Skip to content

Refine apdex metrics for redis-cache

In gitlab-com/runbooks!5560 (merged) we broke down the cache metrics by store, initially to differentiate between RepositoryCacheStore and the main RedisCacheStore store. We noticed that this would also have the side effect of making the redis-cache instance more accurate as we'd exclude LocalStore and MemoryStore metrics, but we didn't realize how large the impact would be. The RPS dropped from ~600k to ~100k, and the Apdex score degraded (in retrospective, predictably, as both LocalStore and MemoryStore are naturally faster than a Redis store), causing production#8574 (closed). We need to refine our Apdex definition to match the actual data.