mvanhorn opened a new pull request, #21714: URL: https://github.com/apache/kafka/pull/21714
When re-authentication fails in the consumer (e.g., temporary OAuth token server unavailability with SASL_OAUTHBEARER), the heartbeat timer is left in an expired state. This causes `timeToNextHeartbeat()` to return 0, leading the consumer poll loop to busy-wait with a 0ms timeout - pinning a CPU core at 100% (ClassicKafkaConsumer) or 40-50% (AsyncKafkaConsumer). ## Changes ### ClassicKafkaConsumer (AbstractCoordinator heartbeat thread) The heartbeat thread's `catch (AuthenticationException)` and `catch (GroupAuthorizationException)` blocks call `setFailureCause()` but never call `heartbeat.failHeartbeat()`. This leaves the heartbeat timer expired. After the user catches the exception and calls `poll()` again, `timeToNextHeartbeat()` returns 0, causing the main thread to spin. **Fix:** Call `heartbeat.failHeartbeat()` before `setFailureCause()` in both exception handlers. This resets the heartbeat timer with exponential backoff, preventing the 0-timeout spin. ### AsyncKafkaConsumer (AbstractHeartbeatRequestManager + StreamsGroupHeartbeatRequestManager) `maximumTimeToWait()` returns 0 when `shouldHeartbeatNow()` is true, even if the heartbeat request is in backoff (can't actually be sent). This causes the background network thread to spin. **Fix:** Check `canSendRequest()` before returning 0. If the request is in backoff, return the remaining backoff time instead. ## Testing - Added `HeartbeatTest.testFailHeartbeatResetsTimerWithBackoff()` - verifies that `failHeartbeat()` resets the timer to a positive backoff value after expiry - Added `ConsumerHeartbeatRequestManagerTest.testMaximumTimeToWaitRespectsBackoffWhenHeartbeatNow()` - verifies that `maximumTimeToWait()` returns a positive value when `shouldHeartbeatNow()` is true but the request is in backoff ## Reproducer See https://github.com/mstruk/kafka-consumer-reproducer for a standalone reproducer demonstrating the CPU spin with SASL_OAUTHBEARER. This contribution was developed with AI assistance (Claude Code). -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
