[
https://issues.apache.org/jira/browse/HDDS-9019?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Siddhant Sangwan updated HDDS-9019:
-----------------------------------
Description:
{{ECUnderReplicationHandler}} checks if current replica datanodes + potential
replica datanodes satisfy placement policy. If they don't, the targets are
rejected and recovery fails. While this check makes some sense, it can block
container recovery. Since under replication is more serious than mis
replication, we should let a new replica be reconstructed even if it doesn't
satisfy placement policy.
Also, we should rely on the policy to suggest a node that satisfies the policy
- if there is none, the policy will fallback and try to suggest a node on a
rack where the container is already present.
was:
{{ECUnderReplicationHandler}} checks if current replica datanodes + potential
replica datanodes satisfies placement policy. If they don't, the targets are
rejected and recovery fails. While this check makes sense, it can block
container recovery. Since under replication is more serious than mis
replication, we should let a new replica be reconstructed even if it doesn't
satisfy placement policy.
Also, we should rely on the policy to suggest a node that satisfies the policy
- if there is none, the policy will fallback and try to suggest a node on a
rack where the container is already present.
> ECUnderReplicationHandler rejects a potential target if it doesn't satisfy
> placement policy
> -------------------------------------------------------------------------------------------
>
> Key: HDDS-9019
> URL: https://issues.apache.org/jira/browse/HDDS-9019
> Project: Apache Ozone
> Issue Type: Bug
> Components: ECOfflineRecovery, SCM
> Reporter: Siddhant Sangwan
> Assignee: Siddhant Sangwan
> Priority: Major
>
> {{ECUnderReplicationHandler}} checks if current replica datanodes + potential
> replica datanodes satisfy placement policy. If they don't, the targets are
> rejected and recovery fails. While this check makes some sense, it can block
> container recovery. Since under replication is more serious than mis
> replication, we should let a new replica be reconstructed even if it doesn't
> satisfy placement policy.
> Also, we should rely on the policy to suggest a node that satisfies the
> policy - if there is none, the policy will fallback and try to suggest a node
> on a rack where the container is already present.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
