GitLab Release Stage, 20 May 2021

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Deploy Approvals Sync – Release Group

Description

Daniel Fosco (Sr. Product Designer), Rayana Verissimo (Product Design Manager), and Shinya Maeda (Sr. Backend Engineer) discuss current design work and existing workflow for deploy approvals.

Agenda Doc: https://docs.google.com/document/d/1N9gh7LlmzyGWIeSl1NLDajZUhC7MpaB_xN7K-LvHm3E/edit#heading=h.hpiee6n2ob2j

Issue:
https://gitlab.com/gitlab-org/gitlab/-/issues/35412/

A

Hello, everyone: this is a call to sing about the current manual, allow and deny action on pipeline lab uh we're talking with hyaluvo product designer man, product design manager and the release group and shenya maida who's. A back-end developer in the release group as well.

A

All right, uh shinya media, um a quick agenda doc, just just with some links for us to take a look at as we discuss um so I'm gonna state the problem, uh the last piece of it. uh As I understood, and then you can correct me if there's anything wrong and then we go from there sounds good.

B

Yes, yes, good. uh Also uh I'd like you, I'd like to show you how uh the typical cicd workflow is built, uh like very simple uh workflow example, um so that you don't have an idea like why and keep saying the same thing again and again, like most rex, oriented approach seems wrong um yeah. So the main purpose in this call for me is uh explaining that thing.

A

Yeah, maybe maybe we can start with that instead, um oh.

B

Okay, okay, cool.

A

Yeah, uh if you want to share your experience, it's even better.

B

B

Okay, so yeah again, let's try to um the role-playing that uh how uh like in a typical development uh developers, operators um a situation like there's a one project and then uh let's say it's: three personas one is developer and then qa, qa artist, qa, checker and then infrastructure engineer. And then the developer is there.

B

The responsibility is to write a code and then commit to master branch, and then infrastructure engineer is to deploy the code, the merged code to the production environment and then the this this approval future is, uh you know, kind of helping, infrastruct infrastructure infrastructure, engineer that like, if there's a imagine that there's a deployment button like manual action manual, job and the influencer infrastructure engineer, can click it and then that kicks the job.

B

But he wants to a require additional approvals from the qa qarthis, the qa checker uh and then, which is uh doesn't exist today in glove futures. So let me show you how it works.

B

I'm going to create a new project.

B

I'm going to deploy apple.

B

uh Test, let's say.

B

I created this new project and, let's say I'm a developer here: uh maybe danielle you can you can be your infrastructure engineer in this case. uh So first we're gonna set up the ci cd workflow.

B

um There's a builder job.

B

And then it's a test job.

B

These two jobs are executed in any branches, especially these jobs are useful in module requests like verifying that code or new changes are correct and then doesn't have any bugs and a deploy. That's a production job.

B

uh This deploys to production, environment and environment production, and imagine that um infrastructure infrastructure engineer doesn't want to deploy from moderate guests. They want to deploy from massive launch. You know this is a the very you know the basic concept of continuous daily value. uh It's even uh the same like auto debugs.

B

The pipeline also has the same configuration. So we say that this production deployment job can only execute it from master branch and then the other one is uh like.

B

Accept accept master like this, so these are jobs for much request, and this job is for uh stable branch and we're gonna.

A

B

B

It takes a long time.

B

All right so uh now I committed the uh the guillotine file to master branch and which creator.

B

This production, sorry, I forward the other.

A

B

Cool now we are seeing that this pipeline is deployed, uh has only deployed stage that which has a production deployment.

B

Job, oh sorry,.

B

And then, if y creator, you know like there's or imagine, there's application, uh ruby on rails or python or whatever things, and then I'm changing the one of the future, which means that I'm going to create a modular request. Let's say that uh this is a like code to change it. It's just a documentation change but just written as a code change so like a future audition and then.

B

I'm going to create a new branch.

B

And they create a much broadcast, so uh here's there. My purpose is that hey this application needs this change feature edition. um You can see that, and then pipeline is runs adam daniel. uh Here's. One thing that I want to emphasize that this pipeline is a pretty merged pipeline.

B

uh This pipeline typically doesn't deploy due production only deploys to ruby apps, which is known production.

B

Typically, this pipeline doesn't require any upper bubbles, because imagine that infrastructure engineer doesn't care that whatever developers create development, environments or testing environments, so uh this pipeline goes. This pipeline is especially for valifying. This change.

B

Let's wait a bit.

B

Maybe I didn't need a future.

A

Well, just one.

