[
https://issues.apache.org/jira/browse/PHOENIX-7887?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Palash Chauhan updated PHOENIX-7887:
------------------------------------
Fix Version/s: (was: 5.4.0)
(was: 5.3.2)
> Add metrics for parent-region replay to enable post-split visibility
> --------------------------------------------------------------------
>
> Key: PHOENIX-7887
> URL: https://issues.apache.org/jira/browse/PHOENIX-7887
> Project: Phoenix
> Issue Type: Sub-task
> Affects Versions: 5.3.1
> Reporter: Palash Chauhan
> Assignee: Palash Chauhan
> Priority: Major
>
> h4. Background
> When an EC-indexed region splits or merges, the new child region's
> {{IndexCDCConsumer}} calls {{replayAndCompleteParentRegions(...)}} →
> {{processPartitionToCompletion(parentId, ...)}} for each ancestor partition
> before entering its main loop. This phase replays the parent's remaining CDC
> events into the EC index tables and marks the parent's tracker entry COMPLETE.
> After PHOENIX-7884, {{cdcIndexUpdateLag}} correctly grows during this window
> (the child region's own writes are accumulating unprocessed and the EC index
> is genuinely stale). However, this is _expected_ operational behavior, not an
> incident — and there's currently no signal that lets ops distinguish
> "consumer is stuck and needs paging" from "consumer is mid-replay after a
> recent split."
> h4. Problem
> There are no metrics that surface parent-replay activity.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
