Rate limit middleware counter persistence during Traefik dynamic configuration reload

Answer

When Traefik’s rate‑limit middleware is re‑loaded by a dynamic configuration change (e.g. touching the file provider), the default behaviour is to reset all leaky‑bucket counters to zero. The middleware instance is discarded and a fresh one is created, so the in‑memory counters are lost. If a persistent store (Redis, DynamoDB, or a shared memory store with persistence) is configured for the rate‑limit store, the counters survive the reload and decay according to the elapsed time since the last request.

Explanation

• In‑memory store (default): counters are stored only in the process memory. A reload recreates the middleware, discarding the old counters. The bucket starts full again, so a burst of requests can pass until the bucket refills.
• Persistent store: counters are written to an external backend. Reloading does not delete the stored state; the bucket continues to decay based on the configured average and burst values. The new middleware instance reads the existing counter values from the backend.
• Graceful shutdown: during a graceful stop, in‑memory counters are also discarded. The next start will behave like a reload with an in‑memory store.

Verification Steps (in a non‑persistent setup)

1. Send a burst of requests that exceeds the configured burst to fill the bucket.
2. Check Traefik metrics traefik_http_request_rate_limit_bucket (or similar) to confirm the bucket is full.
3. Reload the dynamic configuration (e.g. touch /etc/traefik/traefik.yml).
4. Re‑check the metric; it should show the bucket reset to 0 or near‑full again.

Verification Steps (with a persistent store)

1. Configure the rate‑limit store to use Redis: store: redis with redis: { address: "redis://127.0.0.1:6379" }.
2. Repeat the burst request and metric check.
3. Reload the configuration.
4. Verify that the metric still reflects the previous bucket state and that the bucket decays over time.

Missing Diagnostic Detail

Could you confirm whether you are using a persistent store for the rate‑limit middleware? This will determine whether the counters survive reloads or are reset.