[
https://issues.apache.org/jira/browse/HDDS-16206?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18105744#comment-18105744
]
Ivan Andika commented on HDDS-16206:
------------------------------------
cc: [~hevinhsu]
> Datanodes in MiniOzoneCluster are always started pre-finalized
> --------------------------------------------------------------
>
> Key: HDDS-16206
> URL: https://issues.apache.org/jira/browse/HDDS-16206
> Project: Apache Ozone
> Issue Type: Bug
> Components: test
> Reporter: Ethan Rose
> Priority: Major
>
> After HDDS-14812, Datanodes in MiniOzoneCluster (not production) are always
> starting pre-finalized, and finalizing in the first heartbeat after they
> register with SCM.
> HDDS-14812 creates the datanode.id file independently to inject fields into
> the MiniOzoneCluster. When the cluster is started after this, Datanodes see
> the file and assume they are being upgraded from an existing installation
> that did not have the upgrade framework, and use metadata layout version 0.
> This test setup is incorrect since it does not reflect how a real cluster is
> constructed. We need a different way to inject these configurations.
> On master this doesn't cause any existing test issues, because SCM will
> instruct the Datanodes to finalize before they can be used to meet safemode
> exit criteria, blocking the tests until Datanodes are on the expected
> version. With ZDU, pre-finalized Datanodes are valid for all existing
> operations and do not block safemode exit. On the ZDU branch this change
> causes intermittent test failures (like {{TestFinalizeBlock}}) because writes
> expecting finalized Datanodes to test new features may still get
> pre-finalized ones if they run before the heartbeat cycle that finalizes the
> Datanodes.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
