Speed up server startup by running asset compilation in parallel

[davestgermain]

On my machine, the devstack server starts up 10 seconds faster.

@singingwolfboy @wedaly

[singingwolfboy]

Seems like there's no way to use keyword arguments with the solution you've proposed. Also, under what circumstances do you expect this to run with concurrent.futures.ThreadPoolExecutor vs threading.Thread?

[davestgermain]

@singingwolfboy the functions didn't need kwargs, but I can adapt it if you want.
The function would run with concurrent.futures if we decide to add that library (it's available as a backport), or when we move to python 3 someday...

[singingwolfboy]

@davestgermain yeah, would you mind adding kwargs support? I want it to be easy to use this run_threaded function for other things, and I think this will help make it easier.

[singingwolfboy]

max_workers=4 seems a fairly arbitrary value. Is there a way we can select a value more intelligently? If not, I'm OK with just setting it to 4, but it's worth asking.

[davestgermain]

4 seemed reasonable given the number of cpus and the fact that most tasks will be IO bound. I believe 4 is the default, anyway.

[singingwolfboy]

Can we do something like:

try:
    import multiprocessing
    num_cpus = multiprocessing.cpu_count()
except ImportError:
    # reasonable default
    num_cpus = 4

with ThreadPoolExecutor(max_workers=num_cpus) as executor:
    # etc

[davestgermain]

@singingwolfboy added kwargs support and cpu_count (multiprocessing is always available in python 2.7)

[davestgermain]

@singingwolfboy Do have any guesses why this change would cause the acceptance test failures? Could it be that there's an order dependency when compiling coffeescript or sass? I made a change assuming that was the case, but the tests still failed.

[singingwolfboy]

Not sure. Maybe the tests are starting up before all the assets are finished being compiled?

[singingwolfboy]

This pull request needs to be rebased. Also, there's some discussion about moving from paver to invoke (see #3444) so perhaps this PR is no longer useful?

[davestgermain]

I never could figure out why the tests were failing. should I remove this PR?

[singingwolfboy]

@davestgermain that seems reasonable