You are viewing a plain text version of this content. The canonical link for it is here.
Posted to oak-issues@jackrabbit.apache.org by "Rishabh Daim (Jira)" <ji...@apache.org> on 2026/10/08 07:26:00 UTC

[jira] [Resolved] (OAK-12444) Fix reference-retention timing in DataStoreCommandTest

     [ https://issues.apache.org/jira/browse/OAK-12444?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Rishabh Daim resolved OAK-12444.
--------------------------------
    Fix Version/s: 2.8.0
       Resolution: Fixed

> Fix reference-retention timing in DataStoreCommandTest
> ------------------------------------------------------
>
>                 Key: OAK-12444
>                 URL: https://issues.apache.org/jira/browse/OAK-12444
>             Project: Jackrabbit Oak
>          Issue Type: Bug
>            Reporter: Rishabh Daim
>            Assignee: Rishabh Daim
>            Priority: Major
>             Fix For: 2.8.0
>
>
> DataStoreCommandTest.gcSweepRefsOld intermittently fails because GC skips sweeping and does not create avail- output. The latest failure used SegmentTar with FileDataStore: [CI failure|https://github.com/apache/jackrabbit-oak/actions/runs/37638190443/job/112849844984?pr=3195].
> The test assumes a new reference record is already old enough for zero-retention GC. Production correctly requires lastModified < cutoff; equal timestamps reject the sweep. The earlier setup delay happens before marking and cannot guarantee this condition.
> Fix the test by completing a mark-only pass and waiting past its stored reference timestamp before sweeping. Keep the output and blob-deletion assertions and the recent-reference test unchanged. No production changes are needed.
> On current trunk, forcing the GC cutoff to the newest reference timestamp reproduces the exact missing avail- failure on both FileDataStore fixtures. The proposed test setup passes the same controlled experiment.
> The test originated in OAK-9040 (commit 3594f92a47020329f137330f03ee16ee8e4a9572). OAK-12122 reported the same failure before S3Mock coverage; its fix added a deterministic random seed and diagnostics but did not address reference age. Slow real AWS S3 requests can hide this timing assumption.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)