Remove stdout log capture

[Lars Butler]

Recently, we revised our logging methods to capture (in a unified fashion) logging info from Python stdout, Python logging, Java stdout, and Java logging. The unified logging is good, but I think we should not be capturing stdout. Here's why:

If you run the OQ engine with cProfile, all of the nice cProfile formatted output is destroyed and gets turned into logging messages.

To reproduce, run this:

$ python -m cProfile bin/openquake --config_file=whatever/config.gem

[Gabriele Favalessa (favalex)]

I agree, capturing stdout make lots of stuff messy (another example is using
the pdb debugger).

The current situation was intended as an intermediate step anyway, with the aim
of using rabbitmq for logging eventually.

I think we are very close: after landing
https://github.com/gem/openquake/pull/150 and
https://github.com/gem/openquake/pull/304 all the logging output will be going
through the logging modules of python and java (no more prints to stdout).

Sending the output to rabbitmq will be a matter of configuring the logging
modules to use the proper backend.

For log4j I found this http://github.com/jbrisbin/vcloud/tree/master/amqp-appender/

and for python this http://notes.variogr.am/post/143623387/broadcasting-your-logs-with-rabbitmq-and-python .

BTW pull request https://github.com/gem/openquake/pull/150 is actually just
removing some obsoleted files. These files have been modified in the meantime,
making the automatic merge impossible. The gist of the pull request would be
on the other hand really easy to replicate, with a simple rm. Thus I suggest
landing 304 and then open another pull request to remove the files, without
bothering with 150 anymore.

[Muharem Hrnjadovic (al-maisan)]

Apparently this is still an issue. When I run this command

    nosetests -s tests/*_unittest.py > test.log 2>&1

I don't see any output in the terminal but the test.log file is not there either.

[Gabriele Favalessa (favalex)]

I put here some comments on the background behind the current situation.

Whatever we end up implementing we have to keep this in mind.

 - If java logging goes directly to its destination without being captured by
   python the lines in the log don't have information about the worker. Hence
   the trick to send java logging to a proxy stdout, capture it in python and
   then spit it out again with the worker id attached

   The code doing the capturing of the java stdout is in openquake/java.py
   _setup_java_capture() called when starting the jvm

 - In the context of nosetests by default a logging capture mechanism is in
   place (see http://somethingaboutorange.com/mrl/projects/nose/0.11.1/plugins/logcapture.html
   ) and that for sure interfers with our own capture.

  It can be disabled with --nologcapture .

[Mattia Barbon (mattia.barbon)]

To add information about the worker to the Java logs, we could put multiprocessing.current_process().name in the Java MDC; the only drawback is that it's not documented that that is the process name used by Celery logs.

[Mattia Barbon (mattia.barbon)]

Most of the functionality is already in master (as of https://github.com/gem/openquake/commit/b0e86fe3d62a9 ); the only missing bit is adding the process name to log messages.

[Mattia Barbon (mattia.barbon)]

Job name added to log with https://github.com/gem/openquake/pull/410