Fix consumer hangs after mailbox failure

[bartoszs321]

Summary

• keep failed single-topic consumer mailboxes draining requests so calls made after shutdown complete instead of hanging
• propagate multi-topic poller failures to the parent consumer and fail pending receives
• drain failed multi-topic mailbox requests and close child consumers when DisposeAsync is called
• add integration coverage for single-topic and multi-topic mailbox failures

Problem

When a consumer mailbox faulted, its reader stopped. Requests posted afterwards were accepted by the unbounded channel but never received a response. This could leave operations such as GetStats, ReceiveAsync, and DisposeAsync waiting indefinitely. A multi-topic poller failure was only logged, leaving the parent consumer apparently ready even though it could no longer deliver messages.

Implementation

After a mailbox exits, a lightweight drain continues reading its channel. Requests that can no longer be performed fail with AlreadyClosedException; close requests complete after the necessary local or child-consumer cleanup. Multi-topic poller failures are posted to the parent mailbox, which moves the consumer to a failed state and completes pending receives.

Verification

• Unit tests: 216 passed, 2 skipped
• Mailbox integration tests against Apache Pulsar 3.3.0: 3 passed

[Copilot]

Copilot review overview

🟡 Changes recommended

Faulted single-topic consumers remain in the wrong lifecycle state after disposal, and pending-receive coverage is incomplete.

Review effort: Balanced
Findings: 2

Open (2)

• Close request leaves disposed consumer in Failed state · New
• Test queued ReceiveAsync failure when poller stops · New

What changed in this PR

Improves consumer failure handling so mailbox failures no longer leave requests hanging.

Changes:

• Drains stopped consumer mailboxes and completes requests.
• Propagates multi-topic poller failures.
• Adds mailbox-failure integration coverage.

File	Description
ConsumerImpl.fs	Drains requests after mailbox termination.
MultiTopicsConsumerImpl.fs	Handles poller failures and child cleanup.
Basic.fs	Tests mailbox-failure behavior.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Lanayx]

This should be NotConnectedException

[bartoszs321]

Updated in 29edfc2

[Lanayx]

Thank you for the PR, would you be able to implement similar behavior for Producer, so it will be single PR for both

[Copilot]

Copilot review overview

🟡 Changes recommended

Stopped consumers can retain pooled payloads, and failed partition cleanup is incorrectly reported as successful.

Review effort: Balanced
Findings: 1

Open (1)

• DisposeAsync hides partition producer close failures · New

Resolved since last review (2)

• Test queued ReceiveAsync failure when poller stops
• Close request leaves disposed consumer in Failed state

Previously missed (1)

In code that hasn't changed since last review

Mailbox failure drops messages without disposing payloads

src/​Pulsar.Client/​Internal/​ConsumerImpl.fs:891

The catch-all also drops queued MessageReceived values without disposing their RawMessage.Payload. These payloads come from MemoryStreamManager, and every normal discard path explicitly disposes them (for example, ConsumerImpl.fs:1002-1052), so messages racing with a mailbox failure retain pooled buffers. Add a stopped-mailbox case that disposes the payload before discarding the message.

[Copilot]

Copilot review overview

🟡 Changes recommended

A partitioned producer request can still hang when that request triggers the mailbox failure, and the multi-topic test does not exercise waiter cleanup.

Review effort: Balanced
Findings: 1

Open (1)

• Drain leaves the faulting request hanging · New

Resolved since last review (1)

• DisposeAsync hides partition producer close failures