We do not currently have any sles12 s390x machines tagged with "test"

[sxa]

At the moment we run the systemtests on sles12 as this has a suitable version of the libffi library available (https://github.com/AdoptOpenJDK/openjdk-infrastructure/issues/19). At present none of those machines has a a tag of test so I cannot use that tag to run the systemtest jobs. If I leave the tag off I end up with things potentially running on master which doesn't work well at all as it doesn't have make installed. I could use build that would stop the other platforms from using the dedicated test machines, so that's not a sensible solutionest either. For now I've set the jobs to use !hg which knocks out three machines including master and is adequate until we get a sles12/s390x box tagged with test

See also the work item about machines tags: #93

[karianna]

@gdams We just need a new box right?

[gdams]

@karianna getting a "new" box on s390x can be slightly difficult ;) I'm happy to look at splitting the boxes between build and test

[karianna]

@gdams - We can (and should) timeshare the rarer boxes but if we can source more of them that would be good in the long term as we'll be building and testing so many different versions (every 6 months a new release).

[sxa]

Yep obviously happy to share, but at the moment I think there's only the one zlinux box that can run the executables we produce so it could become a bottleneck

[gdams]

@sxa555 are the Ubuntu boxes not going to work then? I was under the impression that we were now building openjdk binaries using Ubuntu 16.04?

[sxa]

Doesn't work out of the box on 16.04, no - it still looks for libffi.so.4 whereas Ubuntu 16.04 has libffi.so.6.0.4. IIRC you can symlink them together and get it working but that's probably not the sensible option.

[smlambert]

Apologies, I had not even seen this issue (prior to creating #85). @tellison if you like, you can close 85 as a duplicate.

[sxa]

Looks like we've got most of the builds moved off SLES12 but https://ci.adoptopenjdk.net/job/openjdk9_build_s390x_linux/ is still running on it - do we know if that's intentional or an oversight? Would be good to move this one forward - at the moment we have three Ubu build machines and this SLES12 one - presumably one of these could be allocated as a dedicated test system now?

[gdams]

I assume that it perhaps has a different compiler level on it? I'll investigate.

[smlambert]

We have created test builds for this platform, but they are currently disabled as there are no s390x machines for test. See this spreadsheet for which test builds are disabled due to no test machines available: https://docs.google.com/spreadsheets/d/1X4CCfvMoCgEavRbvejHrTvPnqj37MB-_C6LB6b8Akkc/edit?usp=sharing

[smlambert]

I have added a note on tests/309, that we are unable to run system or openjdk_regression tests against the openjdk8-openj9 release 0.8 candidate build for s390x_linux at this time as we do not have a test machine for this platform.

[gdams]

@smlambert you do now :) https://ci.adoptopenjdk.net/computer/test-marist-ubuntu1604-s390x-1/

[smlambert]

Awesome! thanks @gdams

[gdams]

closing issue for now

[sxa]

The OpenJ9 are happy on there although our latest released HotSpot build does not. Somewhat worrying it would appear that the HotSpot (Or zero!) VM crashes when you try to run it on that Ubuntu machine at the moment. That'll probably need another issue raised ...