GitLab Developer Evangelism Team, 11 Feb 2021

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: What's next for GitLab Pipeline Editor?

Description

Michael, Developer Evangelist at GitLab, meets with Nadia and Dov from Verify:Pipeline Authoring to discuss the launch of the new Pipeline Editor and what's next to make Pipeline Editor a must-try feature for even more GitLab CI/CD users. 💫

Learn more about Pipeline Editor: https://docs.gitlab.com/ee/ci/pipeline_editor/

A

B

So I was just uh opening up the agenda. um I shared the agenda in zoom chat and also edited to the event in the calendar.

B

So um since you interact so much with the developer community, michael and we've just gone through a major release of a new feature. uh The pipeline editor nbc and we've been getting like. uh Obviously, we know that there are many things that are quite rough still in the experience, but we will be quickly iterating on it, but generally we've been getting some pretty good feedback, so I was wondering if uh you could tell us kind of generally.

B

What is your impression and what are you hearing and uh what are the major challenges that you see, or maybe some opportunities that you see as well for us at pipeline authoring.

A

So I think- and I didn't really expect that, but I think it was one of the most liberated features for the ci authoring, especially in a way that you can visualize. What's uh what's what's going on like when you're riding the yammer jobs and and other things you know the typical workflows you're committed, and then you wait and either it breaks or it works. And then you say: okay, maybe I want to optimize this, and this has a certain feeling of like code refactoring.

A

uh You want to optimize the pipeline, um but maybe you could like have a live preview and while you type, uh you can also see uh what the changes are and this um I think this is like the it's actually the same. You either write the ci configuration yaml or you write source code. um It totally helps us have something visualized or just like preview. Markdown um years ago we didn't have that, and I was writing markdown and then committing it and on on there, on the git platform on guitar or gitlab was like.

A

Oh, it doesn't look good, it's broken. So having some sort of live preview, um I think, is one of the.

A

I would say: killer fee killer feature sounds odd, but it's it excites you because it makes your life easier and you can also with the the live lending. It's the second part you get immediate feedback and say: hey, this keyword doesn't sound right or maybe you, like. You, made a mistake with the indent, because it's not the top, it's two two spaces in yammer.

A

Who knows that? I didn't know it some years ago, some years ago, so I think adding um not only adding the ci pipeline editor but adding or combining what's already. There um is a huge step forward um and even so um like this is like the negative experience which are shared or the not so good experience. When you want to kick off this thing, when you have an empty state, you have nothing and you want to start it. You cannot do it.

A

You need to find a workaround which is like okay, let's create an empty gitlab ci yaml file, um but it's I know that we like work in small iterations and want to make it happen, um and I haven't checked the issue for 13.9 yet, but I'm hoping that we can like overcome this quickly and say: hey if you want to try it out um and integrate it in your workflow, even though it's not perfect yet um you're just going ahead, um I also tried to integrate it into the switzerland meetup workshop.

A

I did I think two weeks ago um and there was a problem because the uh the workshop exercises navigate into merchandise and at a certain state. When I wanted to introduce the pipeline editor, I only had a merge request, but I didn't want to merge it back to the main or master branch I wanted to.

A

Like start start going and saying hey, we want to try it out, and I couldn't so like making this more complete or making this more accessible um will also help like creating an exercise creating a blog post which just teases you and says: hey. Try it out you just like you navigate into your repository view.

A

You pick the pr ci pipeline editor on, on the left hand, side on the menu you can see the empty state, maybe the empty state changes into a suggested.

A

I have detected your source code dc plus plus here is the c plus plus template thing and the whole experience evolves, and I think if we hit the point of users, naturally using it in the workflows, we also get more feedback when we add things later on. So the initial adoption rate is higher, and this is this is why I'm like pushing so hard on on the empty state thing, um because I think when everyone can just use it, and even so it's it might like break things.

A

So it's not that clear that you need, for example, the the needs for for the asynchronous connections.

