[
https://issues.apache.org/jira/browse/HDDS-15561?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18093765#comment-18093765
]
Attila Doroszlai commented on HDDS-15561:
-----------------------------------------
Thanks [~chungen] for the details. I think the inherent problem with
MiniOzoneChaosCluster is that it's a single process, which runs both the test
and several (possibly) faulty components. A runaway component may severely
affect the test controller. It is better to separate these, even better to
separate components. I prefer Docker Compose (or Kubernetes) for the cluster,
and running tests / injecting faults externally.
Acceptance tests use Byteman for some fault injection tests, see:
-
https://github.com/apache/ozone/blob/dbcd9b36d7796dd779e611c91af05e33d13774e7/hadoop-ozone/dist/src/main/compose/ozonesecure-ha/test.sh#L56-L57
-
https://github.com/apache/ozone/tree/master/hadoop-ozone/dist/src/main/smoketest/ozone-fi
> Introduce a linearizability checker to validate Ozone
> -----------------------------------------------------
>
> Key: HDDS-15561
> URL: https://issues.apache.org/jira/browse/HDDS-15561
> Project: Apache Ozone
> Issue Type: Test
> Reporter: Chung-En Lee
> Assignee: Chung-En Lee
> Priority: Major
>
> To catch subtle consistency regressions like {{HDDS-15052}} without the
> massive engineering overhead of full DST (Deterministic Simulation Testing),
> we propose introducing a post-facto *Linearizability Checker* (e.g.,
> Porcupine, Elle, or Java alternatives).
> By acting as a black-box oracle, the checker will analyze concurrent client
> operation histories generated by our existing {{MiniOzoneCluster or docker
> cluster}} to verify strict correctness with zero core code invasiveness.
> Importantly, this checker will also serve as a foundational building block
> for future DST efforts, as any deterministic simulation framework will still
> require such an oracle to validate its execution histories.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
