maria-developers team mailing list archive

[Kristian Nielsen <knielsen@xxxxxxxxxxxxxxx>]

Jonas Oreland <jonaso@xxxxxxxxxx> writes:

> i addressed your comments and I uploaded 2 new patches to JIRA.
> 1) a new "complete patch"
> 2) a patch that is changes from v1 to v2.


> the problem was that I didn't update info->last_pos and linfo->pos
> correctly in send_format_descriptor_event
> so the offsets were wrong (in the old file)...how ever the result file
> still changes, since the old code was also incorrect :-)

> this also made me find a bug that when seeking i forgot to update linfo->pos

Ok, cool.

> ok, I readded the assert,
> and the code crashed on the slave side.
> I updated comment with stack trace,
> but left it at that.

> +   * there should/must be mysql_mutex_assert_owner(&LOCK_log) here...
> +   * but code violates this! (scary monsters and super creeps!)
> +   *
> +   * example stacktrace:
> +   * #8  MYSQL_BIN_LOG::is_active
> +   * #9  MYSQL_BIN_LOG::can_purge_log
> +   * #10 MYSQL_BIN_LOG::purge_logs
> +   * #11 MYSQL_BIN_LOG::purge_first_log
> +   * #12 next_event
> +   * #13 exec_relay_log_event

Ok, I see. Right, seems there is a potential issue between the slave IO thread
moving to a new relay log, and the SQL thread checking if it can purge an old
log. Thanks for checking.

At least the master side seems fixed now, with your patch.

> you take

Ok, I filed it as MDEV-7310 so it does not get lost.

> i think i'm fine either way, put it wherever you like.
> (at least currently I'm mainly interested in the review feedback)

Ok.

I checked the "changes from v1 to v2 diff", and it looks fine to me now.
I can apply it along with the other related patches.

I can also give it a spin through our Buildbot first to check for any issues.

 - Kristian.