> Hm, so the "fix" is just to raise the minimum value for innodb_dict_size to
> 100M to lower probability of a race condition. But people in the bug report
> say the crash is reproducible with 128M and 100M. So either we are fixing a
> different problem (i.e. the one found by dbpq and the original one are
> different), or 100M is not sufficient to avoid the crash?
There's a couple of fixes: some of which is the original Yasufumi fix and the next part is setting a minimum size (which is what we were hitting for some of the dbqp test).
> - percona_innodb_dict_size_basic.reject was added by mistake?
fixed.
> - it looks like -master.opt don't really need --innodb. But need new
> lines :)
fixed.
> - I think the warning about innodb_dict_size_limit should be prefixed
> with "InnoDB:" or "InnoDB: Warning: ", just for consistency with
> other messages
> Hm, so the "fix" is just to raise the minimum value for innodb_dict_size to
> 100M to lower probability of a race condition. But people in the bug report
> say the crash is reproducible with 128M and 100M. So either we are fixing a
> different problem (i.e. the one found by dbpq and the original one are
> different), or 100M is not sufficient to avoid the crash?
There's a couple of fixes: some of which is the original Yasufumi fix and the next part is setting a minimum size (which is what we were hitting for some of the dbqp test).
> - percona_ innodb_ dict_size_ basic.reject was added by mistake?
fixed.
> - it looks like -master.opt don't really need --innodb. But need new
> lines :)
fixed.
> - I think the warning about innodb_ dict_size_ limit should be prefixed
> with "InnoDB:" or "InnoDB: Warning: ", just for consistency with
> other messages
fixed.