[
https://issues.apache.org/jira/browse/HBASE-29838?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18054208#comment-18054208
]
Nick Dimiduk commented on HBASE-29838:
--------------------------------------
The basic structure is there and working. However, our GHA runners are
hard-capped at 6h time limit. I have asked Infra to provision a self-hosted
runner.
However, I'm thinking that we can partition up the test suite into 4 chunks and
run them in parallel. According to the last completely successful nightly build
of master, the total suit takes about 15 hours, so 4 groups of 4h each is what
we'll target. This gives ample room for new, uncategorized tests to run as
well. The parallelism should be a lot nicer experience for devs as well -- we
can significantly tighten the feedback loop on a contribution.
I think we'll need a maintenance job that parses old run logs and tracks the
runtime information. It'll work a bit like the find flakeys job, as a kind of
CI maintenance. We can host this on the nightlies server or in a new CI branch
in the repository.
[~zhangduo], [~psomogyi], [~wchevreuil], [~apurtell], [~nihaljain] WDYT?
> Run Hadoop Check as a GitHub Action
> -----------------------------------
>
> Key: HBASE-29838
> URL: https://issues.apache.org/jira/browse/HBASE-29838
> Project: HBase
> Issue Type: Task
> Components: build
> Reporter: Nick Dimiduk
> Assignee: Nick Dimiduk
> Priority: Major
> Labels: pull-request-available
>
> Following on the heels of HBASE-29787. The other thing our PR pre-commit does
> is run the full build and applicable tests with each supported JDK/Hadoop
> version combination. Implement these as a workflow as well.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
