Page MenuHomePhabricator

Exceptional SQL time-out limits for volunteers doing special analysis
Closed, InvalidPublic

Description

The current default time-out for SQL queries on WMF-Labs is 30 minutes. For any analysis or research that involves the logging table, it is quite normal for even basic queries to exceed the time-out limit, leading to various reports being broken up into multiple sliced queries and re-assembled. This work-around may increase the operational demand.

I propose that volunteers with a track record of report writing and special analysis to support their wiki communities are allowed an exception to their relevant labs accounts on request. A longer time-out, such as 2 hours, and a request that regular bot-based reports aim to stay within the 30 minutes may resolve this frequently encountered issue.

Event Timeline

This is not set to 30 minutes- I remember last time we had load issues (mostly memory issues crashing the server continuously) with labsdb replicas we lowered it to 2 hours, and we aim to increase it with the new analytics service (See https://www.mediawiki.org/wiki/Wikimedia_Engineering/2016-17_Q2_Goals#Technical_Operations, "Provide at least 2 separate service endpoints: one for slow, long running queries; and another for quick, web requests").

Maybe your user is having specific issues or your client is killing things with that timeout? I would encourage to report that if that is the case and we can investigate.

@Steinsplitter we will ask for beta testers soon, please stay tuned.

I am closing this as invalid, as I didn't have more feedback from the original reporter. Please reopen a new one with something like "My tool has db issues after 30 minutes", *if* that is the case, but the limitations indicated are not there and certainly not on purpose, and we aim to increase them anyways in the future, as explained above.