foo2 (3eb78381) · Commits · Software / OSDI20 Artifacts / mariadb

sql/ha_innodb.cc

+1 −230

Original line number	Diff line number	Diff line
		@@ -1741,25 +1741,6 @@ innobase_report_binlog_offset_and_commit(
		trx->mysql_log_file_name = log_file_name;
		trx->mysql_log_offset = (ib_longlong)end_offset;

		#ifdef HAVE_REPLICATION
		if (thd->variables.sync_replication) {
		/* Let us store the binlog file name and the position, so that
		we know how long to wait for the binlog to the replicated to
		the slave in synchronous replication. */

		if (trx->repl_wait_binlog_name == NULL) {

		trx->repl_wait_binlog_name =
		(char*)mem_alloc_noninline(FN_REFLEN + 100);
		}

		ut_a(strlen(log_file_name) < FN_REFLEN + 100);

		strcpy(trx->repl_wait_binlog_name, log_file_name);

		trx->repl_wait_binlog_pos = (ib_longlong)end_offset;
		}
		#endif /* HAVE_REPLICATION */
		trx->flush_log_later = TRUE;

		innobase_commit(thd, TRUE);
		@@ -1828,219 +1809,9 @@ innobase_commit_complete(
		trx_commit_complete_for_mysql(trx);
		}

		#ifdef HAVE_REPLICATION
		if (thd->variables.sync_replication
		&& trx->repl_wait_binlog_name
		&& innobase_repl_state != 0) {

		struct timespec abstime;
		int cmp;
		int ret;

		/* In synchronous replication, let us wait until the MySQL
		replication has sent the relevant binlog segment to the
		replication slave. */

		pthread_mutex_lock(&innobase_repl_cond_mutex);
		try_again:
		if (innobase_repl_state == 0) {

		pthread_mutex_unlock(&innobase_repl_cond_mutex);

		return(0);
		}

		cmp = strcmp(innobase_repl_file_name,
		trx->repl_wait_binlog_name);
		if (cmp > 0
		\|\| (cmp == 0 && innobase_repl_pos
		>= (my_off_t)trx->repl_wait_binlog_pos)) {
		/* We have already sent the relevant binlog to the
		slave: no need to wait here */

		pthread_mutex_unlock(&innobase_repl_cond_mutex);

		/* printf("Binlog now sent\n"); */

		return(0);
		}

		/* Let us update the info about the minimum binlog position
		of waiting threads in the innobase_repl_... variables */

		if (innobase_repl_wait_file_name_inited != 0) {
		cmp = strcmp(trx->repl_wait_binlog_name,
		innobase_repl_wait_file_name);
		if (cmp < 0
		\|\| (cmp == 0 && (my_off_t)trx->repl_wait_binlog_pos
		<= innobase_repl_wait_pos)) {
		/* This thd has an even lower position, let
		us update the minimum info */

		strcpy(innobase_repl_wait_file_name,
		trx->repl_wait_binlog_name);

		innobase_repl_wait_pos =
		trx->repl_wait_binlog_pos;
		}
		} else {
		strcpy(innobase_repl_wait_file_name,
		trx->repl_wait_binlog_name);

		innobase_repl_wait_pos = trx->repl_wait_binlog_pos;

		innobase_repl_wait_file_name_inited = 1;
		}
		set_timespec(abstime, thd->variables.sync_replication_timeout);

		/* Let us suspend this thread to wait on the condition;
		when replication has progressed far enough, we will release
		these waiting threads. The following call
		pthread_cond_timedwait also atomically unlocks
		innobase_repl_cond_mutex. */

		innobase_repl_n_wait_threads++;

		/* printf("Waiting for binlog to be sent\n"); */

		ret = pthread_cond_timedwait(&innobase_repl_cond,
		&innobase_repl_cond_mutex, &abstime);
		innobase_repl_n_wait_threads--;

		if (ret != 0) {
		ut_print_timestamp(stderr);

		sql_print_error("MySQL synchronous replication was "
		"not able to send the binlog to the "
		"slave within the timeout %lu. We "
		"assume that the slave has become "
		"inaccessible, and switch off "
		"synchronous replication until the "
		"communication to the slave works "
		"again. MySQL synchronous replication "
		"has sent binlog to the slave up to "
		"file %s, position %lu. This "
		"transaction needs it to be sent up "
		"to file %s, position %lu.",
		thd->variables.sync_replication_timeout,
		innobase_repl_file_name,
		(ulong) innobase_repl_pos,
		trx->repl_wait_binlog_name,
		(ulong) trx->repl_wait_binlog_pos);

		innobase_repl_state = 0;

		pthread_mutex_unlock(&innobase_repl_cond_mutex);

		return(0);
		}

		goto try_again;
		}
		#endif // HAVE_REPLICATION
		return(0);
		}

		#ifdef HAVE_REPLICATION
		/*********************************************************************
		In synchronous replication, reports to InnoDB up to which binlog position
		we have sent the binlog to the slave. Note that replication is synchronous
		for one slave only. For other slaves, we do nothing in this function. This
		function is used in a replication master. */

		int
		innobase_repl_report_sent_binlog(
		/=============================/
		/* out: 0 */
		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 cmp;
		ibool can_release_threads = 0;

