Summary

This update covers the last 2 weeks: there wasn't much progress due to holidays and vacations. With the end of year approaching, more vacations will reduce our output.

The end of year freeze implies that some of the work that is ready will still wait until January to be deployed in production. In particular, the Search Update Pipeline will wait until January to be deployed on the first production wiki.

What we've accomplished

Improve multilingual zero-results rate

First implementation of ICU token repair is passing unit tests - https://phabricator.wikimedia.org/T332337

WDQS graph splitting

Investigation of IGUANA as a SPARQL testing framework, writing a tool to convert IGUANA RDF results into a tabular CSV format - https://phabricator.wikimedia.org/T351894
Create a tool that records and compares a set of sparql query results (in progress) - https://phabricator.wikimedia.org/T351819
Completed evaluation of IGUANA and TFT as test frameworks. TFT is not fit for our purpose, but IGUANA might work. See ticket for more details - https://phabricator.wikimedia.org/T349519
We've been observing performance degradation on the newer WDQS servers, which is probably linked to CPU starvation, with newer servers having more cores but slower - https://phabricator.wikimedia.org/T336443

Search Update Pipeline

Pipeline has been stable for a week - https://phabricator.wikimedia.org/T347075
The Flink Elasticsearch sink does not provide an easy way to ignore/retry individual flink failures (https://issues.apache.org/jira/browse/FLINK-32028) workaround added
Found (and fixed) an issue in how EventBus processes changes involving redirects - https://phabricator.wikimedia.org/T350299
Minor fixes to corner cases in the pipeline
Create/adapt index comparing script for checking correctness - https://phabricator.wikimedia.org/T349838

Search SLOs

Reviewed our current way to collect SLI and decide it is sufficient - https://phabricator.wikimedia.org/T335499

Misc

Lexeme searches prefer forms over lemmas - https://phabricator.wikimedia.org/T348877

Operations

Reduce the number of false positive alerts for p95 latency on Search - https://phabricator.wikimedia.org/T349340