[
https://issues.apache.org/jira/browse/HDDS-14416?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18053830#comment-18053830
]
Ashish Kumar commented on HDDS-14416:
-------------------------------------
[~ivanandika] The case in HDDS-13928 is more of orphan blocks inside CLOSED
containers. And these blocks are orphan because of missing delete commands.
This specific scenario is handled in container reconciliation. DN replica will
keep track of deletes and during reconciliation if any DN is having extra block
will get deleted.
[HDDS-14416|https://issues.apache.org/jira/browse/HDDS-14416] deals with when
container has already DELETED and DN don't have any delete information, in this
case reconciliation can't help.
> Disk Space Not Fully Reclaimed After Key Deletion
> -------------------------------------------------
>
> Key: HDDS-14416
> URL: https://issues.apache.org/jira/browse/HDDS-14416
> Project: Apache Ozone
> Issue Type: Bug
> Reporter: Ashish Kumar
> Assignee: Ashish Kumar
> Priority: Major
> Labels: pull-request-available
> Fix For: 2.2.0
>
>
> *Scenario:* When any particular DN is down for long time. During this period
> RM replicates data from down DN to some other DNs.
> If data deletion happens during this period, deletion happens only from
> available DN. And also if container becomes empty, container gets deleted.
> If DN that were previously down were brought back online. This DN starts
> reporting replicas to SCM.
> Due to a recent [change|https://issues.apache.org/jira/browse/HDDS-12421],
> SCM accepts replica reports for containers that were already in the DELETED
> state. As a result, containers previously marked as DELETED were transitioned
> back to the CLOSED state.
> Replication Manager then treated these containers as valid and replicated
> them to other DataNodes.
> This results in orphan containers and causing extra space utilization.
> In case of Ratis, to minimize orphan containers we can safely delete replica
> in first case from the below table.
> ||Container State(SCM)||Replica State (DN)||Replica BCSID vs Container
> BCSID||Action (SCM)||
> |DELETED|CLOSED|Replica BCSID <= Container BCSID|Delete Replica|
> |DELETED|CLOSED|Replica BCSID > Container BCSID|Do Not Delete Replica, keep
> current behaviour|
> |DELETED|State other than CLOSED|Replica BCSID <= Container BCSID|Do Not
> Delete Replica, keep current behaviour|
> |DELETED|State other than CLOSED|Replica BCSID > Container BCSID|Do Not
> Delete Replica, keep current behaviour|
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
