Large Instances Meeting Notes 2024-04-30

We had a new participant from eduNEXT, Jorge Fandiño, so we had a round of introductions.

Updates from each org

eduNEXT / @Jhony Avella - monitoring the effort to migrate packages to python 3.12 as part of the Redwood release effort. Also working to publish an open source repo with a set of k6 tests that we use to check Open edX instances - hoping to share in our next meeting. Hopefully it could become a shared/standard tool set for testing Open edX.

2U - has not had anyone joining these meetings lately. We should see if there is someone else who’d be able to join, as we always appreciated having their perspective. TODO: ask them.

OpenCraft / @Gábor Boros - continued working on migrating parts of internal “Grove” software to Harmony. Ran into the known issue with Elasticsearch certificates, but found a workaround that seems to work for now (PR incoming).

[Discussion of Meilisearch: @Felipe Montoya asked: in a world where it replaces Elasticsearch, since it’s so lightweight, could we just run many small instances of Meilisearch to support many Open edX instances, instead of one large cluster? @Braden MacDonald : Absolutely, you can do either approach. Meilisearch has really good multi-tenancy if you want to share one large instance (it doesn’t support clustering/replication yet). But since it’s so lightweight, you can deploy one Meilisearch instance per Open edX instance and it would be more reliable without using too many resources. You could even deploy one instance per use case (studio search, learner courseware search, forum search). @Felipe Montoya What do you think makes the most sense? @Braden MacDonald I’d probably default to one instance of Meilisearch per Open edX instance and see how that goes.]

[Discussion about sharing secrets for Elasticsearch certificates on k8s.]

Raccoon Gang / @Maksim Sokolskiy - focused on testing a large installation on k8s. Should soon be able to publish their locust load testing scripts etc., which may be useful as an alternative to the k6 tests. Also, Raccoon Gang tried AWS serverless Aurora and Redis - found that it works well for large instances so far, and seems like it can save costs.

[Continued discussion of issues with celery tasks, and figuring out how we can fix it in tutor or Harmony. What Raccoon Gang is doing is just copying the config from the old “configuration” repo, adapted to Kubernetes.]

[We’ll review Harmony project issues in more detail next meeting]