2 test failures in Python 3.4 in parallel group

[andrewclegg]

Not sure where you are on Python 3.4, but I'm getting a couple of test failures. This is on Ubuntu 12.04, and only occurs when I kick off the tests from within an IPython notebook -- even though they're in a totally separate Python process started via subprocess.call.

Steps:

Clone Snake Charmer: https://github.com/andrewclegg/snake-charmer.git

Checkout this revision (so as to avoid any confounding factors): 7d5e90e8b8816b0a28b037cf3094cdf4f24bf19f

Start up a VM: vagrant up

... wait about an hour ...

Open notebook browser on VM: http://localhost:8834/tree

Open "Snake Charmer QA" notebook

Run the first code cell (under "Initialization"), then the second (under "Serial tests")

The first test to complete should be

python3.4 `which iptest3` --all -j

with exit code 1.

When it completes, check out log/charmed34/test_output/ipython.out (inside your snake-charmer directory)

I see these two errors:

https://gist.github.com/andrewclegg/823d005491850d7f66cb

One of them is an instance of the error described here -- #5291 -- the other is in test_elapsed_single and looks maybe unrelated?

The interesting thing is that when I ssh into the VM and run the tests from bash, using the exact same command line I passed to subprocess.call, those tests pass.

[andrewclegg]

Another data point:

The scenario I described above, and the test transcript I included, were on a Snake Charmer VM with 16 vCPUs. That is, I did this

export CHARMER_CPUS=16

before spinning it up. (Also CHARMER_RAM=8000.) I didn't think to include this info in the OP as it didn't seem relevant.

I recently retried on a fresh VM with only 12 vCPUs, and all of the tests passed -- not as much as a warning anywhere.

From my point of view I'm happy to consider this closed, as my top priority is getting tests passing in Snake Charmer, which this has achieved :-) ... But if you would rather leave it open, since it suggests something a bit wrong in IPython.parallel, no complaints from me.

[minrk]

The first failure is a real bug that has been eluding me for months. The second failure is probably just an issue in the test, not properly handling slow or overworked test bots.