GitLab Delivery Team Fast Boot, 30 Apr 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Delivery Fast Boot: Day 4 branches discussion

Description

The issues listing the work done are contained in epic https://gitlab.com/groups/gitlab-org/release/-/epics/17

A

Remember how we discussed that we were creating these Auto deploy branches on right now for different repositories, given ze give up on the musket, Lev and deployer correct.

B

And the only reason you really wanted to do.

A

It on deployer was so that when you scan the list of pipelines, you can be like hawk. That's an auto, that's just a convenient. My problem is: is that when you trigger the omnibus pipeline with a tag I need to trigger from omnibus the deployer player, that's the way it works.

B

Right like after the packages built many of us I trigger.

A

The deploy a pipeline.

B

A

Need the passive branch to say like use the auto deploy branch, it's a tag pipeline I, don't know how like how do I know what the latest all of the play branch is or because, like of course, the pipeline is running on the auto deploy branch, but it's a tag. It is triggered by the timing right and so what I have is the tag name. I have the rough name.

B

But I don't have a branch name so I'm.

A

B

Into the trigger, as a variable, it's.

A

Not well, first of all, it's not a trigger right. It's a tag. It's okay, a pipeline right, so.

B

C

Doesn't have the branch and if it was a branch when they thought this is a branch build it.

A

Was easy right because would be when we did this as a branch pipeline excited to just I, just use the same branch name, because I have range names all in the box sent. But in this case it's attack. So my my proposal for this would be to maybe not create the auto deploy branch on deployer, but instead have a different branch that remains the same, and then we just use that for now and then whenever we need to update that branch, because.

B

C

We're not cherry-picking.

A

Or doing anything with the deploy right in.

C

In the deployer, the only reason why we only to do this is cosmetics, because forgive look, it's not.

B

C

Us to filter out pipelines properly yeah, it's.

A

Also, thinking like how to have a branch name called, maybe audit to play or is.

B

It called may be honest, yeah.

A

Maybe more, why don't we dad.

C

What are you tagging deployer.

A

A

So if we tag the deployer I can pass the tag name, but what would the IC so it'd be the same tag as the tag on omnibus, alright and and if I pass the tag which corresponds to a ref and I? Look at the pipeline screen, though I see the time I usually see the tag in the.

C

Bike lanes and then you have one on one relationship between a package that we want to deploy and the deploy by plan kind.

A

Of like that's, what would be this attack here, which.

C

Every time we have.

C

Anything you have like one evaluation. The problems are with this approach. The problems are, you have a breakage of deployer and you have a tag you have to go to the deployer, remove the tag and then trigger.

B

A lien on the bus again.

C

To retake it, that's actually true, so that's one problem, but inside the deployer.

C

A

If we expose the deployer rafted customers like in the version string, yeah I need to start with this one branch like and see where it takes us or even master and then we'll just I, don't know, I mean I.

C

Mean it's really difficult jar. It's.

C

The reason why I had to do this.

B

C

C

We probably haven't just cut back, been like you're like this, is.

C

The choice we make on this now is gonna have simply longer lasting implications that will live beyond us.

C

Hey what's wrong with this well I mean actually suggesting they're, just a branch just.

A

Congratulate and then we have, you know I, like a branch prefix like I, don't quite /, whatever no just Auto deploy. So when I trigger the fight fight and deployer. It's always the owner of a branch named Auto deploy and it's updated completely. Independent of the must.

C

A

Know so Daniel what and.

B

So, in other words, if there's a bug on deployer.

A

We would just urge that change it using the owner to play branch, then. Finally, we wait.

C

As well, how are you going to follow multiple deployment they're happening at the same time on the deployment branch, though I think.

A

We had that problem before as well, because we were going to be using one like what you mean within a week right. We only got one branch, so I think we had that same problem. I, don't know how like we could. You could even like.

C

We're going to have moody blue deploys within a day even, and one of them are going to use the auto deploy branch. So so, right now like.

A

Man, I, don't think there'll, be many deploys happening simultaneously on the auto deploy branch right, because this is only for pre, prod or stadium, and then what I'm thinking is is that as we expand on the deploy to Canary and production, maybe the pipeline just gets longer. So then you won't really ever have more than one auto deploy a pipeline happen at once.

A

Right now, it would be like this with an interruption because it D triggered from chat ops, but maybe maybe.

B

This should also.

A

Be over an employee, I, don't know, that's.

C

B

A

Mean we're overloading branches here to like do it legally. That's really! It's remise.

B

C

It eats really sucky yeah.

B

If we go to the branch, how and how often.

A

It would be manual and but we don't frequently merge it, changes to the deploy. Oh right so part of the process. I.

A

Don't know if everything is everything a master everything's release. I can see. You know that, but but I also think that part of the confusion right now is that we have testified lines that run in dry mode. Whenever we have a branch and those like those, we see the branch name when we merge into master yeah I mean like it's really just the production deploys and the staging deploys and the canary deploys that all look the same, because they're all master and.

