module-launcher 0.1.0

reduces startup and import times by maintaining a pool of python processes

python-module-launcher was a failed experiment in improving the startup performance of python programs. It was concerned with two things:

• eliminate interpreter startup times

• reduce module import times

Eliminating startup times is achieved by maintaining a pool of worker processes. Reducing module load times is achieved by importing modules immediately after a worker is spawned (python does not import modules that are already in sys.modules).

I ended up trying this approach because of a few small Qt dialogs that weren’t launching as fast as I would like them too. The results were an 100ms reduction in startup time (PySide is a hefty package to import). At that point I abandoned this project (there are a lot of unresolved issues such as handling return codes and proper shutdown)

Usage

Usage: python-module-launcher-2.7 [-hvdpws] socket [mod, ...]

Options:
  -h, --help           show this help message and exit
  -v, --version        show version and exit
  -d, --daemonize      run in background (default: no)
  -p, --pidfile        pid file to use if running as a daemon
  -w, --workers <num>  worker processes to start (default: 5)
  -s, --spare <num>    spare worker processes to maintain (default: 3)

Arguments:
  socket               location of AF_UNIX socket
  mod*                 modules to pre-import in every worker

Examples:
  python-module-launcher-2.7 -f launcher.sock numpy PySide.QtCore PySide.QtGui
  echo -n "module path.to.module" | nc -U launcher.sock

You can send the following commands to the unix socket:

file path/to/script.py arg1 arg2 ...
module path.to.module arg1 arg2 ...
entrypoint name==0.1.0 console_scripts name arg1 arg2 ...

License

python-module-launcher is released under the terms of the New BSD License.