Wikimedia Release Engineering Team/Deployment pipeline/2017-10-03
Appearance
Last Time
[edit]RelEng
[edit]- Blubber in ci (just saw Giuseppe made a patch, thanks!)
- Service pipline job: https://integration.wikimedia.org/ci/job/service-pipeline/
- Stumbling blocks:
- Establish secure way of passing registry credentials from Jenkins to Docker
- docker login uses ~/.docker/config.yaml
- problematic for ci since users are shared between jobs, can work around with labels but that's shitty
- wrapper script that has access to a root 400 creds and do docker push from contint1001
- Namespace?
- Establish secure way of passing registry credentials from Jenkins to Docker
- Stumbling blocks:
Ops
[edit]- First draft of design document https://wikitech.wikimedia.org/wiki/Streamlined_Service_Delivery_Design
- Add stuff !!! Some sections are still empty
question: what happens when someone posts a malicious change?
[edit]- Do we build and push the container?
- tyler/marko: Only build and push on CI +2
- marko: Should run a subset of tests based on an upload
- tyler: current setup has jenkins +1 for unknowns, jenkins +2 for known, and CR +2 tests
- dan: issue of trust (what's being submitted), issue of atomicity -- shitty concurrency
- joe: staging cluster is +2 from user and pipeline, possibly also have integration e2e tests before deployment
- dan +1 -- is there something that's needed for pre-merge?
- alex: maybe not e2e tests on staging
- joe: maybe use a namespaces/tags to control push
- dan: feedback loop is much wider if we're running e2e tests post-merge
Services
[edit]- none (yet)