C

I, don't know how to solve that. How is how these different to detect proposal, because if you have to have a branch where you have to take a manual action every time, you change something. How is that different from you going to the tag page? Removing that tag and triggering the thing again, you.

A

Right I mean it's not that much different and if there's.

C

Nothing else tag actually like has this one-on-one relationship.

A

Okay, I'm yeah I think it's a good idea. I was just a little worried about the leaving and moving times, but maybe let's just see if we have to do that often I, don't think we will I think it's more likely that you know you know we'll have a change that will require a new.

C

Create an issue and explain the reasons why we went with this versus this.

A

But Sicily we have attracts when we need to, but now not all Robert. This means you have another thing to the open it schnapps yeah, but we can do that. All other things like.

C

But we don't do anything from release tools in Australia. We do.

C

So we now have three instances: 311 yeah, yeah, okay,.

A

We have enough time today, let's say Airport I think it's not that big of a deal. It's just you.

C

Know, what's the BU that we do already because member.

A

We create the auto deploy branch, it hops, the deployer yeah.

C

Yeah the problem I have with this is he's becoming impossible, humanely to follow the lines. So, yes, we will that's what mermaid graphic for.

A

Yes, so we have like release tools.

A

Just for my understanding.

C

Then we tagged and released was the tag we get tagged. We get tagged in release tools, EE right, we get diversions, add them to omnibus and tag that we don't.

A

You know we use shop or what we're talking protecting an omnibus whenever we need. We.

C

Are not tagging me? We are finding the latest commits in DL to deploy branch in EE and using that shop and then passing that popular nut.

A

That's fine we don't have to, but why wouldn't we like the deformation right? Yes, because.

C

It's really hard to know.

C

You're right but I guess it's like it's the.

A

Information, so if we just like omnibus, then the the EE Brown is just there's a version file. Also I actually thought we were tagging as well, but maybe we don't need to just.

C

We are adding one by one, you via two more. What was trying to repeat is we passed, diversions, omnibus, right and then tag omnibus.

B

C

Subversion commander right.

B

C

The original files through the API yeah so.

B

It's not in the.

C

Release tool per se is triggering the API and then we tell on the bus and that is going to trigger a deploy or by plan, but I guess having another step in the release tool that will add attack to deployers. But no that's the problem. We trigger a pipeline from omnibus.

C

C

We would have to change that behavior. Then we trigger a pipeline. You know me, but no stripper pipeline.

A

It's well triggered in the sense that we tied omnibus, which then results on the pipeline right better.

B

The problem when the triggers itself yeah.

A

Well, it's a tag, so this there's a on its like.

C

So in omnibus, in omnibus, when we create a package right, you have the job inside that pipeline and you decide that pipeline that says deploy to staging. What does that one do? How does that phone interact with ops? It triggers its triggers our pipeline in ops. It ups? Yes, but that's.

A

C

Point if it triggers a by planning ops, you cannot stag that, because it's a random event.

A

That would trigger a pipeline in ops with the there's, a there's, a field which says like what branch you want to use master. You can choose whatever branch you write, but then we're managing branch not play tag but I assume it's RF, I assume as for breath, right because the branches are rough. So I assume that if I I assumed, if I put a tag in there, it'll just work. Okay,.

C

I sold I'm. So if that's the case.

B

We talked briefly yesterday about instead of using major/minor, we use the date.

C

Please I just want to have a working pipeline today, automated once we have that working. We can then make a change with the dates, and we can. We can I, don't.

A

Think it's a good idea, actually I think the the the the major minor tracks to the month. So it is essentially a date right.

C

B

Very deterministic, like it sounds, point innovate that was point about having an 11th in our little arrange that it doesn't go. Production is now essentially what will be 11:11, yeah I, think Vincent I.

C

Think he does I just don't think the timing is right for us to meet fly to change this around as well. I. Don't.

A

Know how that like you're gonna, have it morph in Auto, deploy at the end near the end of the month, you're going to have an orphan Auto, deploy, maybe multiple seasons now this, if any help, even they, they never go to production right that have to take changes on them, but those changes are also paid to are also a mass murder in that. So.

B

Eventually, like.

A

It's gonna be fine right, like eventually auto deploy will be successful and then we'll have all the changes. So we're done except we're.

B

Upon the autumn, upon the amount of begging, we're posting licenses like hey, this is in this branch. It'll go into this version, then, okay, that's.

A

What I think we need more feedback plate? We need like almost three feedbacks right picked employees when they start disappoint finish. You know. Do.

C

Place finish for maybe just deploy finish, employer finished and deployed to production, the branch anything.

B

But in that case that do one way we.

A

Handled ever merit merit, I, had a long discussion with multiple long discussions about this and I was on the data side and then I think I think came around with it was.

B

A

I think it kind of makes sense to have because they might confuse people that have the date. In addition, a major a minor I.

C

Don't think it's a good time for this to be considered changing. That's.

B

C

Thing right so I'm open for revisiting it. When we actually see the impact.

C

So I don't disagree without, but I want to see the impact first and if it's just as simple as changing the major minor in to date, simple right, we only have three instances in and six repositories what can go wrong.