Version: 8.3.2

Problem Summary

The ICapPublisher.PublishDelayAsync method in the .NET Core CAP library exhibits thread-safety and cancellation handling issues when the delay time is less than 1 minute. These issues can lead to silent failures and message accumulation in the database.

Detailed Problems

• Thread-safety Issue:

  • The Dispatcher.EnqueueToScheduler method uses a non-thread-safe PriorityQueue to manage delayed messages.
  • Under race conditions, the PriorityQueue can return null messages.
  • This causes the fire-and-forget task (Dispatcher: line 80) to fail silently, with no error logging.
  • As a result, delayed message processing halts, and future scheduled messages are not sent until the application restarts.

• Cancellation Handling Issue:

  • The _delayCts cancellation token is initialized only once.
  • Once _delayCts is cancelled, subsequent calls to await Task.Delay(new TimeSpan(delayTime), _delayCts.Token) throw an OperationCanceledException.
  • The exception is ignored in the catch block, leading to a while (TryPeek) loop that spins continuously until the delay time expires.

• Consequences of Silent Failures:

  • Delayed messages continue to accumulate in the database without being processed.
  • Upon application restart, querying delayed messages without a limit causes the database to timeout.
  • Manual intervention is required to delete unprocessed messages and restore application functionality.

Steps to Reproduce

1. Call ICapPublisher.PublishDelayAsync concurrently with delay times less than 1 minute.
2. Observe that:
   • Some messages are not processed.
   • Future scheduled messages are not sent.

I have fixed the above problems while maintaining existing functionality as closely as possible. I will open a pull request with tests and code changes to address the issue.