1170861 - Create a new integration repository for autoland

Status: RESOLVED FIXED

(Release Engineering :: General, defect)

Description

[Gregory Szorc [:gps]]

Dan Minor is working on Autoland. The tentative plan is to land to inbound. However, I'd like us to *consider* standing up a new integration repository just for Autoland.

While we can use inbound for Autoland, I'd kinda like to have a separate repository. I'd like to have a separate repository that is "owned" by Autoland and free from "contamination" from humans. If all goes according to plan, Autoland-landed commits will introduce fewer regressions and result in fewer tree closures. If Autoland uses inbound and inbound continues to have numerous tree closures due to non-Autoland-landed commits, this somewhat undermines some of the advantages of Autoland, making it not as useful as it could be. Also, having an Autoland repo separate from inbound (and fx-team) will isolate workflow differences and help us measure the efficacy of Autoland avoiding tree closures.

How much effort is it to stand up an inbound-like repo/branch/tree for Autoland? If it's a lot, please push back and I'll likely withdraw this request.

Needinfo catlee for RelEng effort opinion and RyanVM for a buy-in from a sheriff.

Comment 1

[Chris AtLee [:catlee]]

+1, I've always assumed we'd implement it this way.

One roadblock may be the global # of builders buildbot allows us to have, so adding a new branch may push us over that limit. But we can figure that out if we need to.

Comment 2

[David Burns :automatedtester]

This feels like this will be "We have 4 ways to solve X, lets standardize it. Now we have 5 ways to solve X" type scenario. 

* How long would this be around for?
** I am going to push to shutdown the integration branches once we are satisfied with Autoland. The extra branches creates a lot of busy work for the sheriffs that I would like to remove.
** Could we reuse another integration branch/Project Branch and stop humans pushing there. FX-Team has ~500 pushes a month compared to Inbound which has ~1500 pushes.
* Would this branch not allow pushes from Humans?

Comment 3

[Wes Kocher (:KWierso) (Not reading bugmail; email directly if needed)]

(In reply to David Burns :automatedtester from comment #2)
> ** Could we reuse another integration branch/Project Branch and stop humans
> pushing there. FX-Team has ~500 pushes a month compared to Inbound which has
> ~1500 pushes.
Callek mentioned that the alder twig is available for use if this branch is just going to be used for standing up autoland. That would probably be easier than getting a new branch set up and working around the builder limits, and doesn't force people that would usually push to fx-team to change to pushing elsewhere in the immediate short term.

Comment 4

[Chris AtLee [:catlee]]

(In reply to David Burns :automatedtester from comment #2)
> This feels like this will be "We have 4 ways to solve X, lets standardize
> it. Now we have 5 ways to solve X" type scenario. 
> 
> * How long would this be around for?
> ** I am going to push to shutdown the integration branches once we are
> satisfied with Autoland. The extra branches creates a lot of busy work for
> the sheriffs that I would like to remove.

Agreed. Once we're satisfied with the results of autoland, we should disable mozilla-inbound (or make mozilla-inbound only landable by autoland).

Comment 5

[Gregory Szorc [:gps]]

Using an existing branch is fine by me. The actual URL doesn't matter to me as long as the builds are identical and the sheriffs are happy.

FWIW, I too will heavily lobby to shut down the integration repositories once Autoland is up and running. After all, we created the integration repos to combat tree closures and if Autoland does its job, tip won't go bad and there will be no reason we can't land directly to central. Also, I hate merge commits and want to see as close to linear repo history as possible.