[
https://issues.apache.org/jira/browse/CASSANDRA-17351?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17490999#comment-17490999
]
Ekaterina Dimitrova edited comment on CASSANDRA-17351 at 2/11/22, 3:55 PM:
---------------------------------------------------------------------------
Alright...this is confusing...
So a few notes:
* It seems in CircleCI J11 executors use the base image, J8 use the one with
dependencies... I don't know of any specific reason for this difference.
But now I ran in Circle with low resources the J11 cqlsh tests and I don't see
the config issues.... Only those we see when running with low resources:
[https://app.circleci.com/pipelines/github/ekaterinadimitrova2/cassandra/1377/workflows/d2a2f507-0264-4fc7-8417-0759c01134b8/jobs/8878]
* Looking [here,
|https://github.com/apache/cassandra-builds/blob/trunk/docker/testing/ubuntu2004_j11.docker]
it seems we setup in the base image the virtual env used by CircleCI. So it
can be we cache that one, but now... I changed to new virtual env and installed
the requirements.txt packages myself in Circle config yesterday so I am lost
about that...
* Looking at the
[cassandra_job_dsl_seed.groovy|https://github.com/apache/cassandra-builds/blob/trunk/jenkins-dsl/cassandra_job_dsl_seed.groovy],
seems to me we use the base image for everything dtest which I assume means -
python upgrade tests, dtests and cqlsh tests.
So seems to me we have two options - I _think_ probably rebuilding the second
image should solve the issue. If we opt in for that one I should probably
change what image we use everywhere in Circle config, to align J8 and J11 if
there are no obstacles and add a remark in the docs we need to rebuild the
second image every time we retag ccm from now on.
Wondering whether we should simply align with the way Jenkins works and it
always works when we update ccm. Or... we should think the other way around
whether we can benefit of moving to the second image in Jenkins and asking
people to rebuild the image on ccm retagging?
was (Author: e.dimitrova):
Alright...this is confusing...
So a few notes:
* It seems J11 executors use the base image, J8 use the one with
dependencies... I don't know of any specific reason for this difference.
But now I ran in Circle with low resources the J11 cqlsh tests and I don't see
the config issues.... Only those we see when running with low resources:
[https://app.circleci.com/pipelines/github/ekaterinadimitrova2/cassandra/1377/workflows/d2a2f507-0264-4fc7-8417-0759c01134b8/jobs/8878]
* Looking [here,
|https://github.com/apache/cassandra-builds/blob/trunk/docker/testing/ubuntu2004_j11.docker]
it seems we setup in the base image the virtual env used by CircleCI. So it
can be we cache that one, but now... I changed to new virtual env and installed
the requirements.txt packages myself in Circle config yesterday so I am lost
about that...
* Looking at the
[cassandra_job_dsl_seed.groovy|https://github.com/apache/cassandra-builds/blob/trunk/jenkins-dsl/cassandra_job_dsl_seed.groovy],
seems to me we use the base image for everything dtest which I assume means -
python upgrade tests, dtests and cqlsh tests.
So seems to me we have two options - I _think_ probably rebuilding the second
image should solve the issue. If we opt in for that one I should probably
change what image we use everywhere in Circle config, to align J8 and J11 if
there are no obstacles and add a remark in the docs we need to rebuild the
second image every time we retag ccm from now on.
Wondering whether we should simply align with the way Jenkins works and it
always works when we update ccm. Or... we should think the other way around
whether we can benefit of moving to the second image in Jenkins and asking
people to rebuild the image on ccm retagging?
> Pip doesn't install ccm every time
> ----------------------------------
>
> Key: CASSANDRA-17351
> URL: https://issues.apache.org/jira/browse/CASSANDRA-17351
> Project: Cassandra
> Issue Type: Task
> Components: CI
> Reporter: Ekaterina Dimitrova
> Priority: Urgent
> Fix For: 4.x
>
>
> In CASSANDRA-16688 we fixed requirements.txt in DTest repo, pip install to
> run without -e for ccm in order to address the moveable tag. That worked for
> some time until last night. I successfully retagged and now ccm is not
> re-installed in Circle CI.
> Jenkins picked stuff though. But it was acting unreliably and weird so
> rerunning now things there. Results pending.
> Now I added -e for ccm in requirements.txt and it worked in CircleCI. Still
> pending results. I don't think this will be permanent solution and I am not
> sure whether it will affect also the previous branches in a way. We need to
> further investigate it and test.
> For now I will ask people to test with -e until we figure it out.
> CC [~mck] , [~bereng] , [~dcapwell] and [~stefan.miklosovic]
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org
For additional commands, e-mail: commits-h...@cassandra.apache.org
