Concurrency
2 admitted requests in flight per client. Bulk is the way to go wider.
2 admitted requests at a time
A client may have 2 admitted requests in flight. A third
concurrent request is rejected with 429 concurrency_limit_exceeded and a short
Retry-After.
How to stay under it
- Prefer one bulk call over many parallel detail calls.
- Keep a small worker pool (2) rather than firing requests in parallel from a loop.
- Treat
concurrency_limit_exceededas backpressure, not as a failure to retry immediately.
Concurrency is aggregated per client, like every other limit in Quotas.