Math develop failed on Jenkins


#1

Can someone look into why Math develop failed on Jenkins? I can’t access Jenkins (or the Columbia compute cluster) from Kenya. Thanks!


#2

I don’t know what this means.


#3

also… the math develop tests are allowing things by without failing them. The math dependencies test and some compiler warnings.


#4

Oh, I think this is a MPI related test failure I might have already pointed @wds15 to - not sure if he took a look or passed the buck or what happened then.

I thought I fixed making the warnings fail the build… Maybe that was for another repo. Will have to take a look when I have access again.


#5

Thanks.

Btw, I still haven’t caught up on the new Jenkins jobs… it’s hard to have a dashboard tell us when things are broken. Is there an easy way to see that sort of status? (I think the pipelines are much better in general, but the old thing just to just have read and blue at the home screen to tell which ones were broken.)


#6

There is one! I don’t remember the link, haha. If you click around you should be able to find one or two. The old homepage doesn’t really work too well because for some odd reason they are considering the status of a multibranch pipeline job to be the status of not just all of the branches but all of the pull requests.


#7

I apparently made it for myself and I only see it when I log in. I think I just starred the 3 branches I cared about and they show up here:

(My hotel internet can access Jenkins!)


#8

Ups… failures on develop due to MPI. Maybe I missed that; where should I look?


#9

I honestly don’t remember which repo this thread came up in, but what I remember of the issue is that it seems like the MPI tests are passing but failing to write a correct xml file, probably due to parallelism(?)


#10

ah that one. I was hoping this would go away by itself … and I thought it did resolve itseld, but I understand from you that it’s not going away?


#11

I don’t think it seems like the kind of thing to resolve itself? It failed on develop a day or so ago, not sure if that’s the last time we saw it: http://d1m1s1b1.stat.columbia.edu:8080/job/Math%20Pipeline/job/develop/88/


#12

Hmm… ok. Then I will probably have to introduce a gtest main which is MPI specific. This is a bit more tedious as the makefiles need to account for that as well; but having wrestled a lot with make lately that should be doable.

Short background: Currently we rely in the MPI tests on globals being initialized before tests start. Those globals modify the behavior of google test to go along with MPI. Turns out that this is likely a bad choice as globals are ill defined in terms of initialization order as I learned the hard way.


#13

C++ is such a mess. Thank you for looking into this.