hadoop-yarn-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Steve Loughran <ste...@hortonworks.com>
Subject All the jenkins tests are failing
Date Thu, 02 Apr 2015 10:51:14 GMT
Can I draw attention to the fact that within the last week, every single Hadoop jenkins run
has started failing.

Ports in use
https://issues.apache.org/jira/browse/YARN-3433
https://issues.apache.org/jira/browse/HADOOP-11788

NPEs
https://issues.apache.org/jira/browse/HADOOP-11789 NPE in TestCryptoStreamsWithOpensslAesCtrCryptoCodec
https://issues.apache.org/jira/browse/HDFS-8043 MiniDFSCluster

MR tests are showing NPEs during FS creation, with ~190 tests down
https://builds.apache.org/view/H-L/view/Hadoop/job/Hadoop-Mapreduce-trunk-Java8/

We are trying to get a release out the door, trying to get jenkins to test every single patch
to see its not a regression -yet with this failure rate its impossible to do so.

The whole purpose of a Continuous Integration system is to keep on top of changes, to catch
regressions on the individual commits, to preflight check new code, to help identify  exactly
when and where something broke. It's been one of the most tangible changes to Software Engineering
in recent years. I remember when Apache Gump used to be able to build the entire ASF source
tree every night -a full stack regression test of everything from Ant up. I remember reading
Fowler's blog posts on it, in 2001, (see http://www.martinfowler.com/articles/continuousIntegration.html
), installing CruiseControl and loving it so much we gladly dedicated 20 pages to in our Ant
book; using it in projects ourselves to implement some of the first continuous deployment
pipelines.

For all this to work, people have to care about the build, to care about the build passing
its tests, and to keep atop of it when it breaks.

Please look at the test reports and help fix those that are breaking. And then start paying
attention to those Jenkins emails

-Steve

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message