Page MenuHomePhabricator

Implement GTID-MySQL support for mediawiki
Open, LowPublic

Description

T135027: MASTER_POS_WAIT() alternative that works cross-DC implemented cross-dc support for MariaDB. It would be important to have the same kind of support for MySQL 5.6/5.7 GTID implementation. While MySQL it not currently used on production, this would be a blocker for it, and a partial or full migration to MySQL in the future is not discarded.

Mediawiki may require 5.6/5.7 in the future (T161232: Bump MediaWiki's minimum supported MySQL Version to 5.5.8), and this would be how it supports those versions, together with T108255: Enable MariaDB/MySQL's Strict Mode.

This task should be relatively trivial to implement (GTID on mysql and mariadb work very similarly from the programmer's perspective), the complex part is testing it.

Event Timeline

jcrespo created this task.Apr 3 2017, 3:07 PM
Restricted Application added a subscriber: Aklapper. · View Herald TranscriptApr 3 2017, 3:07 PM
Paladox added a subscriber: Paladox.Apr 3 2017, 3:11 PM
Krinkle triaged this task as Low priority.Jul 23 2019, 5:29 PM
Krinkle added a subscriber: Krinkle.

As I understand it, this task tracks the feature request of wikimedia/rdbms supporting replication lag detection for MySQL's GTID feature.

It currently has support for MariaDB's GTID feature already, which is what we use at WMF.

Restricted Application added a project: Platform Engineering. · View Herald TranscriptJul 23 2019, 5:29 PM
WDoranWMF added a subscriber: WDoranWMF.

@Krinkle Would this tasks live more logically within Performance than CPT?

This is not currently something needed by production, and not needed for any current goals or roadmap of perf team. We also don't have prior expertise or experience with MySQL's GTID specifically (as far as I know) that would make this easier for us implement.

It is an idea for later that CPT might to implement as part of general low-prio maintenance, or if it ends up requested by third parties in a way that CPT wants to support, perhaps with higher prio.

Unless that's the case, I suppose this can just sit in the backlog for now.