Read the queue state
Ready messages are waiting in the queue; unacknowledged messages have been delivered and await acknowledgement. A growing total can have different causes depending on which group is increasing. Capture the queue name, rates, consumer count and time window.
If ready messages rise while unacknowledged messages stay low, investigate whether consumers are connected, permitted to consume and able to receive work. If unacknowledged messages rise, inspect processing time, blocked downstream systems, acknowledgement behavior and consumer failures.
Inspect prefetch in context
Prefetch limits the number of outstanding unacknowledged deliveries according to the consumer and channel configuration. A low value can limit parallel work; a very high value can concentrate work in consumers and increase recovery time after a failure. The right value depends on message size, processing time, memory and ordering requirements.
Check application logs and downstream latency before simply increasing consumers or prefetch. Scaling consumers can overload a database or external API. Acknowledging early may reduce the visible unacknowledged count while losing work if the process fails before completing the business operation.
Recover safely
Identify poison messages and retry behavior. Test a bounded change, monitor publish, deliver and acknowledgement rates, and reconcile completed business operations. Preserve message identifiers and avoid dumping sensitive payloads into logs.
Document the owner and recovery procedure. Queue depth falling is useful evidence, but confirm the intended work actually completed and duplicates were handled.
Further reading
Read the official documentation. For help applying these checks, explore Clouvenix services.