Hi Libing,
Nice work, I have some comments.
Li-Bing.Song@stripped wrote:
> #At file:///home/anders/work/bzrwork/worktree5/mysql-6.0-codebase-bugfixing/ based on
> revid:alik@ibmvm-20091008093643-fjns7fky4mc00jfw
>
> 3649 Li-Bing.Song@stripped 2009-10-26
> Bug#47699 rpl.rpl_backup_block fails sporadically
>
> After 'START SLAVE' is executed, An unexpected error 'RESTORE_ON_MASTER'
> appears in Last_Error. On Slave, The Last_Error is cleared after
> 'START SLAVE [SQL_THREAD]' has returned. 'SHOW SLAVE STATUS' sometimes is
> executed earlier than that Last_Error is cleared. So the last error which
> has happened before 'START SLAVE' still appears in the slave status.
>
> This patch wrote code to clear error before 'START SLAVE [SQL_THREAD]'
> returns.
>
> modified:
> sql/slave.cc
> === modified file 'sql/slave.cc'
> --- a/sql/slave.cc 2009-10-01 23:15:18 +0000
> +++ b/sql/slave.cc 2009-10-26 06:00:33 +0000
> @@ -3162,6 +3162,18 @@ pthread_handler_t handle_slave_sql(void
> rli->slave_run_id++;
> rli->slave_running = 1;
>
> + /*
> + Reset errors for a clean start (otherwise, if the master is idle, the SQL
> + thread may execute no Query_log_event, so the error will remain even
> + though there's no problem anymore). Do not reset the master timestamp
> + (imagine the slave has caught everything, the STOP SLAVE and START SLAVE:
> + as we are not sure that we are going to receive a query, we want to
> + remember the last master timestamp (to say how many seconds behind we are
> + now.
> + But the master timestamp is reset by RESET SLAVE & CHANGE MASTER.
> + */
> + rli->clear_error();
> +
> pthread_detach_this_thread();
> if (init_slave_thread(thd, SLAVE_THD_SQL))
> {
> @@ -3193,18 +3205,6 @@ pthread_handler_t handle_slave_sql(void
> pthread_mutex_unlock(&rli->run_lock);
> pthread_cond_broadcast(&rli->start_cond);
I think you only need move the block below before the above two
statements.
>
> - /*
> - Reset errors for a clean start (otherwise, if the master is idle, the SQL
> - thread may execute no Query_log_event, so the error will remain even
> - though there's no problem anymore). Do not reset the master timestamp
> - (imagine the slave has caught everything, the STOP SLAVE and START SLAVE:
> - as we are not sure that we are going to receive a query, we want to
> - remember the last master timestamp (to say how many seconds behind we are
> - now.
> - But the master timestamp is reset by RESET SLAVE & CHANGE MASTER.
> - */
> - rli->clear_error();
> -
> //tell the I/O thread to take relay_log_space_limit into account from now on
> pthread_mutex_lock(&rli->log_space_lock);
> rli->ignore_log_space_limit= 0;
>
--
MySQL Code Commits Mailing List
For list archives: http://lists.mysql.com/commits
To unsubscribe: http://lists.mysql.com/commits?unsub=1
