Stephen O'Donnell created HDDS-7919:
---------------------------------------
Summary: EC: ECPipelineProvider.createForRead should filter out
dead replicas and sort replicas
Key: HDDS-7919
URL: https://issues.apache.org/jira/browse/HDDS-7919
Project: Apache Ozone
Issue Type: Sub-task
Components: SCM
Reporter: Stephen O'Donnell
ECPipelineProvider.createForRead() is used to get a pipeline to read an EC
block group.
When there are nodes in maintenance, it is possible for there to be replica
details in SCM for dead maintenance nodes, and they should not be returned to
the client as they cannot be read.
Additionally, it is possible for there to be multiple copies for a given
replica index. Eg an IN_SERVICE replica and a DECOMMISSIONED or
DECOMMISSIONING. As things stand, there is no order applied to the replicas,
but the client reading code will tend to use the first replica it loads from
the pipeline.
Therefore we should exclude dead replicas, and then ensure that where there are
multiple copies for a given replica index, HEALTHY (rather than STALE) are
sorted first, and then sort by the SERVICE_STATE, putting IN_SERVICE before
MAINTENANCE etc.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]