juju-deployer output should be unbuffered

[Tom Haddon]

Currently we do this in mojo/phase.py:

cmd = ['python', deployer, '-c', configs[1], "-l"]

If we change this to this we'd get unbuffered output, which would be helpful when you're sending the output of a mojo run to a log file:

cmd = ['python', '-u', deployer, '-c', configs[1], "-l"]

[Tom Haddon (mthaddon)]

Actually, seems it's line 369:

        cmd = ('python', deployer)

Which should be:

        cmd = ('python', '-u', deployer)

[Tom Haddon (mthaddon)]

Hmm, tried that and unfortunately it didn't work. Will need to look into the bicommand function in mojo/utils.py

[Paul Collins (pjdc)]

It's a slightly outsized hammer, but setting PYTHONUNBUFFERED in mojo's environment filters down and affects its juju-deployer invocations suitably.

[David Ames (thedac)]

The problem seems to be related to a python script with logging calling another python script with logging.

Juju deployer output is buffered and is flushed at the end of the run like the following example:

http://pastebin.ubuntu.com/14674519/

When an error occurs this can be hidden by the flushed output.