I am self-hosting medplum and noticed that it creates a Login record on redis server everytime there is a login from a client. Over a few days, this can end up with thousands of login keys. Is this expected? Are they not supposed to expire? Should we delete them manually?

Redis has evictions, and if it gets full will start evicting old logins, etc.

That's right. I believe you can tune your redis instance to choose the memory limit for eviction (https://redis.io/docs/reference/eviction/) Medplum uses volatile-lru for its eviction policy (https://docs.redis.com/latest/rs/databases/memory-performance/eviction-policy/)

>Medplum uses volatile-lru for its eviction policy On server start, bullmq throws a [warning](https://github.com/taskforcesh/bullmq/blob/9307fef3e5f9e5f929fd8755078913a33c2dc783/src/classes/redis-connection.ts#L352)

IMPORTANT! Eviction policy is volatile-lru. It should be "noeviction"

. Running against GCP managed redis but looks like its configured similarly to the cdk script. Also looks similar to the terraform/gcp version in the open PR (except that one's configured to [volatile-ttl](https://github.com/medplum/medplum/pull/5390/files#diff-a58ec1195c15117097420636a74a91d74cb589ad621b4afa72937b8a322881e0R15) 🤔 ). Think this warning is benign? Happen to have any tips on suppressing if so?

@nathan-watkins unfortunately this is a known issue wiht bullmq, and there's a ticket filed against them. You're right that this is benign, but the fix would have to happen upstream

Thank you! Yeah saw this [issue](https://github.com/taskforcesh/bullmq/issues/2737). I'm not sure there's a "fix" short of standing up a second redis instance just for bullmq with

noeviction

or possibly updating the global

console.warn

to intercept it, which seems like overkill