Rsyslog has introduced ratelimit in version 8.2602.0 (available in forky)
This has potential to be a drop-in improvement to the long standing issue where an error/bug/etc will spike log volume on the mw-k8s topic and cause log ingest to lag across all mediawiki channels and levels.
With this we could start setting rate limits before logs are produced to Kafka, based on combinations of log fields. e.g. each unique [channel]+[level] combination should not exceed N logs per M minutes.
Rough todo
- backport >= 8.2602.0 package to trixie
- update rsyslog in container
- draft/test POC ratelimit configuration
- establish sensible log rate limit scheme
- set up metrics/monitoring as related to ratelimit, consider alerting/logging/etc when enforced etc.
- deploy initial ratelimit config