[jira] [Commented] (KUDU-3795) Scenarios in heavy-update-compaction-itest fail from time to time

["ASF subversion and git services (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/KUDU-3795?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18110859#comment-18110859 ] 

ASF subversion and git services commented on KUDU-3795:
-------------------------------------------------------

Commit 12f7a3316bac4c18063bb944f621305e8117d087 in kudu's branch refs/heads/master from Ashwani Raina
[ https://gitbox.apache.org/repos/asf?p=kudu.git;h=12f7a3316 ]

KUDU-3795 Fix ENOSPC on dist-test due to compaction tests

HighDeltaVolumeWithFrequentFlushForRowsetCompaction generates large
amounts of on-disk delta data. When multiple test instances run
concurrently on dist-test slaves sharing the same tmpfs, their combined
disk usage can exhaust available space and trigger ENOSPC, causing
misleading failures.
The test is removed from compaction-highmem-test.cc and preserved as a
standalone benchmark in src/kudu/benchmarks/compaction_highmem_benchmark.cc.
The benchmark is built via add_executable (not ADD_KUDU_TEST) and is
therefore never scheduled by the CI runner. It exercises the full
intended workload — kRowsPerRowset=1000 with 64 KiB wide UNDO-delta
mutations — and pins FLAGS_memory_limit_hard_bytes to 20 GiB so that
peak compaction memory naturally crosses the hard limit and emits the
"beyond hard memory limit" warning without any artificial threshold
manipulation.
tablet.h declares CompactionHighMemBenchmark as a friend class so the
benchmark can access the same private test-only tablet APIs used by the
original test (FlushAllDMSForTests, GetRowSetsForTests).

Change-Id: I38df827a72626840a4da49e8a7b9d7f699525555
Reviewed-on: http://gerrit.cloudera.org:8080/24679
Tested-by: Alexey Serbin <al...@apache.org>
Reviewed-by: Alexey Serbin <al...@apache.org>


> Scenarios in heavy-update-compaction-itest fail from time to time
> -----------------------------------------------------------------
>
>                 Key: KUDU-3795
>                 URL: https://issues.apache.org/jira/browse/KUDU-3795
>             Project: Kudu
>          Issue Type: Bug
>          Components: test
>            Reporter: Alexey Serbin
>            Assignee: Ashwani Raina
>            Priority: Major
>             Fix For: 1.19.0
>
>         Attachments: heavy-update-compaction-itest.20260804.txt.xz
>
>
> It seems long run times of the newly introduced scenarios in introduced heavy-update-compaction-itest lead to running out of space.  At least, the following error manifested itself at least once in ASAN build:
> {noformat}
> [ RUN      ] EvictionPolicyTypes/HeavyUpdateCompactionITestWithEvictionPolicy.TestHeavyUpdateCompaction/0
> ...
> F20260805 02:55:17.034668 20926 tablet_replica_mm_ops.cc:287] Check failed: tablet->HasBeenStopped() Unrecoverable flush failure caused by error: IO error: Unable to allocate new delta data writable_block: No directories could be added (1 dirs total, 0 failed, 1 full): IO error: No directories available in 098265df96824e218d5bd2329db27d5b's directory group (1 dirs total, 0 failed, 1 full) (error 28) (error 19)
> {noformat}
> Full log is attached.



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