zookeeper-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "ASF GitHub Bot (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (ZOOKEEPER-2849) Quorum port binding needs exponential back-off retry
Date Thu, 16 Nov 2017 22:52:01 GMT

    [ https://issues.apache.org/jira/browse/ZOOKEEPER-2849?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16256100#comment-16256100
] 

ASF GitHub Bot commented on ZOOKEEPER-2849:
-------------------------------------------

Github user afine commented on a diff in the pull request:

    https://github.com/apache/zookeeper/pull/419#discussion_r151560463
  
    --- Diff: src/java/main/org/apache/zookeeper/server/quorum/QuorumPeer.java ---
    @@ -946,7 +946,8 @@ protected Election createElectionAlgorithm(int electionAlgorithm){
                     le = new AuthFastLeaderElection(this, true);
                     break;
                 case 3:
    -                qcm = new QuorumCnxManager(this);
    +                qcm = new QuorumCnxManager(this,
    +                    ExponentialBackoffStrategy.builder().build());
    --- End diff --
    
    For instance, it may be nice to have the default strategy perform the existing behavior.


> Quorum port binding needs exponential back-off retry
> ----------------------------------------------------
>
>                 Key: ZOOKEEPER-2849
>                 URL: https://issues.apache.org/jira/browse/ZOOKEEPER-2849
>             Project: ZooKeeper
>          Issue Type: Improvement
>          Components: quorum
>    Affects Versions: 3.4.6, 3.5.3
>            Reporter: Brian Lininger
>            Assignee: Brian Lininger
>            Priority: Minor
>
> Recently we upgraded the AWS instance type we use for running out ZooKeeper nodes, and
by doing so we're intermittently hitting an issue where ZooKeeper cannot bind to the server
election port because the IP is incorrect.  This is due to name resolution in Route53 not
being in sync when ZooKeeper starts on the more powerful EC2 instances.  Currently in QuorumCnxManager.Listener,
we only attempt to bind 3 times with a 1s sleep between retries, which is not long enough.
 
> I'm proposing to change this to follow an exponential back-off type strategy where each
failed attempt causes a longer sleep between retry attempts.  This would allow for Zookeeper
to gracefully recover when the host is misconfigured, and subsequently corrected, without
requiring the process to be restarted while also minimizing the impact to the running instance.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

Mime
View raw message