[MESOS-3910] Libprocess: Implement cleanup of the SocketManager in process::finalize

• Type: Task
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 1.2.0
• Component/s: libprocess, test
• Labels: mesosphere
• Epic Link: libprocess-finalize
• Sprint: Mesosphere Sprint 40, Mesosphere Sprint 41, Mesosphere Sprint 42
• Story Points: 5
• Assignee: Joseph Wu
• Reporter: Joseph Wu
• Shepherd: Joris Van Remoortere
• Created: 13/Nov/15 00:33
• Updated: 26/Nov/18 12:20
• Resolved: 26/Nov/18 12:20

Description

The socket_manager and process_manager are intricately tied together. Currently, only the process_manager is cleaned up by process::finalize.

To clean up the socket_manager, we must close all sockets and deallocate any existing HttpProxy or Encoder objects. And we should prevent further objects from being created/tracked by the socket_manager.

Proposal

1. Clean up all processes other than gc. This will clear all links and delete all HttpProxy s while socket_manager still exists.
2. Close all sockets via SocketManager::close. All of socket_manager 's state is cleaned up via SocketManager::close, including termination of HttpProxy (termination is idempotent, meaning that killing HttpProxy s via process_manager is safe).
3. At this point, socket_manager should be empty and only the gc process should be running. (Since we're finalizing, assume there are no threads trying to spawn processes.) socket_manager can be deleted.
4. gc can be deleted. This is currently a leaked pointer, so we'll also need to track and delete that.
5. process_manager should be devoid of processes, so we can proceed with cleanup (join threads, stop the EventLoop, etc).

Issue Links

• relates to: MESOS-4603 Libprocess tests may crash when during finalization.
• relates to: MESOS-5144 Cleanup memory leaks in libprocess finalize()