[MESOS-8729] Libprocess: deadlock in process::finalize

• Type: Bug
• Status: Accepted
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 1.6.0
• Component/s: libprocess
• Labels: deadlock libprocess

Environment

The issue has been reproduced on Ubuntu 16.04, master branch, commit `42848653b2`. 

• Epic Link: Fix process::finalize
• Story Points: 5
• Assignee: Andrei Budnik
• Reporter: Andrei Budnik
• Shepherd: Benjamin Mahler
• Created: 23/Mar/18 20:02
• Updated: 23/Nov/20 20:55

Description

Since we are calling `libprocess::finalize()` before returning from the IOSwitchboard's main function, we expect that all http responses are going to be sent back to clients before IOSwitchboard terminates. However, after adding `libprocess::finalize()` we have seen that IOSwitchboard might get stuck in `libprocess::finalize()`. See attached stacktrace.

Issue Links

• blocks: MESOS-8545 AgentAPIStreamingTest.AttachInputToNestedContainerSession is flaky.
• causes: MESOS-8723 ROOT_HealthCheckUsingPersistentVolume is flaky.
• relates to: MESOS-10201 Leaving libprocess running after main() exit may make a program crash when global mutexes in glog are destroyed.