You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@hudi.apache.org by "sivabalan narayanan (Jira)" <ji...@apache.org> on 2023/01/03 03:05:00 UTC
[jira] [Updated] (HUDI-5433) Fix the way we deduce the pending instants for MDT writes
[ https://issues.apache.org/jira/browse/HUDI-5433?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
sivabalan narayanan updated HUDI-5433:
--------------------------------------
Reviewers: Raymond Xu
> Fix the way we deduce the pending instants for MDT writes
> ---------------------------------------------------------
>
> Key: HUDI-5433
> URL: https://issues.apache.org/jira/browse/HUDI-5433
> Project: Apache Hudi
> Issue Type: Bug
> Components: metadata
> Reporter: sivabalan narayanan
> Assignee: sivabalan narayanan
> Priority: Blocker
> Labels: pull-request-available
> Fix For: 0.13.0
>
>
> we trigger compaction in MDT, only when there are no pending inflights apart from the one thats currently updating the MDT. So we use below code snippet for it.
>
> {code:java}
> List<HoodieInstant> pendingInstants = dataMetaClient.reloadActiveTimeline().filterInflightsAndRequested()
> .findInstantsBefore(instantTime).getInstants(); {code}
> As you could see, we use "findInstantsBefore" which could not yield right results at all times.
>
> So, we need to find all inflight instants and see if there are any except the current commit thats updating the MDT. If there are any, we should defer compaction.
> Impact:
> writes to MDT might fail if there was any missed inflight and later it was rolledback. Users have to disable MDT and make progress.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)