TST : first pass updating to use travis containers #4355

tacaswell · 2015-04-19T19:05:20Z

This will not work currently, waiting on travis to whitelist needed
packages.

This is waiting on ~~travis-ci/travis-ci#3711 travis-ci/travis-ci#3710 travis-ci/travis-ci#3709 travis-ci/travis-ci#3708 travis-ci/travis-ci#3707~~

sort out humor sans + debian-sid issues

This will not work currently, waiting on travis to whitelist needed packages.

tacaswell · 2015-04-19T19:09:21Z

Apparently the container-based tests start faster and run on better hardware.

tacaswell · 2015-04-20T14:09:27Z

@meatballhat Thank you for the prompt response!

tacaswell · 2015-04-21T01:27:02Z

This seems to have odd intermittent latex-related failures (but enough restarts fixes it).

tacaswell · 2015-04-21T01:31:56Z

The issue above is that to install humor sans we need to enable debian-sid as a source, but if we do that then we get newer versions of a bunch of other stuff, which sets off a nastier dependency cascade which results in failure to install the apt related packages.

@jenshnielsen Does humor sans just get used in building the docs?

OceanWolf · 2015-04-21T01:50:48Z

@tacaswell actually we don't need sid for that, it exists in stable-backports, a.k.a. wheezy-backports and testing a.k.a. jessie.

https://packages.debian.org/search?keywords=humor+sans&searchon=names&suite=all&section=all&sourceid=mozilla-search

tacaswell · 2015-04-21T02:08:37Z

the travis build are ubuntu precise so I suspect even those channels would cause trouble

tacaswell · 2015-04-21T02:10:14Z

Dumping error, restarted test, only on 2.7 this time.


