I noticed with preload that if a worker fails to boot the arbiter will
dutifully respawn it. Only to watch it die again. This patch makes it so
if a worker process exits before making it to the run loop the process
will exit with a special exit code. When the arbiter sees this flag it
will shutdown all processes assuming there is an unrecoverable error.
* All configuration and bootup is handled by the Application objects.
* There is now a strict ordering on the precedence of configuration
settings:
1. Each option is given a default value of some sort in options.ini
2. Any detectable framework configuration settings override the hard
coded defaults for options. Currently, only Paster applications
have support for this.
3. Anything that is specified in a Gunicorn configuration file (by
default gunicorn.conf.py) overrides what was possibly set by a
framework specific configuration source.
4. Anything specified on the command line reins supreme. The command
line is the final authority on a given configuration option.
Though, not all configuration options are available via command
line.
* Configuration metadata is pulled from an options.ini. In the future I'll
use this to build the example gunicorn.conf.py and the config.rst file
in docs/site/config.rst.
I haven't tested the differences thoroughly. The next item on my agenda
is to figure out a way to start testing Gunicorn that doesn't make my
eyes bleed.
An out of file descriptors error was causing the async acceptor loops to
die. The notification process was unaffected so the workers didn't die.
Async workers hitting an error in the acceptor now kill themselves and
rely on the arbiter to restart a new worker in their stead.
