Push backoff policy
When a subscription endpoint becomes unavailable, Simple Message Queue (formerly MNS) automatically reduces the push rate to protect overall service availability. This mechanism is the push backoff policy.
How it works
The push backoff algorithm adjusts two parameters based on the number of consecutive failed pushes:
| Parameter | Normal state | Under backoff |
|---|---|---|
| Push interval | 0 to 500 ms | Increases gradually, up to 60 seconds |
| Messages per push | Up to 32 | Decreases gradually, down to 1 |
As consecutive failures increase, delivery slows until the endpoint recovers.
Fast recovery
When Simple Message Queue (formerly MNS) detects that the endpoint is available again, a fast recovery algorithm restores both the push interval and message count to normal values within 100 seconds.
Impact on retry behavior
During push backoff, the retry interval increases beyond the values defined in the retry policy (NotifyStrategy), and the actual number of retries per message decreases. Configure a dead-letter queue for each subscription to capture messages that fail delivery during backoff.
Causes of subscription unavailability
Deleted queue endpoint
If the subscribed queue has been deleted, the subscription becomes unavailable immediately.
Before you delete a queue, verify that no active topics have subscriptions pointing to it. Deleting a subscribed queue disrupts message delivery for all associated topic subscriptions.
HTTP/S endpoint failures
When message pushes to an HTTP/S endpoint fail repeatedly, Simple Message Queue (formerly MNS) progressively reduces the push rate. This can cause a cascading message backlog:
New messages continue to arrive while the push rate is throttled.
The system retries old messages first. Many of these may have already expired.
Because the push rate is low, new messages cannot be delivered in time. They also accumulate and eventually expire.
This cycle can halt your subscription service.
To recover from an HTTP/S backoff scenario:
Fix the endpoint. Make sure your HTTP/S server is running and responds with a
2xxstatus code to incoming push requests.Reset the consumer offset. In the Simple Message Queue (formerly MNS) console, reset the subscription consumer offset to the current time. This skips all accumulated messages, including expired ones, and resumes delivery from the latest messages.
Usage notes
The push backoff mechanism is automatic and cannot be manually enabled or disabled.
Push backoff applies to all messages for an unavailable subscription to a specific topic, not individual messages.
If a subscription remains unavailable longer than the message retention period, undelivered messages are discarded. To prevent message loss, maintain subscriber availability or configure a dead-letter queue so that these messages are routed there instead of being lost.