Limits, backpressure and usage

Obtain operating limits with access and inspect hello.limits. Safety caps are distinct from pricing, accepted throughput and latency SLA.

Limits cover authentication/subscription/send time, message and replay-page size, replay lease lifetime and byte budget, unacknowledged count/bytes/age, and account resources shared across keys. A slow consumer can be disconnected; it must resume from committed state after improving processing.

Usage boundary

The current interface counts application batch text frames and bytes successfully accepted by the server’s local transport. Live and replay are tracked separately. Physically sending data again during replay consumes usage again.

Control frames do not count as billable batches, but remain bounded. ACK does not cancel usage or determine whether an accepted send is counted. Transport acceptance does not prove client receipt or fsync. Prices, invoicing and commercial rounding cannot be inferred from these counters.

Keep the consumer bounded

Use suitable filters, a finite local queue and idempotent page persistence at checkpoints. Track local commit time and checkpoint age. Avoid slow external actions directly inside each socket callback.

Retry only when allowed by the error frame, with finite backoff. See processing and errors.