[
https://issues.apache.org/jira/browse/HDDS-4738?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Siyao Meng resolved HDDS-4738.
------------------------------
Target Version/s: (was: 2.3.0)
Resolution: Won't Fix
Closing as Won't Fix. Reviewing against current master, both intents in the
description are already covered, and the literal "mark the node DEAD in SCM"
approach isn't the direction the project took.
*No new allocation to a node with no healthy volumes: done.*
{{SCMNodeManager.pipelineLimit()}} returns 0 when {{getHealthyVolumeCount() ==
0}} (HDDS-3810), and container placement's space check skips {{failed=true}}
volumes, so such a node gets no new pipelines or containers.
*Marking DEAD: handled differently (and better).* SCM doesn't change node
health based on disk state. A datanode that loses its last tolerated volume
self-shuts down ({{{}MutableVolumeSet.hasEnoughVolumes(){}}} →
{{{}handleFatalVolumeFailures(){}}}, HDDS-5219) and is then marked STALE/DEAD
via heartbeat timeout. A node with at least one healthy volume keeps serving
its good volumes while excluded from new allocation. Force-marking the whole
node DEAD on disk failure would needlessly re-replicate data still readable on
healthy volumes.
Feel free to reopen if a proactive SCM-side reaction (vs. the current
heartbeat-timeout path) or metadata-volume failure tracking proves necessary.
> Mark Datanode with no heathy data or metadata disks as dead in SCM
> ------------------------------------------------------------------
>
> Key: HDDS-4738
> URL: https://issues.apache.org/jira/browse/HDDS-4738
> Project: Apache Ozone
> Issue Type: Sub-task
> Components: SCM
> Reporter: Shashikant Banerjee
> Assignee: Shashikant Banerjee
> Priority: Major
>
> If a datanode reports no healthy data disks or metadata disks, it should not
> participate in any new pipeline allocation strategy. The idea here is to mark
> such nodes as DEAD.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