A

um These are things which can be like ruled out or improved by feedback, but I think yeah, the um I haven't, checked the linkedin blog the link linkedin post lately, but I think this had four four and a half thousand unique views on linkedin, which is a lot because I have, I think, one thousand followers or one thousand contacts um and it like it was bumping through the linkedin sphere and I think especially for um business and enterprise customers, or also open source users.

A

Of course, um this increases productivity because oftentimes when you want to look also there's a different use case, I'm not a developer, I'm a manager and I want to see how the pipeline could be looking um and someone says yeah.

A

I've prepared this change for you um and you can just like see it. um Hopefully, in the future, like I have um the pipeline editor opened up, and I say I'm I want to see- or I want to edit a specific branch or a specific, much request in that regard, um just like verifying that the change um is doing something useful and probably I don't. I don't really know how the format works.

A

I just want to see, like the connections between the jobs um brings me to another year having some live feedback, but this is something we discussed for later.

B

A

B

A

um No, I was thinking oh this one is coming. I think I did um I'm not sure if uh the the full merged yammer um so when you're using includes and and other things, um it's often really hard to figure out what the compiled yammer code looks like, um and I think I've seen I've seen this being announced attacked to 13.9, I'm not sure if it's ready yet, but I think this will also add a new long avoided feature to the pipeline editor and drive the adoption and saying hey.

A

If you wanna wanna, do live, editing and live lending amazing, you have a preview of what's coming um and you can see. The full extent of using uh anchors extends includes everything which makes your life easier, but also it's more complex to debug. Later on.

A

When say, I don't know why this image is used or why this exit code over this like script, generates a strange exit code, and then you figure that you kind of included and overwrote a variable and things such when you yeah when you program yammer, when you develop yammer code, you can certainly make make mistakes, and I think one thing we also tried out lately was if a specific include would override something else and it didn't work as we would have expected to.

A

I need to look it up, and the thing is, if I have the possibility to see the compiled code, the end result, which is actually then sent to the runner. um This would be super helpful and with like having this possibility now. I think this will also like drive users more users to use eclipse icd in general, but also drive the pipeline editor, because it's in there and when I'm in there and exploring the compiled view, I will totally look into the rest.

C

Yeah, okay idea on annual points like uh what we have seen is cracking the telemetry information on the usage.

C

So it seems that we are are able to capture like new users in terms of new users that are using the pipeline editor, because, if you want to before the pipeline editor, if you wanted to update or configure your pipeline using gitlab ui, you had options either go to the single file editor or they were by the majority of the users choose neither and they went to they're using their own editors like fbs code or sublime text and then like atom. Oh, there are a lot.

C

You know that still I think the majority of the users will prefer to use that. But what we've seen is. Do we we're able to capture like additional users, because if I'm looking at the total number of users that use the web ide and the single file editor versus now, the users that are using the pipeline editor?

C

I'm seeing that? Yes, there is like a decline from the amount of users that are using use the single file later on the web id. But there is, like a value, add for new users, so if I'm combining the decline together, I'm not getting to the number of users that are now using the fiber nato, which leads me to a conclusion that hey like the fiber. Nato is now becoming an attractive feature in github ui for updating for working on your pipeline.

C

So that's a that's something that actually change a bit our prioritization, because now we know that this is a an attractive feature. What we are trying to do is we are trying to capture now more users. So how can we funnel and redirect the user to the python editor like? There are many way, many ways where you can edit your pipeline.

C

So let's make sure that whenever you click I want to edit my pipeline, we will take you to the pipeline editor and the getting started experience. The empty page is super critical, because this is like the first thing you interact with the pipeline editor. We want you to stay there. We want to want you to go away and say: okay, now I'll create a new one, maybe I'll push it and I'll continue to work on that. Maybe I'll use the single file editor and I'll continue to use it.

C

And the third thing is the branch selector. As you mentioned, many users are not using master and, yes, we, it was an mvc branch. Selector is something that we are aiming toward in the coming iteration. In terms of the last feature you mentioned about, the merge yammer feature uh yeah, it is, I think, in review should be scheduled for the end of this iteration, so there is post almost ready.

