pt-table-checksum ignores --lock-wait-timeout

[Jaime Sicam]

Line 5940-5949:

$sql = 'SHOW SESSION VARIABLES LIKE "innodb_lock_wait_timeout"';
      PTDEBUG && _d($dbh, $sql);
      my (undef, $lock_wait_timeout) = $dbh->selectrow_array($sql);
      PTDEBUG && _d('innodb_lock_wait_timeout', $lock_wait_timeout);
      if ( ($lock_wait_timeout || 0) > $o->get('lock-wait-timeout') ) {
         $sql = 'SET SESSION innodb_lock_wait_timeout=1';
         eval {
            PTDEBUG && _d($dbh, $sql);
            $dbh->do($sql);
         };

So if you set, --lock-wait-timeout to a value less than the current innodb_lock_wait_timeout, then the session variable innodb_lock_wait_timeout will be set to 1

If you want to use the existing innodb_lock_wait_timeout value, you need to set --lock-wait-timeout to value that is equal or greater than the existing innodb_lock_wait_timeout

[Baron Schwartz (baron-xaprb)]

This code is very surprising to me. It should set the value of lock_wait_timeout to the value of the option regardless of the option's value, instead of using a hard-coded constant. It should also not do any greater-than comparison before the attempt to set the value. Instead, the greater-than comparison should be done in the error-handling code in case the attempt to set the value causes an error. The error should be displayed only if the value fails to be set and as a result is greater than what the user specified.

Furthermore, we need to get rid of the double-quotes in the SQL.

Finally, we need to document the error-if-failure-and-timeout-is-undesirably-large behavior.

[Baron Schwartz (baron-xaprb)]

Daniel, this is serious enough (and I think simple enough to fix) that I'm targeting it to the current release if you can fit it in. Otherwise you can retarget.

[Daniel Nichter (daniel-nichter)]

This code and the hard-coded value 1 are related to:

17a: The tool should set innodb_lock_wait_timeout=1 automatically to avoid blocking other queries if it is blocked itself. (https://blueprints.launchpad.net/percona-toolkit/+spec/redesign-pt-table-checkum)

Iirc, the original ideas was to always set lock_wait_timeout=1 regardless of the server, but I snuck in the --lock-wait-timeout option and this code to make it configurable primarily because this var is only dynamic with the InnoDB plugin and I needed to change it for testing (to avoid warnings), but it looks like I implemented the option too specifically for my own needs. Therefore, yes, I'll fix it to make it sane for the user.

[Shahriyar Rzayev (rzayev-sehriyar)]

Percona now uses JIRA for bug reports so this bug report is migrated to: https://jira.percona.com/browse/PT-493