LLNL
unifyfs-stage fails to transfer in Travis; not on real machines
System information
| Type | Version/Name |
|---|---|
| Operating System | |
| OS Version | |
| Architecture | Travis build environment |
| UnifyFS Version | dev on 2020jun18 |
Describe the problem you're observing
The unit test 0700 fails when unifyfs-stage tries to transfer the file out of the unifyfs file system.
Describe how to reproduce the problem
Just put test 0700 in without "expect fail" and it will fail the overall tests.
Include any warning or errors or releveant debugging data
Here's the error message given by unifyfs-stage:
failed to transfer file (src=/tmp/tmp.FaQ7lHYYg4/mount/intermediate.file, dst=/tmp/tmp.FaQ7lHYYg4/stage_destination_0700/destination_0700.file): Bad file descriptor
data transfer failed (Bad file descriptor)
I have run these same tests on summitdev and Cameron has run them on one of the machines at Livermore, and they run fine. Somehow the Travis environment is tripping something up.
I'm creating this issue as a to-do to track this down.
Adding a reference to #511 for quick access to where this work was done.
Also noting that this test was changed to use the test_might_fail directive to essentially label it as "unstable."
There is one stage test currently set as test_might_fail, but I don't recall why this decision was made. Might have been a failure that only showed up on Travis and not locally and never got debugged on Travis.
https://github.com/LLNL/UnifyFS/blob/b5d723dd451585e09fe65c69b4297e027ca00334/t/0700-unifyfs-stage-full.t#L60-L62
Essentially, even if this test fails on Travis, it will show as passing, which is something we should probably adjust in the long run.
Originally posted by @CamStan in https://github.com/LLNL/UnifyFS/issues/614#issuecomment-814292501
9300-unifyfs-stage-isolated.t and 0700-unifyfs-stage-full.t fail on Summit and seem to be set as test_might_fail so that they pass on Travis. This might be a separate issue, and if so, I can create another.
This is being addressed in the discussion of PR 614.
Jenna and I talked yesterday. She'll be creating a separate branch and PR to deal with this once PR 614 has gone in.
As far as we can tell in our testing on Summit, the test_might_fail tag can removed harmlessly.