Tuning Software for Production

Run MPF in production mode

YAML is quite slow to parse and reading configs dominates the startup time of MPF and MPF-MC. This is mostly fine during development and we can partially mitigate the costs by caching. However, things are different when running a production machine as caching will not work on a cold boot with a typical read-only setup. Usually production machine setups use less beefy computers with slower disks which makes thinks even worse.

Starting with version 0.54 MPF has a production mode which will use pre-compiled config bundles for much faster start-up times. Additionally, this will disable some expensive config and runtime validations to increase performance. Furthermore this will reduce the amount of debug output.

First run mpf build production_bundle which will create mpf_config.bundle and mpf_mc_config.bundle. You have to recreate those files after every config, mode or show change. Those bundles include all yaml files but not any other assets (such as videos or sounds). Second, add the -P flag to the commandline to run MPF in production mode.

MPF will also try to keep running in some cases instead of exiting the game. This will not be helpful to find bug but a when you ship machines you won’t see the log anyway. Finally, MPF will try to initialize for 30s and then exit in case something went wrong. You can use that to run MPF in a while loop or to reboot your PC in case initialization went wrong.

Run MPF without text UI

Text UI costs some performance so disable it in production or on less powerful hardware in general. You can do this by adding the -t flag to the MPF commandline.

Run MPF with pypy

This will currently only work for MPF and not for MPF-MC because kivy is not yet compatible to pypy. Performance and latency improvements are around 10x in our benchmarks so this might be essential on low-end hardware. Download pypy and install it. Since pypy is a separate python environment you need to install pip and reinstall all pip packages for pypy.

curl https://bootstrap.pypa.io/get-pip.py -o get-pip.py
pypy get-pip.py
pypy -m pip install mpf

Afterward, you can start MPF within your game folder using:

pypy -m mpf game

Instead of the mpf command just use pypy -m mpf. For pip use pypy -m pip.

You still need to run MPF-MC using normal python. This might change in the future.

Install the latest Python version

For instance, MPF runs significantly faster on Python 3.6 than on 3.5. Similarly, 3.5 is faster than 3.4. We expect the same for the next releases. You might not need this if you are using pypy.

Install uvloop

When running MPF on linux install uvloop to reduce latency and increase throughput for I/O operations. That will keep your game responsive:

pip3 install uvloop

MPF will use uvloop once it is available. Requires at least Python 3.5.

Some random hints

  • Optimise assets for your hardware. Audio should have the same sampling rate as your hardware is using. Images and videos should be at native resolution to prevent scaling up or down.
  • Reencode your videos in a codec which can be efficiently decoded on your target hardware.
  • Let us know if we missed something here.