Wikimedia Release Engineering Team/Deployment pipeline/2018-03-15

2018-03-15

Last Time

• 2018-03-01

RelEng

• noticed helm is packaged -- unblocked \o/
• Blubber updates
  • Support for python in addition to npm
  • Changes in permissions scheme to lockdown production images (more at https://phabricator.wikimedia.org/T187372 )
  • internal simplification/refactoring
  • Upcoming: Probably about time to move blubber to gerrit
  • https://phabricator.wikimedia.org/D999 blubber policy
• thcipriani officially worried about quarterly goals
  • Minikube packaging
    • requires golang 1.9 to build
    • https://gerrit.wikimedia.org/g/operations/debs/minikube
    • https://phabricator.wikimedia.org/T184457
  • puppet work for integration (install helm, minikube, k8s-client on integration docker machines, ensure minikube started)
  • Update pipeline to include helm test for minikube

Note: for building images for production do not use the cache at all, layers fine, cache bad

Random neat thing

• Blue Ocean makes pipline look fancy :)  https://integration.wikimedia.org/ci/blue/organizations/jenkins/service-pipeline/detail/service-pipeline/41/pipeline

Fuzzy Future thinking

• What/Who triggers new deployments?
  • Click a button on Jenkins
  • Automagic deployment via Jenkins after tagging
  • Git wrangle/manually run helm on a helmdeployment1001
• Where is this run from?
  • Who's jenkins? Our jenkins? Contint1001?
    • we currently push from there
    • that box does a lot, probably too much

• Split off things that do the work from things that do the testing

SREs

• mathoid deployed on kubernetes@eqiad cluster
• working on the configuration a bit
• tiller can be a pain (client-go kubernetes libraries in general)
• Enabled ServiceAccount admission controller for that
• pybal ready to push traffic
• docker-ce from 17.06.2 to 17.12.1 for CI

As Always

• Release Pipeline Workboard
• Meeting notes