This is enabled on the beta cluster and works as expected.

This is done. I no longer see ReadingLists ("Saved pages") as a beta feature but still have access to my ReadingLists.

This looks good to me on the beta cluster:

I have restored the user's reading lists

@dancy Thank you! This is very helpful and we will be careful to avoid any adding tests to ReadingLists that could be problematic (e.g. flaky tests).

I think we have enough metrics in grafana and we are seeing improvements in the cache miss rate after increasing the cache TTL. I do not think extra logging will help in any meaningful way.

Looking at the metrics in grafana for the bloom filter cache misses, the vast majority of cache misses are because the cache entry is absent (vs stale).

Here is a task for running the script for the affected user: https://phabricator.wikimedia.org/T434018

Should this task be renamed to focus initial efforts on Minerva?

Also, with mediawiki.popover, is this globally available (e.g. for gadgets) or can we limit it to only things that need to use this?

I have some questions about this:

I set a time on the calendar for next week to review this with Jon and the team.

Alternative approach considered was to add a daily testing job (see https://phabricator.wikimedia.org/T406169), but it seems undesirable to make changes to the current zuul setup (since upgrade is happening).

@Peter also we do not have any selenium tests, but have a set of api integration tests and jest (and moving those to vitest)

@Peter In https://integration.wikimedia.org/ci/job/mwext-phpunit-coverage-publish/63714/console, the tests took 10.483s with 450 tests (2671 assertions).

What is not answered by existing metrics is "what triggered rebuild?" (cache miss/absent or user saved/unsaved a page), and the outcome of the rebuild job to know "How often do the jobs complete successfully or fail?"

there already is instrumentation for the job queue in general, including being able to look at specific jobs.

Increasing the cache TTL for the bloom filter to 30 days appears to be helping to improve the cache miss ratio.

I added panels to the bloom filter grafana dashboard for cache entry age, cache entries over 7 days (for the cache TTL change), and cache miss reason

For the "Serial API calls on mount", we are looking at that in https://phabricator.wikimedia.org/T406775 (feedback welcome there)

I noticed the unused $list variable in HookHandler when looking at the bloom filter. Here is a patch for that: