
+        ha_repl_report_sent_binlog(thd, log_file_name, my_b_tell(&log));
+
       DBUG_PRINT("info", ("log event code %d",
 			  (*packet)[LOG_EVENT_OFFSET+1] ));
       if ((*packet)[LOG_EVENT_OFFSET+1] == LOAD_EVENT)
@@ -640,6 +656,12 @@ impossible position";
 	    goto err;
 	  }
 
+          printf("Second loop: %s: Current log position %lu\n", log_file_name,
+                  (ulong)my_b_tell(&log));
+
+          if (thd->variables.sync_replication)
+            ha_repl_report_sent_binlog(thd, log_file_name, my_b_tell(&log));
+
 	  if ((*packet)[LOG_EVENT_OFFSET+1] == LOAD_EVENT)
 	  {
 	    if (send_file(thd))
@@ -704,12 +726,22 @@ impossible position";
 	my_errno= ER_MASTER_FATAL_ERROR_READING_BINLOG;
 	goto err;
       }
+
+      if (thd->variables.sync_replication)
+        ha_repl_report_sent_binlog(thd, log_file_name, 0);
+
+      printf("Binlog file name of a new binlog %s\n", log_file_name);
+
       packet->length(0);
       packet->append('\0');
     }
   }
 
 end:
+  printf("Ending replication\n");
+  if (thd->variables.sync_replication)
+    ha_repl_report_replication_stop(thd);
+
   end_io_cache(&log);
   (void)my_close(file, MYF(MY_WME));
 
@@ -721,6 +753,11 @@ impossible position";
   DBUG_VOID_RETURN;
 
 err:
+  if (thd->variables.sync_replication)
+    ha_repl_report_replication_stop(thd);
+
+  printf("Ending replication in error %s\n", errmsg);
+
   thd->proc_info = "Waiting to finalize termination";
   end_io_cache(&log);
   /*