[jira] [Created] (AMBARI-18728) During cluster install, Components get timed out icon while starting

["Andrew Onischuk (JIRA)" <ji...@apache.org>]

Andrew Onischuk created AMBARI-18728:
----------------------------------------

             Summary: During cluster install, Components get timed out icon while starting
                 Key: AMBARI-18728
                 URL: https://issues.apache.org/jira/browse/AMBARI-18728
             Project: Ambari
          Issue Type: Bug
            Reporter: Andrew Onischuk
            Assignee: Andrew Onischuk
             Fix For: 2.5.0
         Attachments: AMBARI-18728.patch

This was caused by a very tricky race-condition in the way python multiprocessing.thread works resulting in deadlock in ambari_agent.ActionQueue thread.
The problem is the below flow:
If this all these three get executed at the same time (a very rear occasion):
1. Process1 executes queue.get(False)
2. Process2 executes queue.put(largeObjectWhichTakesLongTimeToPut)
3. Someone kills Process2.

This results in deadlock in process1 get. Which is caused by queue locks/semaphores to being released during put of process2.

I have wrote a script test_race_condition.py to emulate this behaviour and indeed could reproduce this and test the fix for it.




--
This message was sent by Atlassian JIRA
(v6.3.4#6332)