CICD made as simple as possible

Missing "features" from idiomatic pipelines

• No dedicated CI build machine
• No dedicated deployment orchestrator
• No Docker
• No Kubernetes
• No container registry

Concepts

• Node daisy chain deployments
  • Only need to compile one time, and no container registry needed because artifacts are just copied from one node to the next.
  • Simple to reason about
  • Don't need to maintain a bunch of pipeline infrastructure, just a reusable scripts and utilities.
• Ubiquitous tools only, meaning the linux distro should already contain everything if not almost everything by default
• Build on Dev machine
  • It is already powerful enough to compile binaries quickly (otherwise you couldn't develop)
  • Build cache is already there, you were just developing remember?
  • Don't need to maintain or rent a separate beefy CI machine
• Hotfixes
  • Ignore soak times and deployment windows and make all nodes in the chain match the desired version
• Pipeline orchestration
  • done through cron

Utilities used:

• apt-get
• basename
• bash
• cat
• chmod
• chown
• cp
• flock
• go
• grep
• ln
• mkdir
• mv
• readlink
• rm
• rsync
• sops
• sort
• ssh
• stat
• sudo
• systemctl
• uname