Project: edge-cache · Branch: fix/redis-failover · Mode: default · Model: gpt-5

Summary

Debugged cache misses spiking during Redis failovers: the client kept a stale slot map. Enabled RouteByLatency with MaxRedirects, added a circuit breaker that falls back to the in-memory LRU, and wrote a chaos test with toxiproxy.

Conversation

Turn 1 — User

Every Redis failover gives us a minute of 100% misses.

Turn 1 — Assistant

The client keeps routing to the demoted primary until the slot map refreshes. Refresh on MOVED, add a breaker that degrades to the local LRU, and reproduce with toxiproxy in CI.

Tools used in this turn:

- Bash: docker compose -f chaos.yml up -d
- Edit: store/redis.go
- Write: store/breaker.go
- Bash: go test ./chaos -run Failover — ok

Connections

- edge-cache
- Redis
- CircuitBreaker
- ChaosTesting