In T349442 explicit DB access of the math extension was removed. After this patch (scheduled for 1.42.0-wmf.14) has been deployed everywhere the tables should be deleted in production. I expect that the table mathlatexml does not exist and that mathoid is empty or does not exist as well.
Progress
| Status | Subtype | Assigned | Task | ||
|---|---|---|---|---|---|
| Resolved | Physikerwelt | T349442 Remove explicit DB access for extension Math | |||
| Resolved | Marostegui | T355050 SQL: DROP TABLE IF EXISTS mathoid, mathlatexml; |
I have dropped these two from labtestwiki (they were empty) and renamed it for now on s1 (db1186):
root@db1186.eqiad.wmnet[enwiki]> show tables like 'T%'; +-----------------------+ | Tables_in_enwiki (T%) | +-----------------------+ | T355050_mathoid | +-----------------------+ 1 row in set (0.001 sec)
The patch hasn't reach production yet. Causing ~200 errors an hour: https://logstash.wikimedia.org/app/discover#/doc/logstash-*/logstash-deploy-1-7.0.0-1-2024.01.16?id=xD33E40BySCoT0gd0Wgb
https://logstash.wikimedia.org/goto/b7ed5371a9dc7bc3a5110ce00f59da02
I'll revert the s1 change. We can retry next week.
The patch needs to go with train, it's in wmf.14, enwiki is on wmf.13 (https://en.wikipedia.org/wiki/Special:Version)
Probably what we should do is amend the doc for dropping tables and state that the tickets should be created once the changes are fully deployed.
I'll do that!
I just added this https://wikitech.wikimedia.org/w/index.php?title=Schema_changes&diff=2139957&oldid=2094041
@Physikerwelt can you please unstall this ticket once the ticket has been deployed? Thank you!
Ok, I think the change goes with the train. So I will unstall this tomorrow https://wikitech.wikimedia.org/wiki/Deployments#deploycal-item-20240118T0900, correct?
Thank you. From a process perspective, I think it would be advisable to add another level of security and instruct the DBA team to double-check the train status. Here the tag MW-1.42-notes (1.42.0-wmf.14; 2024-01-16) is a bit misleading. If the submitter error probability is at 50% (Unfortunately, the manual compliance rate from non-wmf-people like me is quite low) and the DBA error probability is at 10%. The overall error probability could be reduced to 5%.
We usually give it the full week in case train gets rolled back or blocked or other issues.
It is still unclear to me if the new cache is used T355673. At least the 200 hits we saw before should end up in BagOStuff cache requests.
I found one related issue https://gerrit.wikimedia.org/r/c/mediawiki/extensions/Math/+/993223 but other than that I think we could try it again.
Done again on db1186:
root@db1186.eqiad.wmnet[enwiki]> show tables like '%math%'; +---------------------------+ | Tables_in_enwiki (%math%) | +---------------------------+ | T355050_mathoid | +---------------------------+ 1 row in set (0.001 sec)
Mentioned in SAL (#wikimedia-operations) [2024-02-05T06:11:25Z] <marostegui> Drop mathoid, mathlatexml tables T355050
Now going for s3, this will take several hours as there are almost 1000 tables there and I've left a sleep between deletions to avoid replication lag.
Mentioned in SAL (#wikimedia-operations) [2024-02-05T06:56:12Z] <marostegui> dbamaint Drop mathoid, mathlatexml tables T355050