Redis eviction policy

I see in the performance tuning document.

Sentry recommends using allkeys-lru. I’m wondering if this is only true for the sentry buffer instances.

We have a shared redis for the buffer and tsdb. So when we reach memory limits, the redis starts evicting tsdb data causing missing histogram on Sentry’s UI.

So would setting volatile-ttl or volatile-lru break the buffer usage in any way?

We have enough space to expand the redis, and that’s what we do, but it breaks the UI histograms every time there a surge of new exceptions and were looking to avoid that.

I just realized that all keys have a TTL on it. Changing the policy would do nothing.