MTuner
By Paul Haile

Table of content

Potential topics for today:

  • Caching and built data distribution
  • Build service maintenance / scaling
  • Build system setup
  • Build systems (Poll)
  • Tracking build versions
  • Starting a build system from scratch (When / Why to do this?)
  • Tools for merging binary files across branches (Repeat from day 1)
  • Branching for releases and deployment

Caching & Distribution

  • Fastbuild
    • Caches results, intermediate files.
    • Really good for clean builds but some report not as beneficial for incremental builds.
    • Good for code and for data build processes. (Easily extendable.)
  • Redis (https://redis.io/) good as a backing store but needs a front end.
  • Simple checksum caching and retrieval still in use by some as well.
  • If you’re using a centralized service, make sure you have adequate resources or you’ll lose any prospective gains to network or I/O congestion.
    • A way to alleviate this is to spend CPU cycles locally to race the cache as well - if a local process of the asset is faster, use that output and cancel the network fetch.

Build service maintenance and scaling

  • Keep the scripts/tools for building the branch directly in the branch.
    • Lets you modify them as needed per branch.
  • Jenkins is a common thread
    • Cascading, smaller specific jobs which can operate on any branch require less overall upkeep.

Build system setup

  • Build system requires significant setup, perl, python, specific libs, etc. - how to reduce setup complexity for users?
  • Docker (https://www.docker.com/) image/container can be used as a pre-built setup.
  • Puppet (https://puppet.com/) for automated setup of the software you need on your dev workstations.
  • Encode the requirements for each output directly in the build system.
    • If you’re missing the tools you need to build something, acquire / build those tools first, as part of the build process itself.

Tracking build versions

  • How to detect/track whats inside a build?
  • For middleware source, build them and upload the output to a common location, then use an npm style distribution/version management of them as needed per-build or branch.
  • Conan.io was suggested for a platform agnostic package manager

Builds system poll

Deciding when to set up a build system from scratch

  • Understand the problem first, lots of assets? Lots of code? Localization? Scaleability? Speed? What is most important to you will guide you to the solution to use.
  • There isn’t really a one size fits all solution out there - it’s important to understand your requirements first, then start looking for off the shelf solutions before diving into custom creations.
  • Consider many builds in flight vs small numbers of builds that you want to go fast (vertical vs horizontal scaling.)
  • From scratch solution, in most cases, “just don’t do it” - off the shelf can take you farther, faster, and most will let you extend from there.

Branching

  • GitFlow - https://datasift.github.io/gitflow/IntroducingGitFlow.html
  • Perforce Streams - https://www.perforce.com/support/tutorial-video-library/video/branching-streams
  • Auto integration tools of some kind are used in most cases for dumb merges to help keep necessary data in sync between branches.
  • Most seem to be shying away of the overhead of maintaining multiple branches for each release.
  • Branch setup based on states of the product works well:
    • Dev (Engineering development) -> Staging (Tools/Engine QA) -> Main (Content creation) -> Release (Prospective release to the public, Game QA) -> Submission (Holding location for when a build is in submission, in case a resubmit is required) -> Live (Matches what players currently have on their machines)

Merging binaries across branches

  • Mostly same discussions as yesterday - avoid this when possible by pulling metadata out into mergable file types, split your assets up more to avoid collisions.
  • See notes from Day 1.