Can Durable Object request serialization mimic connection pool exhaustion under burst traffic?

Yes, a single Cloudflare Workers Durable Object can produce latency spikes that resemble connection‑pool exhaustion when burst traffic forces its internal request queue to fill.

Each Durable Object runs on a single‑threaded event loop, so all incoming calls are serialized. Cloudflare buffers concurrent calls in an internal queue; when that queue exceeds its undocumented per‑object capacity the runtime returns a REQUEST_QUEUE_FULL error or times out, creating the same intermittent 5xx/429 patterns seen with exhausted TCP connection pools. Because the queue depth limit is not published and varies by plan and runtime version, the symptom is indistinguishable from a true connection‑pool limit without deeper diagnostics.

1. Reproduce a burst of requests to a single Durable Object key (e.g., using a script that fires many parallel fetches).
2. Capture request latency, success rate, and any error codes (particularly REQUEST_QUEUE_FULL or timeout) via Worker logs or the Cloudflare API.
3. Compare the observed queue depth or error frequency across different Worker plans (e.g., Pro vs. Business) to see if the behavior changes.
4. If possible, enable request tracing or use the “workers_dev” sandbox to inspect the internal queue length directly.

Please provide any logs or error payloads that indicate the Durable Object is hitting the REQUEST_QUEUE_FULL condition, as that would confirm the queuing behavior rather than a generic network timeout.