WL#1595 "Optionally fsync() the binlog after every statement":

New option --sync-binlog=x (and global settable variable) which will fsync the binlog
after every x-th disk write to it. That is, if in autocommit mode, after every x-th statement
written to the binlog; if using transactions, after every x-th transaction written to the binlog.
x==0 means no fsync. x==1 is the slowest.
There is no test added for this, I have just checked that it works as --sync-binlog=1 dramatically
slows down mysqld.
Made sync-frm a global settable variable.

5 files changed, 58 insertions, 7 deletions

diff --git a/sql/log.cc b/sql/log.cc
index e7a142230b1..47a6a4a9b4c 100644
--- a/sql/log.cc
+++ b/sql/log.cc
@@ -31,6 +31,7 @@
 #include <m_ctype.h>				// For test_if_number
 
 MYSQL_LOG mysql_log,mysql_update_log,mysql_slow_log,mysql_bin_log;
+ulong sync_binlog_counter= 0;
 
 static bool test_if_number(const char *str,
 			   long *res, bool allow_wildcards);
@@ -1164,6 +1165,13 @@ bool MYSQL_LOG::write(THD *thd,enum enum_server_command command,
 }
 
 
+inline bool sync_binlog(IO_CACHE *cache)
+{
+  return (sync_binlog_period &&
+          (sync_binlog_period == ++sync_binlog_counter) &&
+          (sync_binlog_counter= 0, my_sync(cache->file, MYF(MY_WME))));
+}
+
 /*
   Write an event to the binary log
 */
@@ -1369,9 +1377,9 @@ COLLATION_CONNECTION=%lu,COLLATION_DATABASE=%lu,COLLATION_SERVER=%lu",
 
