Wikimedia Release Engineering Team/MediaWiki on Kubernetes/Meeting notes/2021-02-10
2021-02-10
[edit]Always
[edit]- Core_Platform_Team/Initiatives/MediaWiki_on_Kubernetes
- Wikimedia_Release_Engineering_Team/MediaWiki_on_Kubernetes
- Workboard
- IRC: #mediawiki-mw-on-k8s connect
TODOs from last time
[edit]- Announce/post Pipeline Office Hours meeting in Slack
- Continue conversation about CD (?)
- We're in the middle-ish of a migration to GitLab. Do we need to wait?
- What are the mechanisms of deployment now? The trustworthy parts, the not so much?Etc.
- Notes from RelEng sync up:
- What model of CD are teams envisioning? Continuous Delivery or Continuous Deployment?
- CDep is upon merge/build, deployments are always automated
- CDel is a human making the decision post merge to start deployment following a number of merges. Do we already have this? Does `ssh deployment.eqiad.wmnet 'helmfile apply ...'` count?
- Rollbacks:
- Are rollbacks automated? What are the benefits to this or risks?
- Concept of rollback windows. A deployer explicitly opens/closes windows wherein automated rollback is allowed. After the window is closed, rollbacks are no longer automated: 1) since no user is there to handle potential fallout; 2) at some point "forward moving resolution" is a more appropriate means to fix things.
- Deployment dependencies:
- Do we envision deployments of separate services being independent of one another or dependent in some cases? If the latter, how do we handle dependency in our CD model?
- What model of CD are teams envisioning? Continuous Delivery or Continuous Deployment?
How long is the window for considering a rolllback of a new deployment?
General
[edit]RelEng
[edit]- Got pipelinelib job working in releases jenkins and publishing to registry
- Working on multi-version build now - need to figure out how to accomplish this in pipelinelib
- Getting localization stuff working with multi-version and getting ready for packaging helm chart
- Adding private settings
Serviceops
[edit]- Joe: We merged change to move apache to use general data structure for site configuration
- Last blocker to writing chart for apache + MediaWiki
- Some things we need to think about
- re: chart. Do we need a dev environment case? E.g. using sqlite
- How are we going to handle the massive amount of logging that MW generates?
- How are we going to monitor MediaWiki in k8s?
- Kunal (have conflict today, sorry): setting up "restricted" namespace in docker-registry is in progress
Platform Engineering
[edit]- Shellbox merged to MW master, awaiting security review: https://phabricator.wikimedia.org/T268092