Deployments/Emergencies

If you're looking for help with an emergency situation, please first try to contact Release Engineering & SRE on libera.chat in #wikimedia-operations ^connect. If that fails, it may be appropriate to use Klaxon.

Deployments

Jump to current event

• Deployment calendar
  • Long-term calendar
  • 1.44 Roadmap
  • Train overview
  • Archive
• Blocking tasks (Current task)
• How to monitor errors
• Holding the train
• 🚨 Emergencies
• How to deploy code
  • Heterogeneous deployment
  • Train deploys
• Backport windows
  • How to deploy backports
• Backport vs Train vs Special Window
  • Train vs Backport
  • Special Windows
  • Backport windows
• Risky change notification

[]

Emergency deployments happen when things need fixing right now, even though deployments aren't happening right now.

How to

To do an emergency release, join IRC and get positive confirmation from both Release Engineering and SRE, and perhaps other specialist teams as well (such as Security) and carry out the deployment with them.

🚨 Step-by-step

• Join #wikimedia-operations ^connect on libera.chat
• Message Tyler (thcipriani) and/or the person assigned to this week's train (see the template below)
• Include a link to the patch you'd like to deploy (and the task if appropriate)
• We'll make sure that someone can help you deploy and that you're clear to deploy

IRC message Template

thcipriani: [FILL IN TRAIN CONDUCTOR]: help! I'd like to do an emergency deploy for https://gerrit.wikimedia.org/r/1234 -- context is T1234

Reasons for an emergency deploy

• Address security issues

  For example, a mis-configuration once meant that a private wiki and all of its content was accidentally made public.

• Avoid data loss / corruption

  For example, a coding error meant that newly-painted pages were being cached in a corrupted form; the longer it went, the more of the site was wrong.

• Maintain availability

  For example, a new feature proved much more popular than planned and the extra load it was causing was threatening to take down the site, so it was temporarily disabled over a holiday, until people were back at work.

• Prevent abuse

  For example, a massive content scraping run from a search engine wasn't responding to automated HTTP 429 speed bumps and so had to be manually blocked until they could adjust their code.

• Major loss of functionality / appearance

  For example, a code efficiency change broke the visual appearance and usability of parts the sites for a large number of logged-out users, and so the change was reverted out of production until it could be fixed.

For deployers

• Rollback first, fix later; maintaining an overall service to our users is the most important focus.
• Prioritise general availability over that of new features; we have a billion readers and only a few users of your new tool, no matter how cool.
• Make on-wiki edits rarely, and only when you really have to; each wiki's editing community expects autonomy.