Matplotlib lockerror

I've been doing some TBS style runs (hundreds of short runs) to try to stress test the code.

I found about 10% of jobs failed with the following trace. It looks like there are too many things trying to access the matplotilb installation.

Any thoughts?

@paul-lasky @gregory.ashton @moritz.huebner?

File "/home/colm.talbot/virtualenvironents/py3.4/lib/python3.4/site-packages/matplotlib/texmanager.py",
line 406, in with Locked(self.texcache):
File "/home/colm.talbot/virtualenvironents/py3.4/lib/python3.4/site-packages/matplotlib/cbook.py",
raise self.TimeoutError(err_str)
matplotlib.cbook.TimeoutError: LOCKERROR: matplotlib is trying to acquire the lock
'/home/colm.talbot/.cache/matplotlib/tex.cache/.matplotlib_lock-*' and has failed.
This maybe due to any other process holding this lock.
If you are sure no other matplotlib process is running try removing these folders and trying again.

This seems to be breaking when dynesty tries to make the trace plots.

It does this before saving the h5 file, so this means that the result is just thrown away.

Can we remove https://git.ligo.org/Monash/tupak/blob/master/tupak/sampler.py#L330?

changed the description

Hmm, that's unfortunate. Sorry about that @colm.talbot, - I added the trace plots.

The reason I made them just run by default is that it relies on the result object made by dynesty which we don't store as it creates pickle errors. I see a few options

1. Remove the traceplot entirely
2. Make the creation of traceplots optional by some keyword arguments
3. Add a command line arg which stop the creation of all plots - useful if running many simulations and you forget to remove plot_corner or something

Closing as this is solved in 733a52e6. This adds a plot=False argument to run_sampler, so only is this is set to true will the traceplots be generated.

closed