Re: bzr commit into mysql-6.0 branch (mats:2674) Bug#37939, Bug#37941, Bug#37940

[Mats Kindahl]

Sven Sandberg wrote:
> Mats Kindahl wrote:
>> #At file:///home/bzr/fix-mysql-6.0-rpl/
>>
>>  2674 Mats Kindahl    2008-07-08
>>       Bug #37939      rpl_row_stop_middle_update fails sporadically on
>> pushbuild
>>       Bug #37940      rpl_dual_pos_advance fails sporadically on
>> pushbuild,fail wait_for_slave_to_stop
>>       Bug #37941      rpl_flushlog_loop fails sporadically on pushbuild
>>             Several tests fail when waiting for the slave to stop in what
>>       appears to be timeouts caused by a timeout value set to low.
>>       This causes false failures when the servers are loaded.
>>             In order to try to avoid false negatives, we increase the
>>       timeout 10 times and also print some more information in the
>>       event that the slave fails to stop when expected to.
>>             We add a printout of the current processes running to be able
>>       to see if any process have been executing for an unexpectedly
>>       long time, and also print the binlog events at the position
>>       indicated by SHOW SLAVE STATUS.
>> modified:
>>   mysql-test/include/wait_for_slave_sql_to_stop.inc
>>
>> per-file messages:
>>   mysql-test/include/wait_for_slave_sql_to_stop.inc
>>     Increase the default timeout to 3000 and allow the timeout counter
>> to be set.
>>     On failure, print process list and also the events on the master
>> following the
>>     binary log position given by the show slave status, unless
>> keep_connection is
>>     set, in which case we just print the process list and the slave
>> status.
>> === modified file 'mysql-test/include/wait_for_slave_sql_to_stop.inc'
>> --- a/mysql-test/include/wait_for_slave_sql_to_stop.inc    2008-02-27
>> 17:46:06 +0000
>> +++ b/mysql-test/include/wait_for_slave_sql_to_stop.inc    2008-07-08
>> 13:36:55 +0000
>> @@ -8,6 +8,10 @@
>>  #      2) loop through looking for
>>  #         sql threads to stop
>>  #      3) If loops too long die.
>> +#
>> +# INPUT:
>> +#    timeout_counter    Number of 1/10 sec to wait
>> +#    keep_connectio     Do not do any "connect"
> 
> typo: keep_connection with n at the end.
> maybe include the dollar signs?

OK

> 
> btw, 1/10 of a second is a weird unit... could we use entire seconds in
> the interface? Buy it's a matter of taste of course....

No, the inc and dec functions operate on integers.

> 
>>  ####################################################
>>  if (!$keep_connection)
>>  {
>> @@ -15,19 +19,39 @@ if (!$keep_connection)
>>  }
>>  let $row_number= 1;
>>  let $run= 1;
>> -let $counter= 300;
>> +if (!$timeout_counter)
>> +{
>> +  let $timeout_counter= 3000;
>> +}
>>  
>>  while ($run)
>>  {
>> +  if (!$keep_connection)
>> +  {
>> +    if (!$timeout_counter)
>> +    {
>> +      connection slave;
> 
> connection slave should be unnecessary here?

Hmmm.... yes

> 
>> +      let $binlog_pos = query_get_value("SHOW SLAVE STATUS",
>> Exec_Master_Log_Pos, 1);
>> +      let $binlog_file = query_get_value("SHOW SLAVE STATUS",
>> Master_Log_File, 1);
>> +    }
>> +  }
>>    let $sql_result= query_get_value("SHOW SLAVE STATUS",
>> Slave_SQL_Running, $row_number);
>>    if (`SELECT '$sql_result' = 'No'`){
>>      let $run= 0;
>>    }
>>    sleep 0.1;
>> -  if (!$counter){
>> +  if (!$timeout_counter){
>>      --echo "Failed while waiting for slave SQL thread to stop"
>> +    if (!$keep_connection)
>> +    {
>> +      connection master;
> 
> I'd suggest adding:
> 
> --echo **** Note that the binlog positions are not read atomically, so
> there is a small risk that the binlog status is incorrect ****

OK.

> 
>> +      eval SHOW BINLOG EVENTS IN '$binlog_file' FROM $binlog_pos
>> LIMIT 5;
>> +      connection slave;
>> +    }
>> +    --echo **** Note that the process list might have changed since
>> the failure was detected **** +    SHOW PROCESSLIST;
>>      query_vertical SHOW SLAVE STATUS;
>>      exit;
>>    }
>> -  dec $counter;
>> +  dec $timeout_counter;
> 
> Better to leave $timeout_counter unmodified to the caller: please copy
> it to $_timeout_caller before using it in this file.

Good point, will do.

> 
>>  }
>>
>>
> 


-- 
Mats Kindahl
Lead Software Developer
Replication Team
MySQL AB, www.mysql.com