Page MenuHomePhabricator

Convert runner-1030.gitlab-runners.eqiad1.wikimedia.cloud to an instance-wide shared runner
Closed, ResolvedPublic

Description

Change runner-1030.gitlab-runners.eqiad1.wikimedia.cloud from a group runner into an instance-wide shared runner.

Event Timeline

Mentioned in SAL (#wikimedia-releng) [2023-01-25T19:37:05Z] <dancy> Taking runner-1030.gitlab-runners.eqiad1.wikimedia.cloud offline for T327949

thcipriani triaged this task as Medium priority.Jan 25 2023, 7:53 PM

Mentioned in SAL (#wikimedia-releng) [2023-01-25T21:35:55Z] <dancy> runner-1030.gitlab-runners.eqiad1.wikimedia.cloud has been re-registered as an instance-wide runner. T327949

dancy renamed this task from Convert runner-1030.gitlab-runners.eqiad1.wikimedia.cloud to a shared runner to Convert runner-1030.gitlab-runners.eqiad1.wikimedia.cloud to an instance-wide shared runner.Jan 25 2023, 9:39 PM
dancy updated the task description. (Show Details)

Change 884037 had a related patch set uploaded (by Jelto; author: Jelto):

[operations/puppet@production] gitlab_runner: add separate ensure for docker::network

https://gerrit.wikimedia.org/r/884037

Change 884037 merged by Jelto:

[operations/puppet@production] gitlab_runner: add separate ensure for docker::network

https://gerrit.wikimedia.org/r/884037

Small hint: de-registration and re-registration broke the networking of runner-1030 because docker network was removed during the re-registration.

I merged the change above and this should not happen again for other runners.

Sorry for reopening this old task. But it seems during the refactoring of the runner registration (T344951) runner-1030 was converted back to a group runner. So all WMCS runners are group runners for /repos again (see admin menu).

@dancy should we make runner-1030 (or more?) instance-wide again? There are some use-cases which would benefit from instance-wide WMCS runners, like some jobs in toolforge-repos, see T362401#9810208.

Sorry for reopening this old task. But it seems during the refactoring of the runner registration (T344951) runner-1030 was converted back to a group runner. So all WMCS runners are group runners for /repos again (see admin menu).

@dancy should we make runner-1030 (or more?) instance-wide again?

Yes please.

There are some use-cases which would benefit from instance-wide WMCS runners, like some jobs in toolforge-repos, see T362401#9810208.

Agreed.

I was unable to add the private registration token to the single host runner-1030 even with help from the cloud team. I temporarily replaced the devtools-wide profile::gitlab::runner::token with the proper tokeen and unregistered and re-registered runner-1030. The runner is available now for all projects, including toolforge-repos.

So I'll close the task (and try to figure out how to properly add host-based secrets on a standalone wmcs puppetmaster).