Andrew in infra made us a label that covers all the hosts save H2. I've updated all the nightly builds to use it.
(specifying a label expression as we do in precommit doesn't work on matrix builds because the & and | from the expression end up in the filesystem path) On Fri, Jan 22, 2016 at 3:08 PM, Sean Busbey <[email protected]> wrote: > we should probably ensure the earlier branch builds also exclude H2. > > I'll leave myself a note to look at it this evening. If anyone gets to it > before then, please update here. > > On Fri, Jan 22, 2016 at 1:33 PM, Stack <[email protected]> wrote: > >> Related to the below, I just changed the trunk matrix build job to exclude >> H2 from the build roster (with Sean's help); it seems to be responsible for >> this failure type -- *Caused by: java.lang.IndexOutOfBoundsException: >> Index: 0, Size: 0* -- in particular. Here is recent example: >> >> https://builds.apache.org/view/H-L/view/HBase/job/HBase-Trunk_matrix/652/jdk=latest1.7,label=Hadoop/console >> >> Lets see if it helps. >> >> While I have your attention, see the nice checkstyle and findbug graph >> trajectories here: >> https://builds.apache.org/view/H-L/view/HBase/job/HBase-Trunk_matrix/ >> >> St.Ack >> >> >> On Tue, Jan 19, 2016 at 7:02 PM, Sean Busbey <[email protected]> wrote: >> >> > On Tue, Jan 19, 2016 at 11:48 AM, Stack <[email protected]> wrote: >> > >> > > On Tue, Jan 19, 2016 at 5:46 AM, Sean Busbey <[email protected]> >> wrote: >> > > >> > > > We could start forcing a clean repository on every build (though this >> > > > seems heavy handed). >> > > > >> > > > IIRC, we ran into this ages ago and it was one particular dependency. >> > > > Presuming we can track down what that was, we could add some >> pre-build >> > > > work that verifies a known good copy of that dependency is present. >> > > > >> > > > For now, I've blacklisted the H2 build host again, since that's the >> > > > only host this has been happening on. We added it back last week so >> > > > that Jon could test a fix from infra. >> > > > >> > > > >> > > Thanks Sean. >> > > >> > > Yeah, clean repo each time would be OTT. >> > > >> > > If you have pointers on how to find the bad dependency, I'll dig. >> > > >> > > Of course, all builds fine locally. >> > > >> > > St.Ack >> > > >> > > >> > > >> > Relevant bits from our old precommit build (lucky we kept it! ;) ) >> > >> > >> > ---- >> > ... >> > # holding place for local repo >> > if [ -d "${WORKSPACE}/maven_repo" ]; then >> > echo "removing hbase artifacts from prior runs." >> > rm -rf "${WORKSPACE}/maven_repo/org/apache/hbase" >> > echo "removing javax.inject in case we got a bad dependency" >> > rm -rf "${WORKSPACE}/maven_repo/javax/inject" >> > # uncomment to list entire contents of repo >> > # find "${WORKSPACE}/maven_repo" >> > else >> > mkdir "${WORKSPACE}/maven_repo" >> > fi >> > >> > ... >> > >> > if [ "$?" -ne "0" ]; then >> > echo "test patch failed, checking javax.inject dependency" >> > find "${WORKSPACE}/maven_repo/javax/inject" >> > cat >> > "${WORKSPACE}/maven_repo/javax/inject/javax.inject/1/javax.inject-1.pom" >> > exit 1 >> > fi >> > --- >> > >> > So it looks like javax:inject was the culprit before. A first step would >> be >> > to remove it from our local repos before running; that'll require looking >> > up how Yetus names the branch-specific maven repos. A better step would >> be >> > to manually install it from a known good location so we can avoid >> whatever >> > nonsense is happening on H2. >> > >> > > > > -- > Sean
