Problem
The proxy starts with cold caches. Every cold-start request pays full upstream
latency. We don't currently know how much that costs us, and the proxy is
deployed multi-instance in environments where CACHE_MODE=LOCAL (Caffeine
per replica) would multiply prewarm load on upstream registries by replica
count.
Goal
Decide whether to build cache pre-heat now, defer it, or skip it entirely.
The design evaluates six options (in-proxy startup listener, scheduled
refresh-ahead, external CronJob hitting an admin endpoint, Redis-coordinated
election, lazy background warming, do nothing) and recommends a path with
explicit prerequisites and revisit criteria.
Design
See docs/designs/021-cache-prehit/DESIGN.md.
Problem
The proxy starts with cold caches. Every cold-start request pays full upstream
latency. We don't currently know how much that costs us, and the proxy is
deployed multi-instance in environments where
CACHE_MODE=LOCAL(Caffeineper replica) would multiply prewarm load on upstream registries by replica
count.
Goal
Decide whether to build cache pre-heat now, defer it, or skip it entirely.
The design evaluates six options (in-proxy startup listener, scheduled
refresh-ahead, external CronJob hitting an admin endpoint, Redis-coordinated
election, lazy background warming, do nothing) and recommends a path with
explicit prerequisites and revisit criteria.
Design
See
docs/designs/021-cache-prehit/DESIGN.md.