[
https://issues.apache.org/jira/browse/PHOENIX-8011?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Andrew Kyle Purtell updated PHOENIX-8011:
-----------------------------------------
Fix Version/s: 5.4.0
> Classify batch origin once and gate the HA write path on the sync-replication
> master switch
> -------------------------------------------------------------------------------------------
>
> Key: PHOENIX-8011
> URL: https://issues.apache.org/jira/browse/PHOENIX-8011
> Project: Phoenix
> Issue Type: Improvement
> Reporter: Tanuj Khurana
> Assignee: Tanuj Khurana
> Priority: Major
> Fix For: 5.4.0
>
>
> The synchronous-replication write path in IndexRegionObserver.preBatchMutate
> re-derived a batch's HA disposition (from _HAGroupName / REPLICATED_MUTATION
> / cluster ids) at several independent sites, and the cluster-role
> mutation-block / staleness gate ran
> whenever _HAGroupName was present — independent of whether the feature was
> actually enabled. This made the rollout window unsafe: clients are upgraded
> first and already attach _HAGroupName before the HA group exists or the
> feature is turned on, so the gate would
> try to resolve a not-yet-created HA group.
> This change classifies each batch's origin exactly once and treats
> SYNCHRONOUS_REPLICATION_ENABLED as the master switch for all HA write-path
> semantics. It also adds commit-id observability to the replication log.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