C

So this is also something that I'm really excited about. So yeah.

A

Yeah, I think I don't know how many users are waiting for it, but in my smaller community and the coffee chat wiki coverage it we have. I get asked often about this or it just naturally comes up when we are doing something in ci, cd and say yeah, probably you just want to have the the entire few yeah. um I I just had another thought when you mention opportunities or like making users uh use the pipeline editor.

A

I know that there is like the the difference between the server which, like generates the yammer and does the linting and the runner, which also receives uh the yaml from the server. But um there is a possibility- and I think it's deprecated- that the runner can like pass the gitlab ci jammer and do something, but it's a limited scope and it doesn't support all the language elements and there is an ongoing discussion somewhere in the runner uh issue. Tracker is pretty old. Where people want that, like the the gitlab runner go, binary can do.

A

Config validation can do. Linting can do create a whole yammer file. If we find a way to provide this cli functionality, either within gilderbrunner or in the cli, we don't have it yet. But there are some community clls out there. So um to like to come to the point.

A

um If we have an api endpoint which provides you with like getting getting the linting getting a live, preview live freeway, doesn't make sense, um but just like linting plus unrolling the entire yaml file and maybe having some sanity checks or getting the feedback like from from the ui representation.

A

This could help everyone who just is on their server next to the runner or somewhere else, and wants to do some validation checks or something like that, um because I think, having two ways to validate a yammer file or getting feedback. How it looks like is not good and um it diverge, so it will divert into the ruby code and the go code, um and you need to manage two two places of like the source of truth.

A

So um the with the pipeline editor, um the gitlab server, is now like the the single source of truth, where the development also happens, um and if you provide some progress programmatic apis. To that, I think our community will catch up on this and find ways to automate that I know that the ci lint endpoint already existed over time, but it was hard to discover- and probably we can think- or we can experiment with features where to the ui pipeline editor and may also make it available as a cli integration.

A

One thing is also like autocomplete, um where you're like on the cli you're typing and in the background the rest api is like asked about.

A

Are you typing a shop name now? Are you typing something else? um Maybe like the drive, and this highly depends on community contributions, I would say, is to to create a cli um and to silly cli then talks to the gitlab server and says I want to validate the snippet. I want to like create something, maybe even update something um just just from the cli. I don't know if we're going there, it could be one of the drivers and saying hey this.

A

The ci pipeline editor is not just ui, it's also api and you can automate it in your workflows.

C

Yeah we we do actually- and there is a there- is now a spike issue that is planned to happen. Maybe this duration of the next on how we can automatically generate a like a schema so that that way, you know as soon as you'll start typing a keyword. We will automatically do like the autocomplete and the idea is to do it like automatically, because you know every time we will introduce a new keyword. We don't want to manually update the schema. What will happen eventually is we'll forget to update the schema it will not be.

C

It will not be up to date. There is a community contribution, uh but it is not up to date. uh It's it's. It is on on github. uh My thought was, if indeed we will develop our own schema, then we can like contribute it upstream and put it in some repo in gitlab and have it like this. This will be like the latest schema store. Users that are using this code, for instance, can load the schema and use it for for writing their own pipeline configuration and yeah.

C

This is definitely some thing that I was thinking about, but still there is a long way for us before we can do that.

A

That's a great thought. I totally forgot about the json-hosted schema file for visual studio code.

A

um This is actually an even better idea to not pull all the time when you're typing and auto complete things, um but if, like the the cri tool which doesn't exist yet um could just like use the json schema, which is hosted somewhere on gitlab.com or maybe just compiled into the binary. um Both. I think.

C

Both both compiled.

A

With the battery.

C

And also a contributed upstream github.

A

And we probably solve one one of one of the highly avoided uh features um because validation, some sort of and having immediate feedback. Not everyone wants to use the pipeline editor, that's totally okay, and if, if we want to want to really like catch up with not only visual studio code but maybe like chat brains, we want to reach the developers where they are at and would not saying hey.

