You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@ozone.apache.org by "Siddhant Sangwan (Jira)" <ji...@apache.org> on 2024/01/30 06:46:00 UTC

[jira] [Commented] (HDDS-9280) Quasi-closed container with unhealthy replicas may remain under-replicated in 4 node cluster

    [ https://issues.apache.org/jira/browse/HDDS-9280?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17812168#comment-17812168 ] 

Siddhant Sangwan commented on HDDS-9280:
----------------------------------------

The attached design document "QUASI_CLOSED Container in a Small Cluster" explains the problem and presents various ways to solve it in Replication Manager. After discussing the doc, we decided it's better to have a long term fix that reconciles a container's replicas in the Datanode to produce one healthy, closed replica - HDDS-10239. We can close this one as a duplicate.

> Quasi-closed container with unhealthy replicas may remain under-replicated in 4 node cluster
> --------------------------------------------------------------------------------------------
>
>                 Key: HDDS-9280
>                 URL: https://issues.apache.org/jira/browse/HDDS-9280
>             Project: Apache Ozone
>          Issue Type: Bug
>          Components: SCM
>            Reporter: Ethan Rose
>            Assignee: Siddhant Sangwan
>            Priority: Major
>         Attachments: QUASI_CLOSED Containers in a Small Cluster.pdf, Quasi_dilemma.png
>
>
> This is a follow-up based on discussion around HDDS-8536 in this [PR|https://github.com/apache/ozone/pull/5255#discussion_r1322165405]. Replication manager will try to keep all replicas with unique origin node IDs for quasi-closed containers that cannot be closed, while fully replicating the healthy replicas. This could require up to 5 replicas to satisfy, for example:
> SCM state: Quasi-closed
> Replica states: quasi-closed(origin 1) quasi-closed(origin 1) quasi-closed(origin 1) unhealthy(origin 2) unhealthy(origin 3)
> In a 4 node cluster, or a 5 node cluster trying to decommission a node, this would remain under-replicated or get stuck. In this case, we need to choose how to proceed.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@ozone.apache.org
For additional commands, e-mail: issues-help@ozone.apache.org