Closing a capacity risk before it became an outage
Localis caches the HTML behind every public code page so lookups stay fast. That cache shares underlying infrastructure with the background job system that keeps our data pipelines running, which means unchecked growth on the cache side becomes capacity the job system needs too.
During a routine review, we found the cache had grown further than actual traffic justified — more pages and images were being built and held in reserve than anyone was using. Left alone, that growth would eventually have exhausted the shared capacity and put the background job system at risk along with it.
We closed the gap before it caused any customer-facing slowdown or outage. The fix was to stop over-provisioning cache work that wasn't earning its cost: build and hold fewer pages ahead of real demand, so the shared system keeps genuine headroom instead of a shrinking one.
Nothing here was a bug in the sense of a wrong answer — every rate we served stayed correct throughout. This was proactive housekeeping: a shared resource was trending toward a ceiling, and we brought it back down before anyone hit it.