Summary

Another WDQS incident this week disrupted our flow of work.

Dealing with page redirect in the context of the Search Update Pipeline is more complex than expected, and involves multiple teams (data engineering, mediawiki core, ML). Hopefully that additional work will benefit more teams, in particular ML.

What we've accomplished

Improve multilingual zero-results rate

documentation and some implementation of the framework to evaluate impact https://www.mediawiki.org/wiki/User:TJones_(WMF)/Notes/Language_Analyzer_Harmonization_Notes

= Search Update Pipeline

Discussions with Data Engineering and Mediawiki core about page redirects - https://phabricator.wikimedia.org/T325315#8871319

Finalize Search SLOs

Cleanup of the final list of SLOs - https://docs.google.com/document/d/1gYROXo8Fl7JSxReHAVI22EhcPvG-INVkq79a1C3tfK0/edit#heading=h.2v65rh3w6ii0

Operations / SRE

Dealing with the WDQS outage - https://wikitech.wikimedia.org/wiki/Incidents/2023-05-23_wdqs_CODFW_5xx_errors
Fixed prometheus metrics not being reported by flink-1.16 - https://phabricator.wikimedia.org/T336872

Misc

Trying to make sense to the perceived limitations of Search when being used in the context of a ChatGPT plugin
The language team finished the work on (https://phabricator.wikimedia.org/T322284) sent a patch to enable this new config. This should allow us in the end to quickly switch traffic from one DC to another (unblocking https://phabricator.wikimedia.org/T143553)
Turkish is reindexed to enable unpacking, with write up on mediawiki https://www.mediawiki.org/wiki/User:TJones_(WMF)/Notes/Unpacking_Notes#better_apostrophe_Impact_on_Turkish_Wikipedia_(T337064) - https://phabricator.wikimedia.org/T337064