[
https://issues.apache.org/jira/browse/HDDS-6384?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Stephen O'Donnell updated HDDS-6384:
------------------------------------
Status: Patch Available (was: In Progress)
> EC: Ensure EC container usage is updated correctly when handling reports
> ------------------------------------------------------------------------
>
> Key: HDDS-6384
> URL: https://issues.apache.org/jira/browse/HDDS-6384
> Project: Apache Ozone
> Issue Type: Sub-task
> Components: SCM
> Reporter: Stephen O'Donnell
> Assignee: Stephen O'Donnell
> Priority: Major
> Labels: pull-request-available
>
> The current container report handling logic in AbstractContainerReportHandler
> does update the container usage and key count correctly in SCM for EC
> containers. The current logic says:
> {code:java}
> for (ContainerReplica r : otherReplicas) {
> // Open containers are generally growing in key count and size, the
> // overall size should be the min of all reported replicas.
> if (containerInfo.getState().equals(HddsProtos.LifeCycleState.OPEN)) {
> usedBytes = Math.min(usedBytes, r.getBytesUsed());
> keyCount = Math.min(keyCount, r.getKeyCount());
> } else {
> // Containers which are not open can only shrink in size, so use the
> // largest values reported.
> usedBytes = Math.max(usedBytes, r.getBytesUsed());
> keyCount = Math.max(keyCount, r.getKeyCount());
> }
> } {code}
> For EC Containers, the bytes used will be different across all the replicas
> due to partial stripes. The first data replica and the parity replicas should
> have the max size, but other data replicas are likely to be smaller.
> Therefore using the min as above will not work.
> We should either use the max value for open containers, or the min of the
> first data and parity group, and ignore the usage in the other replicas.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
