You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@nifi.apache.org by markap14 <gi...@git.apache.org> on 2018/04/19 13:09:58 UTC

[GitHub] nifi pull request #2646: NIFI-5096: Periodically poll ZooKeeper to determine...

GitHub user markap14 opened a pull request:

    https://github.com/apache/nifi/pull/2646

    NIFI-5096: Periodically poll ZooKeeper to determine the leader for ea…

    …ch registered role in Leader Election. This avoids a condition whereby a node may occasionally fail to receive notification that it is no longer the elected leader.
    
    Thank you for submitting a contribution to Apache NiFi.
    
    In order to streamline the review of the contribution we ask you
    to ensure the following steps have been taken:
    
    ### For all changes:
    - [ ] Is there a JIRA ticket associated with this PR? Is it referenced 
         in the commit message?
    
    - [ ] Does your PR title start with NIFI-XXXX where XXXX is the JIRA number you are trying to resolve? Pay particular attention to the hyphen "-" character.
    
    - [ ] Has your PR been rebased against the latest commit within the target branch (typically master)?
    
    - [ ] Is your initial contribution a single, squashed commit?
    
    ### For code changes:
    - [ ] Have you ensured that the full suite of tests is executed via mvn -Pcontrib-check clean install at the root nifi folder?
    - [ ] Have you written or updated unit tests to verify your changes?
    - [ ] If adding new dependencies to the code, are these dependencies licensed in a way that is compatible for inclusion under [ASF 2.0](http://www.apache.org/legal/resolved.html#category-a)? 
    - [ ] If applicable, have you updated the LICENSE file, including the main LICENSE file under nifi-assembly?
    - [ ] If applicable, have you updated the NOTICE file, including the main NOTICE file found under nifi-assembly?
    - [ ] If adding new Properties, have you added .displayName in addition to .name (programmatic access) for each of the new properties?
    
    ### For documentation related changes:
    - [ ] Have you ensured that format looks appropriate for the output in which it is rendered?
    
    ### Note:
    Please ensure that once the PR is submitted, you check travis-ci for build issues and submit an update to your PR as soon as possible.


You can merge this pull request into a Git repository by running:

    $ git pull https://github.com/markap14/nifi NIFI-5096

Alternatively you can review and apply these changes as the patch at:

    https://github.com/apache/nifi/pull/2646.patch

To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:

    This closes #2646
    
----
commit 757c4e2445d052b593fbea8f0d9a36bac001f44f
Author: Mark Payne <ma...@...>
Date:   2018-04-19T13:05:32Z

    NIFI-5096: Periodically poll ZooKeeper to determine the leader for each registered role in Leader Election. This avoids a condition whereby a node may occasionally fail to receive notification that it is no longer the elected leader.

----


---

[GitHub] nifi issue #2646: NIFI-5096: Periodically poll ZooKeeper to determine the le...

Posted by mcgilman <gi...@git.apache.org>.
Github user mcgilman commented on the issue:

    https://github.com/apache/nifi/pull/2646
  
    @markap14 I see. It appears then that the underlying issue is that either (1) the stateChange method is not being invoked or (2) the leader thread interruption is not happening/working. We could include your proposed changes and update our implementation of stateChanged to set `leader` to false when the `newState` is LOST or SUSPENDED before invoking super. Additionally, in `takeLeadership` we should loop while not stopped and is leader. This should help if the underlying issue was (2) while the polling could act as additional insurance.


---

[GitHub] nifi issue #2646: NIFI-5096: Periodically poll ZooKeeper to determine the le...

Posted by mcgilman <gi...@git.apache.org>.
Github user mcgilman commented on the issue:

    https://github.com/apache/nifi/pull/2646
  
    Thanks @markap14! This has been merged to master.


---

[GitHub] nifi pull request #2646: NIFI-5096: Periodically poll ZooKeeper to determine...

Posted by asfgit <gi...@git.apache.org>.
Github user asfgit closed the pull request at:

    https://github.com/apache/nifi/pull/2646


---

[GitHub] nifi issue #2646: NIFI-5096: Periodically poll ZooKeeper to determine the le...

Posted by mcgilman <gi...@git.apache.org>.
Github user mcgilman commented on the issue:

    https://github.com/apache/nifi/pull/2646
  
    @markap14 These changes look like they will reduce the time when there are multiple primaries (leaders) in the event the existing primary isn't notified of the change. That's definitely an improvement, however, I was curious if you tried incorporating a `ConnectionStateListener` for the node to be more proactively notified when the connection is SUSPENDED or LOST. Here's an SO post [1] where it is discussed.
    
    [1] https://stackoverflow.com/questions/41042798/how-to-handle-apache-curator-distributed-lock-loss-of-connection


---

[GitHub] nifi issue #2646: NIFI-5096: Periodically poll ZooKeeper to determine the le...

Posted by markap14 <gi...@git.apache.org>.
Github user markap14 commented on the issue:

    https://github.com/apache/nifi/pull/2646
  
    @mcgilman I agree. I have pushed a new commit that does just that.


---

[GitHub] nifi issue #2646: NIFI-5096: Periodically poll ZooKeeper to determine the le...

Posted by markap14 <gi...@git.apache.org>.
Github user markap14 commented on the issue:

    https://github.com/apache/nifi/pull/2646
  
    @mcgilman we do indeed implement the ConnectionStateListener, but we do so only to log the fact and then call super.stateChanged(). When we call super.stateChanged(), that will throw CancelLeadershipException, which in turn is supposed to interrupt our listener. We followed the "Error Handling" guidance provided by Apache Curator: https://curator.apache.org/curator-recipes/leader-election.html
    
    So we are handling the SUSPENDED and LOST scenarios as is recommended. And this works 99% of the time. Unfortunately, we do occasionally see scenarios where it does not interrupt the thread and as such the node believes that it retains the lock. It's not clear, when this happens, if the thread just wasn't interrupted for some reason, or if the notification of SUSPENDED/LOST never was received, or what exactly is occurring that prevents our ElectionListener from being interrupted.
    
    That's why I went with the solution of periodically polling ZooKeeper, to check the state. That way, whatever the cause of the thread not being interrupted, we still will break out. If you think it makes sense, though, we can detect the LOST state specifically and have that trigger us to leave the election, in addition to polling?


---

[GitHub] nifi issue #2646: NIFI-5096: Periodically poll ZooKeeper to determine the le...

Posted by mcgilman <gi...@git.apache.org>.
Github user mcgilman commented on the issue:

    https://github.com/apache/nifi/pull/2646
  
    @markap14 Looks like our `LeaderSelectionListener` does implement `ConnectionStateListener`. Do we need to update code on our side to explicitly give up leadership in the SUSPENDED or LOST scenarios?


---