1601776 - coalescing: never supersed tasks on tip revision

Status: NEW

(Release Engineering :: General, defect)

Description

[Sebastian Hengst [:aryx] (needinfo me if it's about an intermittent or backout)]

https://treeherder.mozilla.org/#/jobs?repo=autoland&group_state=expanded&resultStatus=superseded%2Cretry%2Ctestfailed%2Cbusted%2Cexception%2Cusercancel&tochange=469a80cf68cabebfebe740c77fb6f488587beb03&fromchange=91db74da5f4a5b10fc47606b168a3ad0618acfd7&selectedJob=279879233

is a task which got set to supersed at 21:04:26 when the push still was the most recent one (tip). That should not happen, the goal is to have test coverage for tip else failures might not get discovered.

Comment 1

[Kendall Libby [:fubar] (he/him)]

Not sure if this is a RelEng thing, but it's not RelOps, so over the fence!

Comment 2

[Tom Prince [:tomprince]]

I believe this is https://github.com/mozilla/tc-coalesce which is a service operated by relops.

Comment 3

[Jake Watkins [:dividehex]]

301 to :pmoore. I wrote it but no longer own it.

Comment 4

[Pete Moore [:pmoore][:pete]]

Coop, any idea who owns/operates this service now?

Comment 5

[Pete Moore [:pmoore][:pete]]

Or do you know, dustin?

Comment 6

[Dustin J. Mitchell [:dustin] (he/him)]

still nope :)

Comment 7

[Chris Cooper [:coop] (he/him)]

We discussed this at the Taskcluster team meeting today. Here's what we came up with:

• If there is an urgent need for maintenance, we can help as we always do.
• Turning this service off is a valid course of action. If it's not providing value, we should decomm.
• If Mozilla is going to maintain the coalescing service, management of the service should live with RelOps or ci-admin since coalescing is only used by the Firefox CI cluster. I will talk to catlee, jlund, fubar, and jmaher to make sure this is OK.
• Superseding/coalescing is a feature we might consider folding into the Taskcluster platform itself at some point in the future. This wouldn't actually change who runs the service though, it would just be baked into Taskcluster instead of standalone. We have lots of higher priority work to get through before we'd consider doing this. Q2 2020 is the earliest I could see us picking this up.