23 September 2015, 13:54

Hi all,

I've put together a patch based on Russell's suggestion that the problem comes from the different default transaction isolation levels in MySQL & Postgres. MySQL defaults to "REPEATABLE READ", while Postgres defaults to the less stringent "READ COMMITTED". I think this likely explains why the problem is so much more prevalent in MySQL sites than in Postgres sites.

The patch for this tells Mahara to use the "READ COMMITTED" transaction isolation level across the site. This should result in the same behavior for MySQL & Postgres sites.

The patch is up on our gerrit site. Please give it a test and let me know if it solves your problems: https://reviews.mahara.org/#/c/5335/3

Cheers,

Aaron

10 November 2015, 17:07

Since we've received reports of people still experiencing this deadlock issue after the patch in 15.04.3 was applied, I've opened a follow-up bug: https://bugs.launchpad.net/mahara/+bug/1514608

18 November 2015, 16:20

Hi all,

I've developed another patch for this issue. It disables Mahara's usage of explicit transactions in MySQL, because the root of the problem seems to be that we have some operations that open up transactions that last too long.

Would anyone like to test it to see if it resolves the problem on their site? https://reviews.mahara.org/#/c/5723/

Cheers,

Aaron