A

B

Well now, this uh pre-march pipeline has passed, meaning that this modular guest is safe to be merged. Now I gonna merge this.

B

And this march action creates that this postmatch pipeline now this is a. What are we going to focus in this context, um so this pipeline run is uh is running on master branch, so you can see that.

B

This pipeline's running it's created by me um so yeah. This is a typical, continuous deployment workflow that, uh after module guesses marched, this postmatch pipeline deploys to production, and you can also see that.

B

Environment is created and then the deployment history shows that uh hey got a diplomat from this. uh uh This comment, which is committed from uh the march request and now uh we're gonna, improve this pipeline because think about that, the problem is that I can deploy to infrastructure. uh I can deploy the production environment but daniel. You don't want me to do that because I'm just developer, so what you're gonna do is uh maybe at first make it manual job so like uh the job won't ex automatically executed, but you have to click it.

B

The play button.

B

B

Let's see and then.

B

Now this uh easier pipeline on the master branch, the stable launch and the production job is shown as a manual so that it won't be automatically executed, even the module. The a new module case is merged. It doesn't deploy automatically you, then you have to uh click this button, but I can also still click it. um That's also a problem right. So uh what you're gonna do is.

B

To create a protected environment.

B

So uh you can protect an environment that only authorizes users can deploy to. You can execute the deployment job. So in this case.

B

Only wait, let me can I invite you guys.

A

Danielle, it's d fosgo.

B

A

G4 squirrel infrastructure, engineer: okay,.

B

B

You as an infrastructure infrastructure, engineer, visits to the project level, configuration protected environments and then set up the new protection rule that only uh only you danielle can execute your job. So you choose you and then click the product button.

B

So now it's production environment is protected, meaning.

B

I can execute this job, so it's cool for now and now.

B

uh If we're seeing this screen as daniel, I can see the play button that I can deploy to the production server, but then the problem in this issue that we don't want to require additional approvals. Let's say that hyena is a qa artist and then how do you want her to join in this loop and then, like a requires elbows or even showing the approval history who upload who is eligible for uploaders etc? um So uh can you cancel showing? uh Can you start sharing a screen in your daniel and then.

A

Yeah, of course,.

B

Okay in this project, uh yep cool, so you can the first question for you as an infrastructure engineer is um how do you know that, like deployments, the manual deployments are available. um Let's say like they're, a bunch of 100 mark wickets have been merged, uh 100 pipelines created on the master blanche and the 100 manual deployment. Jobs are waiting for you to click the execute job and, in addition, like highness approval, how do you see that.

A

Right yeah, you would have to I mean I would probably show one see 100 of these right in the pipeline view.

B

Yeah, so even the pipeline view is very inconvenient that you're sh you're, seeing that pipelines on master branch and also huge avalanche, but you don't want to you, you don't care uh future lunch. What do you care is only master plans? What happening on master branch, so one answer for this problem is that you visit the environment page.

B

uh Can you go to the? Can you open the left, pane and then go to operations, environments.

B

And then click the production.

B

uh uh Yeah, it doesn't even show that the available um deployment jobs- uh these are only finished deployment, jobs, so yeah, it's this page, is even incompetent. um So what do you mean.

A

So uh one one question I have is this is a use case. We have today right of infrastructure engineers that have ci cd workflow, where every production deploy has to be manually approved. So the way the only way they have to do this today is to come into all the pipelines for the project, the pre-merged ones and post merge ones and then just scan which ones are production and just click either here or here for all of them is that.

B

It! Yes! Yes, yes, yes,.

C

What happens on the on the filter pipelines that search bar? What is it and what can users search for there.

A

I guess it could filter by skipped, but I I assume if there are other stages in the pre-merged ones, that are also skipped. They would also show up here.

B

Maybe you, if you don't buy branch names in this case.

C

Yeah, but that's another okay, it would be possible, but it's an extra step and not everyone would like to go and keep searching for things. Yeah. The most intuitive.

B

Yeah, uh the most reasonable feature in this case is that showing all available deployments job in environment page not the pipeline page yeah. So um I think that infrastructure engineer wants to take an action in that page, not instead of pipeline or not even modular guest, so um yeah. This is a reason why what what I suggest why I suggested that maybe we should build this feature around environment, uh slash deployment page, which you uh you can click from that you know left pane operations, environment.

A

Yeah, I think it makes a lot of sense. For me, the the piece that that was missing for me to understand was the difference between the pre-merge and the post-merge pipeline, because that was very clear for me. I would see two two pipeline widgets in the merge request, and I was just assuming that the pipeline was running twice, maybe because the merge request was updated or, for some other reason, uh so yeah having the post merge pipeline running the production deployment only on the main branch.

