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
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 java_capture( ) called when starting the jvm
_setup_
- In the context of nosetests by default a logging capture mechanism is in somethingabouto range.com/ mrl/projects/ nose/0. 11.1/plugins/ logcapture. html
place (see http://
) and that for sure interfers with our own capture.
It can be disabled with --nologcapture .