This follow-up task from an incident report has not been updated recently. If it is no longer valid, please add a comment explaining why. If it is still valid, please prioritize it appropriately relative to your other work. If you have any questions, feel free to ask me (Greg Grossmeier).

Adding a reference to the outage mentioned:

https://wikitech.wikimedia.org/wiki/Incident_documentation/20150825-Redis

I created https://phabricator.wikimedia.org/T148637 to add Redis metrics to Prometheus.

Should we merge this with T148637 ?

To clarify, this task and the linked incident, are about the rdb* hosts. These are known to MW as redis_lock and in monitoring as redis_misc. I'm mentioning this because it means the task remains relevant after T267581: Phase out "redis_sessions" cluster and away from memcached cluster.

Removing SRE, triaging to serviceops. redis_misc is in our care as a team and we should decide what we want to do regarding better monitoring of it.

We're already alerting on disk space for all servers, not sure why this would be different.

Adding an alert on a full memory for the redis datastore can work, but needs silencing of the instances used by ORES.

I'd keep it simple instead of making the whole thing overly complex, thus just alerting on

redis_memory_used_bytes / redis_memory_max_bytes > 0.98

and manually silencing the three ores-related instances.

Change 901141 had a related patch set uploaded (by Giuseppe Lavagetto; author: Giuseppe Lavagetto):

[operations/alerts@master] sre: add redis memory full alert

https://gerrit.wikimedia.org/r/901141

Change 901141 merged by jenkins-bot:

[operations/alerts@master] sre: add redis memory full alert

https://gerrit.wikimedia.org/r/901141