On Mon, Jul 7, 2014 at 4:50 PM, Toshio ITO <toshio9....@toshiba.co.jp> wrote: > Hi Roman, > > I previously reported some cases where Giraph unit tests failed. > > http://mail-archives.apache.org/mod_mbox/giraph-user/201406.mbox/%3c87a990dkni.wl%25toshio9....@toshiba.co.jp%3E
This thread talks about hadoop_0.20.203 profile. I am not sure this version of Hadoop gets a lot of attention in Giraph community. Personally, I'd definitely not treat any failures in that profile as release blockers. > http://mail-archives.apache.org/mod_mbox/giraph-user/201407.mbox/%3C8761jgqjec.wl%25toshio9.ito%40toshiba.co.jp%3E This one is more interesting. I can't reproduce your failures in Rexster I/O Format in my environment. As for -Dprop.mapred.job.tracker=localhost:54311 -- I've never seen anybody running Giraph unit tests that way. You're right that in theory it should work and it would be useful for us to understand why it fails. I may be able to look at it, but since it happens to be a pretty non-orthodox way of running Unit test, I don't think it'll be a release blocker all by itself. > Because it seems I'm almost the only one in the user mailing list > who cares about the unit tests I think all of us do, but the thing is -- we run them as pure unit tests -- you run then as combination of unit/system tests. If you can make them work both ways that would be appreciated regardless of whether your fixes end up in 1.1.0 or not. > I just wonder whether it is normal (or expected) for the unit tests > to fail at this stage of development (release-1.1.0-RC0). Pure unit tests are definitely expected to pass. They do pass on our Jenkins, hence my suspicion that something's is wrong with your env. Thanks, Roman.