======================================================================
ERROR: matplotlib.tests.test_backend_ps.test_savefig_to_stringio_with_usetex
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/travis/virtualenv/python2.7.9/lib/python2.7/site-packages/nose/case.py", line 197, in runTest
    self.test(*self.arg)
  File "/home/travis/virtualenv/python2.7.9/lib/python2.7/site-packages/matplotlib-1.5.dev1-py2.7-linux-x86_64.egg/matplotlib/testing/decorators.py", line 110, in wrapped_function
    func(*args, **kwargs)
  File "/home/travis/virtualenv/python2.7.9/lib/python2.7/site-packages/matplotlib-1.5.dev1-py2.7-linux-x86_64.egg/matplotlib/testing/decorators.py", line 51, in failer
    result = f(*args, **kwargs)
  File "/home/travis/virtualenv/python2.7.9/lib/python2.7/site-packages/matplotlib-1.5.dev1-py2.7-linux-x86_64.egg/matplotlib/tests/test_backend_ps.py", line 73, in test_savefig_to_stringio_with_usetex
    _test_savefig_to_stringio()
  File "/home/travis/virtualenv/python2.7.9/lib/python2.7/site-packages/matplotlib-1.5.dev1-py2.7-linux-x86_64.egg/matplotlib/tests/test_backend_ps.py", line 36, in _test_savefig_to_stringio
    plt.savefig(buffer, format=format)
  File "/home/travis/virtualenv/python2.7.9/lib/python2.7/site-packages/matplotlib-1.5.dev1-py2.7-linux-x86_64.egg/matplotlib/pyplot.py", line 577, in savefig
    res = fig.savefig(*args, **kwargs)
  File "/home/travis/virtualenv/python2.7.9/lib/python2.7/site-packages/matplotlib-1.5.dev1-py2.7-linux-x86_64.egg/matplotlib/figure.py", line 1509, in savefig
    self.canvas.print_figure(*args, **kwargs)
  File "/home/travis/virtualenv/python2.7.9/lib/python2.7/site-packages/matplotlib-1.5.dev1-py2.7-linux-x86_64.egg/matplotlib/backend_bases.py", line 2225, in print_figure
    **kwargs)
  File "/home/travis/virtualenv/python2.7.9/lib/python2.7/site-packages/matplotlib-1.5.dev1-py2.7-linux-x86_64.egg/matplotlib/backends/backend_ps.py", line 992, in print_ps
    return self._print_ps(outfile, 'ps', *args, **kwargs)
  File "/home/travis/virtualenv/python2.7.9/lib/python2.7/site-packages/matplotlib-1.5.dev1-py2.7-linux-x86_64.egg/matplotlib/backends/backend_ps.py", line 1019, in _print_ps
    **kwargs)
  File "/home/travis/virtualenv/python2.7.9/lib/python2.7/site-packages/matplotlib-1.5.dev1-py2.7-linux-x86_64.egg/matplotlib/backends/backend_ps.py", line 1298, in _print_figure_tex
    self.figure.draw(renderer)
  File "/home/travis/virtualenv/python2.7.9/lib/python2.7/site-packages/matplotlib-1.5.dev1-py2.7-linux-x86_64.egg/matplotlib/artist.py", line 60, in draw_wrapper
    draw(artist, renderer, *args, **kwargs)
  File "/home/travis/virtualenv/python2.7.9/lib/python2.7/site-packages/matplotlib-1.5.dev1-py2.7-linux-x86_64.egg/matplotlib/figure.py", line 1107, in draw
    func(*args)
  File "/home/travis/virtualenv/python2.7.9/lib/python2.7/site-packages/matplotlib-1.5.dev1-py2.7-linux-x86_64.egg/matplotlib/artist.py", line 60, in draw_wrapper
    draw(artist, renderer, *args, **kwargs)
  File "/home/travis/virtualenv/python2.7.9/lib/python2.7/site-packages/matplotlib-1.5.dev1-py2.7-linux-x86_64.egg/matplotlib/axes/_base.py", line 2110, in draw
    a.draw(renderer)
  File "/home/travis/virtualenv/python2.7.9/lib/python2.7/site-packages/matplotlib-1.5.dev1-py2.7-linux-x86_64.egg/matplotlib/artist.py", line 60, in draw_wrapper
    draw(artist, renderer, *args, **kwargs)
  File "/home/travis/virtualenv/python2.7.9/lib/python2.7/site-packages/matplotlib-1.5.dev1-py2.7-linux-x86_64.egg/matplotlib/axis.py", line 1092, in draw
    renderer)
  File "/home/travis/virtualenv/python2.7.9/lib/python2.7/site-packages/matplotlib-1.5.dev1-py2.7-linux-x86_64.egg/matplotlib/axis.py", line 1042, in _get_tick_bboxes
    extent = tick.label1.get_window_extent(renderer)
  File "/home/travis/virtualenv/python2.7.9/lib/python2.7/site-packages/matplotlib-1.5.dev1-py2.7-linux-x86_64.egg/matplotlib/text.py", line 809, in get_window_extent
    bbox, info, descent = self._get_layout(self._renderer)
  File "/home/travis/virtualenv/python2.7.9/lib/python2.7/site-packages/matplotlib-1.5.dev1-py2.7-linux-x86_64.egg/matplotlib/text.py", line 329, in _get_layout
    ismath=ismath)
  File "/home/travis/virtualenv/python2.7.9/lib/python2.7/site-packages/matplotlib-1.5.dev1-py2.7-linux-x86_64.egg/matplotlib/backends/backend_ps.py", line 345, in get_text_width_height_descent
    renderer=self)
  File "/home/travis/virtualenv/python2.7.9/lib/python2.7/site-packages/matplotlib-1.5.dev1-py2.7-linux-x86_64.egg/matplotlib/texmanager.py", line 670, in get_text_width_height_descent
    dvifile = self.make_dvi(tex, fontsize)
  File "/home/travis/virtualenv/python2.7.9/lib/python2.7/site-packages/matplotlib-1.5.dev1-py2.7-linux-x86_64.egg/matplotlib/texmanager.py", line 417, in make_dvi
    report))
RuntimeError: LaTeX was not able to process the following string:
'$0.0$'
Here is the full report generated by LaTeX: 

No latex error report available.

----------------------------------------------------------------------

Travis build machines only have 2 cores

OceanWolf · 2015-04-21T02:17:28Z

Well for debian, backport packages have a higher pin and thus don't install unless you force it to install from backports, or you can try getting it from utopic http://packages.ubuntu.com/search?keywords=humor+sans&searchon=names&suite=all&section=all

Only a font file so you might not even need to download it from the sources. Looking at the website, all the example figures get generated using the font, so yes, to generate those figures you will need the font installed. see http://matplotlib.org/xkcd/examples/api/barchart_demo.html

tacaswell · 2015-04-21T03:06:54Z

I ended up just grabbing the .deb, unpacking and then copying it to the correct place to be found by fc-cache.

