Deployment tooling/Cabal/2015-04-13

April 13th

• Services need better visibility and hookability:
  • Better feedback and interactivity from the git deploy cli
    • one aspect of this is non-blocking interactivity of git deploy, it should report thorough, real-time feedback and allow deployer to take actions during the deployment process, not just after
  • Needs to have:
    • rolling restarts
    • 'canary' testing
    • easy semi-automatic rollback when things go bad
    • emergency abort if deployer notices something went wrong
    • ability to isolate and investigate failed nodes when the majority of nodes succeed but a few failed.

Homework:

• Mukunda and Chad will work further on the mediawiki/core and extensions release branching strategy
  • Mukunda is experimenting with git-subtree to create a composite release repo with all deployed extensions merged into one place
• Tyler will attempt to document the hooks in trebuchet which we can use to provide custom behavior for individual services or projects
  • This is already in the trebuchet core and it's configurable, just not very discoverable right now
• Dan to sit in on deployments to get a better sense for the current practices