Understand acknowledgement, backlog and recovery.
Use real receipts and set clear delivery boundaries.
Native durable queues
FlowPort uses native durable storage without SQLite. Source acknowledgement confirms persistence on the current Worker. Worker queues are independent and are not automatically replicated.
Logical quota is not physical disk usage. Failed batches consume capacity. Reserve disk and queue headroom and monitor storage health and backpressure.
Acknowledgement and retry
Downstream acknowledgement follows each protocol, such as HTTP success, Kafka ISR acknowledgement or an object write. There is no cross-destination transaction or absolute ordering.
Retries and crash recovery may produce duplicates. Use downstream deduplication or idempotency where required. Delivered points divided by received batches is not a success rate.
Failed records and recovery
Correct connectivity, credentials or format before replaying affected failures. Successful branches do not need to resend. Deleting failed data discards it; review the selected scope.
Stopping a flow stops new ingestion while accepted data continues processing and delivery. Drain and confirm no backlog before removing nodes or data directories.
Metrics and diagnostics
Overview provides state and throughput charts. Runtime animations represent sampled activity, not individual record tracing. Prometheus /metrics is available; configure authentication and scraping for the deployed role.
Connection testing separates network reachability from protocol requests. Read-only checks cannot prove write permission. Sampling does not add consumers, acknowledge messages or advance cursors.