[
https://issues.apache.org/jira/browse/HDDS-668?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17755264#comment-17755264
]
Stephen O'Donnell commented on HDDS-668:
----------------------------------------
At the moment, RM makes decisions on which replicas to copy based on the BCSD.
Deleted transaction ids are not in the picture and I have never heard of anyone
mentioned they need to be.
[~nanda] Do you think this is still relevant? I'm not too familiar with the
block delete path. However I think there are situations where some block
deletes can get missed. Even with maintenance mode, a replica can be offline
temporarily and then come back, and it would still not be replicated. Or a node
is down and comes back online - the "down" replica could have missed block
deletes.
In theory RM could check the deleted transaction ID, but with placement policy,
and then under and mis replicated handling it could start to get complex.
Have there been other thoughts in the past about "lost block deletes" to clear
up orphan blocks?
> Replica Manager should use replica with latest delete transactionID
> -------------------------------------------------------------------
>
> Key: HDDS-668
> URL: https://issues.apache.org/jira/browse/HDDS-668
> Project: Apache Ozone
> Issue Type: Bug
> Reporter: Lokesh Jain
> Assignee: Stephen O'Donnell
> Priority: Major
>
> Currently replica manager does not use delete trasactionID for choosing the
> replica which will be replicated. This Jira aims to store delete
> transactionID for each replica so that replica manager can choose replica
> with latest delete transactionID.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
