See also T105859: Some Flow topics are not being created properly in core page and revision tables which may or may be related.
Description
Status | Subtype | Assigned | Task | ||
---|---|---|---|---|---|
Resolved | DannyH | T104279 Make non-existent subpages in Flow occupied namespaces Flow again | |||
Resolved | Legoktm | T104278 E7. Run populateContentModel.php for page on all WMF wikis | |||
Resolved | Legoktm | T103388 E11. Write maintenance script to populate page.page_content_model, revision.rev_content_*, archive.ar_content_* |
Event Timeline
Why do we need to populate archive? Shouldn't we only do that when we plan on converting a namespace?
I think that is likely to be forgotten. I would much rather it be consistent across all the wikis.
It shouldn't be overly expensive if we don't do revision.
[10:38:27] <legoktm> !log running foreachwikiindblist /home/legoktm/largebutnotenwiki.dblist populateContentModel.php --ns=all --table=page
For the record, commonswiki has more pages than enwiki...oh well.
[14:57:50] <legoktm> !log running mwscript populateContentModel.php --wiki=enwiki --ns=all --table=page
@hoo found the culprit, Flow's ContentHandlerDefaultModelFor hook invoked LinkCache, which will grow indefinitely: T106998: LinkCache should be LRU based to avoid indefinitely growing and causing memory issues.
Restarted the script with LinkCache::singleton()->clear() statements added to avoid OOM'ing again. Not going to backport the LinkCache fix since it has scary perf implications.