For the future reader:
As pointed out by Shixiong on the PR, we had to filter the new modules
from the MiMa builds in
https://github.com/apache/spark/blob/76fbe121ab2e829f75389c6fd466afcf2ec3b9d1/project/SparkBuild.scala#L335
Thanks for the feedback Shixiong, the PR build is green now.
Cheers,
Martin
On 05.06.19 12:46, Martin Junghanns wrote:
Hi all,
While introducing new modules for Spark Graph
(https://github.com/apache/spark/pull/24490),
builds started to turn red because of failing MiMa (binary
compatibility) tests.
We had one successful build, however, all subsequent builds fail:
Successful run:
https://github.com/apache/spark/pull/24490#issuecomment-498075761
First failed run:
https://github.com/apache/spark/pull/24490#issuecomment-498104960
Both builds run on the same commit:
https://github.com/apache/spark/commit/bf7db46d7dc548ffecb84d5696281951a46e5ae7
The build log contains:
[error] running
/home/jenkins/workspace/SparkPullRequestBuilder/dev/mima -Phadoop-2.7
-Pkubernetes -Phive-thriftserver -Pkinesis-asl -Pyarn
-Pspark-ganglia-lgpl -Phive -Pmesos ; received return code 1
I tried to reproduce this locally, but got return code 0 without any
output when running
'dev/mima'
and
'dev/mima -Phadoop-2.7 -Pkubernetes -Phive-thriftserver -Pkinesis-asl
-Pyarn -Pspark-ganglia-lgpl -Phive -Pmesos'
Any ideas what could cause CI to change its behaviour between builds?
Are we missing something in the way we introduced the modules? Any
help would be appreciated.
Thanks,
Martin