     if (file == &log_file) // we are writing to the real log (disk)
     {
-      if (flush_io_cache(file))
+      if (flush_io_cache(file) || sync_binlog(file))
 	goto err;
- 
+
       if (opt_using_transactions && !my_b_tell(&thd->transaction.trans_log))
       {
         /*
@@ -1529,7 +1537,8 @@ bool MYSQL_LOG::write(THD *thd, IO_CACHE *cache, bool commit_or_rollback)
                             commit_or_rollback ? 6        : 8, 
                             TRUE);
       qinfo.set_log_pos(this);
-      if (qinfo.write(&log_file) || flush_io_cache(&log_file))
+      if (qinfo.write(&log_file) || flush_io_cache(&log_file) ||
+          sync_binlog(&log_file))
 	goto err;
     }
     if (cache->error)				// Error on read
diff --git a/sql/mysql_priv.h b/sql/mysql_priv.h
index f9dd4ded94a..f0497925523 100644
--- a/sql/mysql_priv.h
+++ b/sql/mysql_priv.h
@@ -858,7 +858,7 @@ extern ulong max_binlog_size, max_relay_log_size;
 extern ulong rpl_recovery_rank, thread_cache_size;
 extern ulong com_stat[(uint) SQLCOM_END], com_other, back_log;
 extern ulong specialflag, current_pid;
-extern ulong expire_logs_days;
+extern ulong expire_logs_days, sync_binlog_period, sync_binlog_counter;
 extern my_bool relay_log_purge;
 extern uint test_flags,select_errors,ha_open_options;
 extern uint protocol_version, mysqld_port, dropping_tables;
diff --git a/sql/mysqld.cc b/sql/mysqld.cc
index 7f91466bb3d..4e9847e99cc 100644
--- a/sql/mysqld.cc
+++ b/sql/mysqld.cc
@@ -306,7 +306,7 @@ ulong binlog_cache_use= 0, binlog_cache_disk_use= 0;
 ulong max_connections,max_used_connections,
       max_connect_errors, max_user_connections = 0;
 ulong thread_id=1L,current_pid;
-ulong slow_launch_threads = 0;
+ulong slow_launch_threads = 0, sync_binlog_period;
 ulong expire_logs_days = 0;
 ulong rpl_recovery_rank=0;
 ulong my_bind_addr;			/* the address we bind to */
@@ -3884,7 +3884,7 @@ enum options_mysqld
   OPT_RANGE_ALLOC_BLOCK_SIZE,
   OPT_QUERY_ALLOC_BLOCK_SIZE, OPT_QUERY_PREALLOC_SIZE,
   OPT_TRANS_ALLOC_BLOCK_SIZE, OPT_TRANS_PREALLOC_SIZE,
-  OPT_SYNC_FRM, OPT_BDB_NOSYNC,
+  OPT_SYNC_FRM, OPT_SYNC_BINLOG, OPT_BDB_NOSYNC,
   OPT_ENABLE_SHARED_MEMORY,
   OPT_SHARED_MEMORY_BASE_NAME,
   OPT_OLD_PASSWORDS,
@@ -4866,6 +4866,12 @@ The minimum value for this variable is 4096.",
    (gptr*) &max_system_variables.sortbuff_size, 0, GET_ULONG, REQUIRED_ARG,
    MAX_SORT_MEMORY, MIN_SORT_MEMORY+MALLOC_OVERHEAD*2, ~0L, MALLOC_OVERHEAD,
    1, 0},
+  {"sync-binlog", OPT_SYNC_BINLOG,
+   "Sync the binlog to disk after every #th event. \
+#=0 (the default) does no sync. Syncing slows MySQL down",
+   (gptr*) &sync_binlog_period,
+   (gptr*) &sync_binlog_period, 0, GET_ULONG, REQUIRED_ARG, 0, 0, ~0L, 0, 1,
+   0},
   {"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,
diff --git a/sql/set_var.cc b/sql/set_var.cc
index 35bcbc9ce34..590b550ac3a 100644
--- a/sql/set_var.cc
+++ b/sql/set_var.cc
@@ -325,6 +325,10 @@ sys_var_thd_table_type  sys_table_type("table_type",
 				       &SV::table_type);
 sys_var_thd_storage_engine sys_storage_engine("storage_engine",
 				       &SV::table_type);
+#ifdef HAVE_REPLICATION
+sys_var_sync_binlog_period sys_sync_binlog_period("sync_binlog", &sync_binlog_period);
+#endif
+sys_var_bool_ptr	sys_sync_frm("sync_frm", &opt_sync_frm);
 sys_var_long_ptr	sys_table_cache_size("table_cache",
 					     &table_cache_size);
 sys_var_long_ptr	sys_thread_cache_size("thread_cache_size",
@@ -573,6 +577,10 @@ sys_var *sys_variables[]=
   &sys_sql_mode,
   &sys_sql_warnings,
   &sys_storage_engine,
+#ifdef HAVE_REPLICATION
+  &sys_sync_binlog_period,
+#endif
+  &sys_sync_frm,
   &sys_table_cache_size,
   &sys_table_type,
   &sys_thread_cache_size,
@@ -788,6 +796,10 @@ struct show_var_st init_vars[]= {
   {sys_sort_buffer.name,      (char*) &sys_sort_buffer, 	    SHOW_SYS},
   {sys_sql_mode.name,         (char*) &sys_sql_mode,                SHOW_SYS},
   {sys_storage_engine.name,   (char*) &sys_storage_engine,          SHOW_SYS},
+#ifdef HAVE_REPLICATION
+  {sys_sync_binlog_period.name,(char*) &sys_sync_binlog_period,     SHOW_SYS},
+#endif
+  {sys_sync_frm.name,         (char*) &sys_sync_frm,               SHOW_SYS},
   {"table_cache",             (char*) &table_cache_size,            SHOW_LONG},
   {sys_table_type.name,	      (char*) &sys_table_type,	            SHOW_SYS},
   {sys_thread_cache_size.name,(char*) &sys_thread_cache_size,       SHOW_SYS},
@@ -2309,6 +2321,22 @@ bool sys_var_slave_skip_counter::update(THD *thd, set_var *var)
   pthread_mutex_unlock(&LOCK_active_mi);
   return 0;
 }
+
+
+bool sys_var_sync_binlog_period::update(THD *thd, set_var *var)
+{
+  pthread_mutex_t *lock_log= mysql_bin_log.get_log_lock();
+  sync_binlog_period= var->save_result.ulong_value;
+  /*
+    Must reset the counter otherwise it may already be beyond the new period
+    and so the new period will not be taken into account. Need mutex otherwise
+    might be cancelled by a simultanate ++ in MYSQL_LOG::write().
+  */
+  pthread_mutex_lock(lock_log);
+  sync_binlog_counter= 0;
+  pthread_mutex_unlock(lock_log);
+  return 0;
+}
 #endif /* HAVE_REPLICATION */
 
 bool sys_var_rand_seed1::update(THD *thd, set_var *var)
diff --git a/sql/set_var.h b/sql/set_var.h
index 64bdfdb718b..9bed6f01dcc 100644
--- a/sql/set_var.h
+++ b/sql/set_var.h
@@ -462,7 +462,7 @@ public:
 };
 
 
-#ifndef EMBEDDED_LIBRARY
+#ifdef HAVE_REPLICATION
 class sys_var_slave_skip_counter :public sys_var
 {
 public:
@@ -475,6 +475,14 @@ public:
     type() or value_ptr()
   */
 };
+
+class sys_var_sync_binlog_period :public sys_var_long_ptr
+{
+public:
+  sys_var_sync_binlog_period(const char *name_arg, ulong *value_ptr)
+    :sys_var_long_ptr(name_arg,value_ptr) {}
+  bool update(THD *thd, set_var *var);
+};
 #endif
 
 class sys_var_rand_seed1 :public sys_var