Page MenuHomePhabricator

TrainBranchBot merges code to mediawiki-config master branch, causing undeployed code problem
Closed, ResolvedPublic

Description

NOTE: According to @brennen in -operations, this was caused by T282824: MW container image build workflow vs docker-registry caching.

It sounds releng's TrainBranchBot has started to merge code to mediawiki-config master branch, causing undeployed code problem.

@Catrope described why this is an issue in https://wikitech.wikimedia.org/wiki/How_to_deploy_code#Problem:_undeployed_code. Long story short, it confuses any deployer who deploys after.

With humans causing this issue, it is usually resolved with reverting the patch and (sometimes) by yelling at people. With bots, this is much more serious -- as the bot will just do a new patch after it is reverted.

For that reason, I've removed the ability of TrainBranchBot to merge there by removing it from wmf-deployment Gerrit group, until this issue is resolved.

Since this probably breaks other parts of the bot (namely, branching), I fill this as a new UBN task.

Event Timeline

Urbanecm triaged this task as Unbreak Now! priority.EditedJun 29 2021, 11:56 PM

Per description.

Mentioned in SAL (#wikimedia-operations) [2021-06-30T00:01:17Z] <urbanecm> (following up previous SAL item) TrainBranchBot was removed from wmf-deployment group because of T285819

Sorry for the confusion this caused. I have disabled the Jenkins job which updates train-versions.json and will look for an alternate way to handle it.

dancy lowered the priority of this task from Unbreak Now! to Medium.Jun 30 2021, 3:27 PM

TrainBranchBot has been readded to the wmf-deployment group in Gerrit. Closing this task.