OceanWolf · 2015-04-21T03:09:43Z

Well it worked!

jenshnielsen · 2015-04-21T06:29:06Z

@tacaswell Yes it is only used to build some of the xkcd and font examples. I'm 👍 on installing by hand.

jenshnielsen · 2015-04-21T06:32:59Z

Looks like google font has moved to github so we should probably change the Felipa-Regular download location to https://github.com/google/fonts/blob/master/ofl/felipa/Felipa-Regular.ttf?raw=true

(That was the best free Cursive font that I could find)

jenshnielsen · 2015-04-21T08:46:49Z

.travis.yml

@@ -25,8 +44,7 @@ matrix:

 install:
  - pip install -q --use-mirrors nose python-dateutil $NUMPY pep8 pyparsing pillow sphinx!=1.3.0
-  - sudo apt-get update && sudo apt-get -qq install inkscape libav-tools gdb mencoder
-  # We use --no-install-recommends to avoid pulling in additional large latex docs that we don't need
+    # We use --no-install-recommends to avoid pulling in additional large latex docs that we don't need


This comment was referring to apt-get. It should probably be deleted now.

- get Felipa-Regular from github instead of (dying) googlecode - Also remove outdated comment

tacaswell · 2015-04-24T02:57:07Z

@jenshnielsen Thoughts? I suspect that the latex related failures were due to processes stepping on each other's toes.

It is not clear that this is actually an improvement for us.

@meatbalhat, should we be seeing speed ups? Is this the direction travis wants us to be moving in any case?

meatballhat · 2015-04-24T03:08:44Z

@tacaswell The container-based infrastructure has much better resource guarantees than the legacy OpenVZ infrastructure, but it's a much younger offering that's undoubtedly going to require more rounds of tuning. The one area where we've gotten reports of degraded performance is with high file IO test suites such as pip. This is because the container-based setup does not currently have a tmpfs mount at /tmp. We're looking to change this, but it's probably at least a week out.

As for processes stepping on each other's toes, is this the issue discussed above such as with matplotlib.tests.test_backend_ps.test_savefig_to_stringio_with_usetex?

Thank you! 🤘 😻

tacaswell · 2015-04-24T03:17:19Z

@meatballhat Yes, exactly those. I don't know how you deal with logging re-started tests, but we were getting more-or-less random errors of that form. Manually restarting the failing test would generally result in a pass. There still some un-restarted examples of this on the matplotlib repo on my personal (tacaswell) account.

It looks like scaling the number of concurrent processes back to 2 (which iirc is the number of cores available, but I could not find the link last time I looked) seems to have fixed the problem with out degrading the run time.

We are probably very high IO as we write out a lot of test images and the read them back in along with a baseline image and compare them, however I don't think we are currently using /tmp as the files get written back in to a folder the source directory (but we could fix that pretty easily and would probably be a big gain going forward once that is set up).

Please let us know if there is anything else useful we can do from our end, travis is an awesome service 😄 🍻 .

meatballhat · 2015-04-24T04:35:50Z

@tacaswell 2 cores is correct. I'm planning to address the tmpfs situation within the next week, which will hopefully have a decent impact 😃.

jenshnielsen · 2015-04-24T07:55:19Z

@tacaswell It seems likely that we have some kind of race condition. If we can cut down to 2 process without performance degradation that sounds like a sensible solution. I guess we still have some kind of race condition causing the (relatively rare) random failures in font tests.

I still think there is an issue with nose running multiple processes on the mac that I have been meaning to investigate for some time. This prevents me from having any chance of replicating the race condition locally.

I also want to investigate parallelising the docs build with the new feature in Sphinx 1.3 but that might require changes to some of the Sphinx plugins.

jenshnielsen · 2015-04-30T13:43:45Z

@tacaswell Do we want to merge this as is? Even if we don't see any performance improvement at this stage the wait time on the container based solution should still be shorter.

tacaswell · 2015-04-30T14:27:20Z

I am in favor of merging this as-is. It does no harm compute time wise, should reduce variation in run time and I bet it will get rid of the random 'no speed up' test failures. It also seems to be what the travis folks want us to do and I am in favor of staying on their good side.

