Rate limits & isolation
Understand fair admission, overload, and safe retries.
Account limits
The Go gateway groups web sessions and API credentials by the same authenticated account. The initial gateway policy is 60 accepted requests per minute, two concurrent requests per account and four global requests. These are process-local controls and can reset on restart; durable quotas must remain authoritative.
No invisible queue
When admission is exhausted the gateway returns HTTP 429 with Retry-After. Use bounded exponential backoff with jitter for explicit admission rejection. Do not retry a generation blindly after a connection drop or timeout.
Cancellation
Abort the HTTP request to close the stream. The Go request context and upstream connection are canceled on disconnect. Backend cancellation must still be observed on the exact public route before production readiness is claimed.