A

That makes a lot more sense than than what I was thinking.

B

Yeah, so um one thing is that we can improve in this page is showing all manual jobs manual like uh deployment jobs waiting for execution. So uh I think it's easy from back-end perspective. We can show that, like these available jobs- um and maybe it's even more convenient to show that all associated much biggest from deployment jobs, so hey this deployment job. No, currently you see that uh the top the latest successful diplomatic job is marcel blanche like marsh branch future.

B

There is a fusion branch name into master, but it's not really helpful, like you really have an idea. What like you want to directly dive into the uh modular constant and what actually changed yeah like especially this is useful for a qa artist like um you know they want us at first. No, they want to know that change lock from the the previous deployment onto the upcoming deployment, so what's being changed, and then okay. So here, though, feature a feature b feature.

B

She was added so they're going to dive into the staging environment and then they're doing qa and later they approve the pro deployment job to the production.

B

um Sorry I skipped the staging workflow, um especially qa engineer, performs a like the common cases. They perform uh the qa testing in on staging environment, um so.

A

So the same, this, the same way, the the devops engineer, would should be able to see all the production deploy requests here on the production environment on the staging environment. The qa person would also be able to see all the requests to go to staging. Is that what you get all right?.

B

A

B

um Staging environment staging a deployment doesn't need to be manual. It can be automatically deployed like what I did at billy. First, uh like every time a new modder gets merged. It will not get deployed to pro uh staging because it doesn't harm any end users. So yeah qa artists can dive into the testing anytime.

A

um One one final question: I have um why: why does this? This job gets the state of skipped when, uh in practice, yeah waiting to be approved or not right.

B

Yeah yeah, so um money on job is a very like fundamental future in pipelines, so it doesn't even understand the concept of deployment it's just just manual and then this job like from pipeline perspective, very fundamental pipeline perspective, it's just manual job and then, if it doesn't need to be executed. So it just skipped right. So this pipeline said it just skipped. um If there is a like the other job uh like testing or build a job, and then it was it if it were executed, then this pipeline status will be like passed or something.

B

So this thing skipped status, isn't really important here, right: okay,.

A

Yeah um all right, thank you. Thank you. So much for this overview was really helpful to to understand more the technical baseline behind this this issue. um Hayana, do you have any other comments.

C

um No not related to to watching yeah explain but uh I'm curious to understand it. Then you know uh what would be like from you to the the next steps from um looking at the problems that you currently had with your design proposals.

A

So let me go back to the.

C

Yeah, by the way, I changed all the agendas. I put your points at the top here.

A

uh So yeah coming coming back to the issue, I think uh we had the benefits with um solution, validation, to see that this approval interface uh worked really well, even though the entry point didn't so much so I think we have. We have the benefit that we can adjust this and move it to the environments page um and and just turn the mvc into that. I think it's what works best at this point um I'll commit to paper all the things sheena said just so.

A

I can understand it better where it fits in their workflow, but I think yeah I'll break down another issue to explore how this can be integrated in the environments page and we go from there.

B

Yep sounds good to me um so, like uh uh reflecting my points, though, uh like how infrastructure the point, one is how infrastructure engineer will find that deployment available deployment, jobs, and the second is how the big approval feature. um Yeah yeah.

C

Yeah- and I think that's uh that'll be interesting- I don't know like from a conversation with kevin yesterday, but also looking at that environment in the plugins pages uh faithless, like that, I think it would be interesting to see um on the insides, but also what are the issues that we already have uh for the the environment's reordering page uh filtered search these things, uh because just if I'm here, if you trade it, but I think there are a lot of different priorities for that page uh in terms of yeah, improving the usability discoverability of the issues, I think, would be nice to um just have.

C

Like a. I know, top five list of boring solutions. You know the low-hanging efforts that we can have to improve uh this findability of the um the manual jobs or the things that need approval on the leads field. That would be super cool.

A

Yeah, I think definitely I have to take another close look at the list of bugs and and small issues we have for the environments page, because I also don't feel 100 comfortable to just add this feature there at the state. It's it's at so yeah. Let's, let's do this auditing and come back to it.

A

All right, thanks again shinya. uh This was really insightful, really helpful. I I wish uh I had called this conversation earlier because it would have saved some time, um but nonetheless it was really good to hear from you.

B

A

That's one conversation yeah all right um have a good day. Talk to you later.

B

A

C