At some point we should look into putting the saved images into tmpfs rather than a real disk which might get some performance back.

jenshnielsen · 2015-04-30T14:30:16Z

Great I will wait a short while to see if there are any objections and then merge if not

TST : first pass updating to use travis containers

TST : first pass updating to use travis containers

2b4450c

This will not work currently, waiting on travis to whitelist needed packages.

tacaswell added the status: needs review label Apr 19, 2015

TST : update .travis.yaml

8ade0a3

TST : nuke debian-sid and humorsans for testing

afca492

tacaswell mentioned this pull request Apr 20, 2015

APT whitelist request for libgvc6 travis-ci/travis-ci#3720

Closed

TST : actually install all of the latex needed

22a7ee7

tacaswell added this to the next point release milestone Apr 21, 2015

MNT : install humor sans 'by hand'

150320f

TST : turn down number of concurrent jobs

fd0dbbf

Travis build machines only have 2 cores

jenshnielsen reviewed Apr 21, 2015
View reviewed changes

tacaswell added 3 commits April 21, 2015 09:56

TST : make number of concurrent procs an ENV

8599784

TST : squelch gcc output, now at the top of logs

7e741ce

TST : update cursive font download location

31d21df

- get Felipa-Regular from github instead of (dying) googlecode - Also remove outdated comment

jenshnielsen added a commit that referenced this pull request May 1, 2015

Merge pull request #4355 from tacaswell/bld_travis_update

6fd57d7

TST : first pass updating to use travis containers

jenshnielsen merged commit 6fd57d7 into matplotlib:master May 1, 2015

tacaswell removed the status: needs review label May 1, 2015

jenshnielsen mentioned this pull request May 2, 2015

Travis docs fixes #4395

Merged

tacaswell deleted the bld_travis_update branch May 16, 2015 03:25

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

TST : first pass updating to use travis containers #4355

TST : first pass updating to use travis containers #4355

tacaswell commented Apr 19, 2015

tacaswell commented Apr 19, 2015

tacaswell commented Apr 20, 2015

tacaswell commented Apr 21, 2015

tacaswell commented Apr 21, 2015

OceanWolf commented Apr 21, 2015

tacaswell commented Apr 21, 2015

tacaswell commented Apr 21, 2015

OceanWolf commented Apr 21, 2015

tacaswell commented Apr 21, 2015

OceanWolf commented Apr 21, 2015

jenshnielsen commented Apr 21, 2015

jenshnielsen commented Apr 21, 2015

jenshnielsen Apr 21, 2015

tacaswell commented Apr 24, 2015

meatballhat commented Apr 24, 2015

tacaswell commented Apr 24, 2015

meatballhat commented Apr 24, 2015

jenshnielsen commented Apr 24, 2015

jenshnielsen commented Apr 30, 2015

tacaswell commented Apr 30, 2015

jenshnielsen commented Apr 30, 2015

TST : first pass updating to use travis containers #4355

TST : first pass updating to use travis containers #4355

Conversation

tacaswell commented Apr 19, 2015

tacaswell commented Apr 19, 2015

tacaswell commented Apr 20, 2015

tacaswell commented Apr 21, 2015

tacaswell commented Apr 21, 2015

OceanWolf commented Apr 21, 2015

tacaswell commented Apr 21, 2015

tacaswell commented Apr 21, 2015

OceanWolf commented Apr 21, 2015

tacaswell commented Apr 21, 2015

OceanWolf commented Apr 21, 2015

jenshnielsen commented Apr 21, 2015

jenshnielsen commented Apr 21, 2015

jenshnielsen Apr 21, 2015

Choose a reason for hiding this comment

tacaswell commented Apr 24, 2015

meatballhat commented Apr 24, 2015

tacaswell commented Apr 24, 2015

meatballhat commented Apr 24, 2015

jenshnielsen commented Apr 24, 2015

jenshnielsen commented Apr 30, 2015

tacaswell commented Apr 30, 2015

jenshnielsen commented Apr 30, 2015