[
https://issues.apache.org/jira/browse/GEODE-4995?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16431370#comment-16431370
]
ASF subversion and git services commented on GEODE-4995:
--------------------------------------------------------
Commit 2e89e840f040800a65721c161d3b4203acfd7ff0 in geode's branch
refs/heads/develop from [~bschuchardt]
[ https://gitbox.apache.org/repos/asf?p=geode.git;h=2e89e84 ]
GEODE-4995 ClusterStartupRule is inefficient when shutting down the cluster
modified after() to shut down non-locators before shutting down locators,
allowing membership coordination to stay in the locator instead of
causing cascading movement of the role.
> ClusterStartupRule is inefficient when shutting down the cluster
> ----------------------------------------------------------------
>
> Key: GEODE-4995
> URL: https://issues.apache.org/jira/browse/GEODE-4995
> Project: Geode
> Issue Type: Test
> Components: tests
> Reporter: Bruce Schuchardt
> Assignee: Bruce Schuchardt
> Priority: Major
> Labels: pull-request-available
> Time Spent: 20m
> Remaining Estimate: 0h
>
> ClusterStartupRule is often used to create a locator and some number of
> servers, ignoring the locator started by DUnitLauncher. This is inefficient
> but it's probably intentional and can be avoided by configuring the servers
> to use the DUnitLauncher's locator.
> However, when shutting down the cluster it is managing ClusterStartupRule
> shuts down the VMs in order starting with vm0. If vm0 happens to be the
> membership coordinator this causes membership to create a new coordinator in
> vm1 and causes distributed lock grantor state to also fail over to vm1. Then
> vm1 is shut down, causing services to fail over to vm2. Etc.
> ClusterStartupRule should be modified to shut down non-locator VMs first and
> then shut down locator VMs.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