		/* If synchronous replication is not switched on, or this thd is
		sending binlog to a slave where we do not need synchronous replication,
		then return immediately */

		if (thd->server_id != thd->variables.sync_replication_slave_id) {

		/* Do nothing */

		return(0);
		}

		pthread_mutex_lock(&innobase_repl_cond_mutex);

		if (innobase_repl_state == 0) {

		ut_print_timestamp(stderr);
		sql_print_warning("Switching MySQL synchronous replication on "
		"again at binlog file %s, position %lu",
		log_file_name, (ulong) end_offset);

		innobase_repl_state = 1;
		}

		/* The position should increase monotonically, since just one thread
		is sending the binlog to the slave for which we want synchronous
		replication. Let us check this, and print an error to the .err log
		if that is not the case. */

		if (innobase_repl_file_name_inited) {
		cmp = strcmp(log_file_name, innobase_repl_file_name);

		if (cmp < 0
		\|\| (cmp == 0 && end_offset < innobase_repl_pos)) {

		ut_print_timestamp(stderr);
		sql_print_error("MySQL synchronous replication has "
		"sent binlog to the slave up to file "
		"%s, position %lu, but now MySQL "
		"reports that it sent the binlog only "
		"up to file %s, position %lu",
		innobase_repl_file_name,
		(ulong) innobase_repl_pos,
		log_file_name, (ulong) end_offset);
		}
		}

		strcpy(innobase_repl_file_name, log_file_name);
		innobase_repl_pos = end_offset;
		innobase_repl_file_name_inited = 1;

		if (innobase_repl_n_wait_threads > 0) {
		/* Let us check if some of the waiting threads doing a trx
		commit can now proceed */

		cmp = strcmp(innobase_repl_file_name,
		innobase_repl_wait_file_name);
		if (cmp > 0
		\|\| (cmp == 0 && innobase_repl_pos
		>= innobase_repl_wait_pos)) {

		/* Yes, at least one waiting thread can now proceed:
		let us release all waiting threads with a broadcast */

		can_release_threads = 1;

		innobase_repl_wait_file_name_inited = 0;
		}
		}

		pthread_mutex_unlock(&innobase_repl_cond_mutex);

		if (can_release_threads) {

		pthread_cond_broadcast(&innobase_repl_cond);
		}

		return(0);
		}
		#endif /* HAVE_REPLICATION */

		/*********************************************************************
		Rolls back a transaction or the latest SQL statement. */

sql/ha_innodb.h

+0 −3

Original line number	Diff line number	Diff line
		@@ -303,9 +303,6 @@ int innobase_rollback_by_xid(
		XID xid); / in : X/Open XA Transaction Identification */


		int innobase_repl_report_sent_binlog(THD thd, char log_file_name,
		my_off_t end_offset);

		/***********************************************************************
		Create a consistent view for a cursor based on current transaction
		which is created if the corresponding MySQL thread still lacks one.

sql/handler.cc

+0 −53

Original line number	Diff line number	Diff line
		@@ -2684,56 +2684,3 @@ TYPELIB *ha_known_exts(void)
		}
		return &known_extensions;
		}


		#ifdef HAVE_REPLICATION
		/*
		Reports to table handlers up to which position we have sent the binlog
		to a slave in replication

		SYNOPSIS
		ha_repl_report_sent_binlog()
		thd thread doing the binlog communication to the slave
		log_file_name binlog file name
		end_offse t the offset in the binlog file up to which we sent the
		contents to the slave

		NOTES
		Only works for InnoDB at the moment

		RETURN VALUE
		Always 0 (= success)
		*/

		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
		return 0;
		#endif
		}


		/*
		Reports to table handlers that we stop replication to a specific slave

		SYNOPSIS
		ha_repl_report_replication_stop()
		thd thread doing the binlog communication to the slave

		NOTES
		Does nothing at the moment

		RETURN VALUE
		Always 0 (= success)

		PARAMETERS
		*/

		int ha_repl_report_replication_stop(THD *thd)
		{
		return 0;
		}
		#endif /* HAVE_REPLICATION */

sql/handler.h

+0 −5

Original line number	Diff line number	Diff line
		@@ -965,8 +965,3 @@ 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

+0 −17

Original line number	Diff line number	Diff line
		@@ -5899,23 +5899,6 @@ 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 HAVE_REPLICATION
		{"sync-replication", OPT_SYNC_REPLICATION,
		"Enable synchronous replication.",
		(gptr*) &global_system_variables.sync_replication,
		(gptr*) &global_system_variables.sync_replication,
		0, GET_ULONG, REQUIRED_ARG, 0, 0, 1, 0, 1, 0},
		{"sync-replication-slave-id", OPT_SYNC_REPLICATION_SLAVE_ID,
		"Synchronous replication is wished for this slave.",
		(gptr*) &global_system_variables.sync_replication_slave_id,
		(gptr*) &global_system_variables.sync_replication_slave_id,
		0, GET_ULONG, REQUIRED_ARG, 0, 0, ~0L, 0, 1, 0},
		{"sync-replication-timeout", OPT_SYNC_REPLICATION_TIMEOUT,
		"Synchronous replication timeout.",
		(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 /* HAVE_REPLICATION */
		{"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, 5121024L,