This is a tracking task for all database errors and other issues related to large transactions or large writes. Large transactions causes issues because they generate lag as bad as the transaction size. To avoid that, a transaction size limit was setup, however, sometimes that limitation can cause issues- the limit should be maintained, but sometimes actions have to be split in smaller transactions or being executed in the background.
Description
Description
| Status | Subtype | Assigned | Task | ||
|---|---|---|---|---|---|
| Resolved | jcrespo | T207940 Large transaction-related errors and other problems (tracking) | |||
| Resolved | PRODUCTION ERROR | • kostajh | T171898 Deleting large watchlist takes > 4 seconds causing rollback due to write time limit | ||
| Resolved | PRODUCTION ERROR | BPirkle | T198176 Mediawiki page deletions should happen in batches of revisions | ||
| Resolved | PRODUCTION ERROR | Catrope | T207941 Spike of DBTransactionSizeError exceptions from /w/api.php from Special:Watchlist |
Event Timeline
Comment Actions
@jcrespo: Is this ticket maybe superseded by Wikimedia-database-issue since https://phabricator.wikimedia.org/T228360 ?
Comment Actions
I personally don't know what Wikimedia-database-issue is, probably the first time I learn about it.
Apparently that exist and also MediaWiki-libs-Rdbms and I have no idea what they are for. We DBA s don't use any of these- @Aklapper please ask whoever created those.
Comment Actions
Proposing to decline this unused "tracking task" which has been unused for >2 years. Project tags should be used instead.
Comment Actions
Sure, this used to be a very recurrent issue at the time, but AFAIK lately db issues seem to be of a different kind (long running times, large number of rows selected, etc. so this is not worth tracking anymore. I am going to resolve it, as the problem was real at the time and was for the most part fixed at T95501.
Comment Actions
For clarity sake (for others if they reach this task), this was about long open transactions, specially in the context of writes (which is one of the causes of replication lag)- causing mysql operational and redundancy problems. Wikimedia-Slow-DB-Query will be mostly for read only SELECTs, which cause performance issues for clients, primarily.