Page MenuHomeMiraheze
Create Task
Maniphest T6006

Migrate to a Kafka Job Queue
Open, LowPublic
Actions

Description

Setting as high due to the recent JobQueue incidents, feel free to re-triage.

See https://wikitech.wikimedia.org/wiki/Kafka_Job_Queue

In addition to giving more resources (T5994), we should consider migrating to the new Kafka system.

It includes de-duplication and concurreny limiting which should in theory allow faster and more balanced processing of jobs due to a more managed load on the system.

I'm still thinking we should add more resources though as while this will ease & manage the pressure in theory by reducing duplicate jobs and processing things in a more balanced way. We still had a very large backlog and are expanding quickly so I still think more resources would be noticeable.

Related Objects

Event Timeline

RhinosF1 triaged this task as High priority.Aug 2 2020, 07:55
RhinosF1 created this task.
Herald added subscribers: Void, Reception123, NDKilla. · View Herald TranscriptAug 2 2020, 07:55
RhinosF1 moved this task from Radar to New Features/Services on the Site Reliability Engineering board.Aug 2 2020, 07:55
RhinosF1 mentioned this in T5994: Upgrade jobrunner capabilities .
John lowered the priority of this task from High to Low.Aug 2 2020, 08:07
John added a subscriber: John.

Would not solve problems, and would require more resources to implement - potentially more than we are able to suitably give.

RhinosF1 added a comment.Aug 2 2020, 08:15

Would not solve problems

There's 100% duplication in the way jobs are processed. If the duplication detections works as I understand, it would reduce the load as less jobs to process. Concurrency limiting would also reduce the amount of resources a single type of job can take up and therefore reduce the impact of a spike in edits on loginwiki.

I also note that I doubt the old system will be maintained or supported forever and that the jobrunner services repo hasn't had a single commit since 2017

John added a comment.Aug 2 2020, 08:26

“ would require more resources to implement - potentially more than we are able to suitably give.” is a vital thing you did not reply to, the more critical part.

RhinosF1 added a comment.EditedAug 2 2020, 08:44
In T6006#117420, @John wrote:

“ would require more resources to implement - potentially more than we are able to suitably give.” is a vital thing you did not reply to, the more critical part.

That's something I'm going to look into more as I think we should seriously consider it.

RobLa added a subscriber: RobLa.Jan 13 2021, 08:06
Herald added a subscriber: Universal_Omega. · View Herald TranscriptJan 13 2021, 08:06
John edited projects, added Infrastructure (SRE); removed Site Reliability Engineering.Jan 30 2021, 11:45
Herald added a project: MediaWiki (SRE). · View Herald TranscriptJan 30 2021, 11:45
Reception123 removed a project: Infrastructure (SRE).Jan 30 2021, 12:04