The instinct when something’s slow is to put Redis in front of it. Sometimes that’s right. Often it’s a way to defer thinking about the actual problem, with the added bonus of now having two: the original slowness plus a cache that drifts out of sync with the truth in ways you can’t predict.

The question that filters most “let’s add a cache” proposals: how does it get invalidated? If the answer is “TTL and we’ll cross our fingers”, you’re not designing a cache, you’re rate-limiting how often your bug appears. A cache without a clear purging story — what triggers it, what gets purged, what happens when purging fails — is a wager that staleness will hurt less than slowness. Sometimes true. Rarely measured.