WIP: load testing and gunicorn support

[chicco785]

No description provided.

[c0c0n3]

Just for future reference, the "cleaned up" version of this PR sits in the following commits of the load-test and gunicorn branches

• port notify load test from WIP: load testing and gunicorn support #354. (a13a8f5) This commit basically lifts the load test as is from this PR and replays it on master.
• make sure load test uses latest ql code and removes containers to avoid undeletable dangling images. (960a148) This is a small fix to the load test to avoid problems w/ dangling docker images.
• port gunicorn setup from WIP: load testing and gunicorn support #354. (6b84e35) This commit basically lifts the Gunicorn setup as is from this PR and replays it on master.
• get rid of supervisord. (4c6f1b1) This is a small improvement over the previous commit since supervisord isn't actually needed. In fact, it'd only monitor the Gunicorn master process which already takes care of monitoring child worker processes. Also Quantum Leap logs were not output to stdout. (This is a consequence of our supervisord config.)

As agreed, we left out the caching stuff, i.e. following changes in this PR

• src/tests/common.py
• src/translators/crate.py
• src/translators/sql_translator.py
• src/translators/table_cache.py
• src/translators/tests/*.py, i.e. all pytests in the translator package. These were all failing anyway because of issues with shared fixtures and fixture execution order, so they'd have needed some massaging to get them to work.

Also notice that there were random failures with the Gunicorn+gevent setup which you can try reproducing by checking out the cleaned up version of commit (6b84e35) mentioned above. Those failures got fixed with the Gunicorn+gthread setup of PR #357 (see 92eb940) and my gut tells me those tests failed for the same reason with the Gunicorn+gevent setup but I had no time to actually verify this.