[
https://issues.apache.org/jira/browse/ZOOKEEPER-5063?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18093079#comment-18093079
]
Li Wang commented on ZOOKEEPER-5063:
------------------------------------
When the Observer Master feature is enabled in ZooKeeper, observers fail to
reconnect to participant leaders after the participant pods are restarted or
rescheduled with different IP addresses. This occurs because observers cache
the resolved IP addresses of observer masters at startup or reconfig and never
re-resolve them during the connection process.
> ZooKeeper observer master needs to dynamically resolve socket addresses
> -----------------------------------------------------------------------
>
> Key: ZOOKEEPER-5063
> URL: https://issues.apache.org/jira/browse/ZOOKEEPER-5063
> Project: ZooKeeper
> Issue Type: Bug
> Components: quorum, server
> Affects Versions: 4.0.0, 3.7.2, 3.10.0, 3.9.5, 3.8.6
> Environment: - Kubernetes deployments where ZooKeeper participants
> run as pods
> - Observer Master feature enabled (`observerMasterPort` configured)
> - DNS-based service discovery
> Reporter: Li Wang
> Assignee: Li Wang
> Priority: Major
>
> Observer Master feature does not dynamically recreate socket addresses when
> connecting to learner master, causing connection failures in Kubernetes
> environments where participant pods get rescheduled with new IP addresses.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
