[
https://issues.apache.org/jira/browse/HBASE-24280?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17095916#comment-17095916
]
Josh Elser commented on HBASE-24280:
------------------------------------
Yeah, I think you're right, Stack (and Nick). I think we just need to be
consistent across all active branches as to how profiles work, and make sure
that we don't have divergent activation rules. Quite the exercise in test infra
:)
If, in master, we want Hadoop3 to be the default (without requiring an explicit
activation of the {{hadoop-3.0}} profile), I think we have to put custom logic
into the yetus wrapper scripts to account for that. (The other option I see
here would be to duplicate a stage in the Jenkinsfile, but that feels gross).
I'll link a PR here just to try to distill some of this (rather than eat
another day until Istvan/Peter come back tomorrow) –
[https://github.com/apache/hbase/pull/1609] is what I think we need to do. If
we
I think Duo's comment here is worthy of a meta-discussion. He's correct that,
functionally, using {{-Dparam=value}} or {{-Pprofile}} name for hadoop profile
activation are equivalent for the Master branch. This is not true for branch-2
releases. Specifically, we use one system property to create mutually exclusive
profile activations. Meaning, you cannot activate both hadoop-2.0 and
hadoop-3.0, unless you explicitly name them both (and that's your own fault if
you do that). Switching the yetus infra to move back to using
{{-Dhadoop.profile}} does not stop Duo from using a profile explicitly on
Master, but I think it will fix the problem we saw here (where this test fails
because we are activating the hadoop2 and hadoop3 profiles at the same time).
> TestSecureRESTServer started failing in nightlies for Hadoop3
> -------------------------------------------------------------
>
> Key: HBASE-24280
> URL: https://issues.apache.org/jira/browse/HBASE-24280
> Project: HBase
> Issue Type: Bug
> Reporter: Josh Elser
> Assignee: Istvan Toth
> Priority: Major
> Fix For: 2.3.0
>
> Attachments: HBASE-24280.master.001.patch,
> TEST-org.apache.hadoop.hbase.rest.TestSecureRESTServer.xml
>
>
> [~ndimiduk] pointed out that, after this change went in, TestSecureRESTServer
> started failing with Hadoop3 on branch-2.3
> https://builds.apache.org/job/HBase%20Nightly/job/branch-2.3/56/
> Of course, I ran this with 1.8.0_241 and Maven 3.6.33 and it passed :) {{mvn
> clean package -Dtest=TestSecureRESTServer -Dhadoop.profile=3.0
> -DfailIfNoTests=false}}
> FYI [~stoty] in case you can repro a failure and want to dig in. Feel free to
> re-assign.
> It looks like we didn't have a nightly run of branch-2.2 due to docker
> container build issues. Will be interesting to see if it fails there. It did
> not fail the master nightly.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
