maria-developers team mailing list archive

[Michael Widenius <michael.widenius@xxxxxxxxx>]

Hi!

On Thu, Aug 18, 2016 at 11:03 PM, Kristian Nielsen
<knielsen@xxxxxxxxxxxxxxx> wrote:
> Michael Widenius <michael.widenius@xxxxxxxxx> writes:
>
>> Do you happen to know of any failures caused by the patch?
>
> As I mentioned on IRC, the assertion triggers during rpl_mdev6020, for
> example here:
>
>   https://buildbot.askmonty.org/buildbot/builders/work-amd64-valgrind/builds/9199/steps/test/logs/stdio
>
> As we discussed, one possible reason is a race because did_mark_start_commit
> is set only after waking up the next thread.

Just a note about rpl_mdev6020

This test started to fail with a timeout, as some parallel threads are
waiting on each-other, between MariaDB 10.0.17 and MariaDB 10.0.18

To get it to fail, one must run with valgrind:
mysql-test-run --valgrind rpl_mdev6020,innodb_plugin,mix

Now working on trying to see what changes caused it to fail.

Regards,
Monty