A

You need to use gitlab and in your browser, but you can just like use whatever you are your you prefer, and there is a bunch of editors out there. Some of them have a gitlab integration or gitlab plugin extension, others don't and maybe they those who don't they didn't find the use case and like making them develop. A plugin um could be like hey. You can live, lend your ci jammer now, because this json file is now there.

A

So I think it's a it's more of a long term vision now, but I think there are. There are good ways to to reach or to to reach out to developers even more and think.

C

Yeah yeah, okay and like the value-add for us like, as I mentioned, we we should reach developers wherever they are.

C

We should not limit them if they're using like other editors, but we can provide them value, add if they are using bit labs so like, for instance, if we take this to the next level, let's say that we have like a json schema, so in if you're using any editor, then you can load the schema and it will be like provide you with an autocomplete which is awesome but like if you use gitlab, we can also provide tooltips links to documentation things that we can do more on on gitlab on gitlab ui.

C

So, like you say, hey we you can do that, but you can even get like an improved experience if you use our gitlab uh in terms of like the linting, there are like two methods uh which I was thinking that we should do. One is the linkedin with the with the with it with our apis, and this is like make sure that your configuration is valid. So let's say that you.

C

You create like an empty stage or an empty job uh so like in. If, if you do like, uh if, for instance, you, you will run a lint with a json schema, it will be okay, but the actual configuration is wrong and it's broken. So with with the with our apis, we can validate the configuration and with the json schema, we can validate the syntax, so it's like client, side and and server side, and the idea is to bring them all together.

C

So you can have like the best of both worlds, because, like autocomplete is much better when you do it on the server side, it's only on the client side and the tooltips, but like validating the configuration is something that you should do on the client on the on the server side. So having this like use like the best of for best of breed the best from both world all combined into like a single editing. Experience is like my uh planned going forward.

A

And I think it's also a great way to improve the the validation and the feedback on errors, because sometimes there is like a cryptic error message and I've tried looking into the code and it's like typically dsl parser and you like overwrite, specific scopes or specific uh entry points for errors um and like having having the single source of truth for everything in there probably also allows us to um to educate contributors to say hey.

A

If you want to improve this error message, this is how it works um and we can can explain like the ruby snippets to them and and things like. So I would love to look into that, because when I was working on isinger we also created a dsl and the compiler parser, and things are things around this. um For me, it's like it's. It's it's hard to dive into the room because and say um how does this work? Is there something like how I can improve that.

C

Okay, switching gears a bit, uh and I know we don't have a lot of time. I actually want to go to the last point that nadia added to the uh the agenda. Maybe we can go back to the next one another we can view it. Async we've received a lot of, like, I would say, feedback like ask even from engineers on like to test the pipeline before they run it, and for me I, like I, I I found it hard to underst to to grasp what that the.

C

What does they mean when they say test it? I mean you cannot really test the pipeline unless you actually run it and we can do the linting and we can do like the the syntax validation but like it feels like that the users want to run the pipeline, maybe locally or in some way, without using the runner. I don't know if it's possible, so I was wondering if this is like a a request that you were hearing and if you think like what is the requirement here and how can we achieve?

C

How can we explore that.

A

I hear that a lot, um and I also heard that with a singer like execute a check but don't process the check result. um I think it's in a highly dynamic environment, it's impossible to provide a static pipeline run test. The thing I could imagine is that users want to run it, but they don't want to see the result in the in the history so because the the pipeline exit state or the job stages exit state overall influences, the reporting influences the history and it's just a simulation.

A

The simulation is just run it with all the like parameters or maybe modified parameters to say hey. This is staging, or this is testing, um but when the the result is fetched, it's not like um published um to to the pipeline result and no one gets notified. It's just the simulation. It's just running the pipeline, um but not triggering any events like um if it's, if it fails or if it's, uh if it's okay, um the problem is this: is uh it costs you resources?

A

So if you're running 200 tests and the pipeline is running for 30 minutes, it's probably not the best idea to provide a simulation button.

A

But like thinking of running a job which executes fast um or just a snippet, just one job I want to execute, um I have form and say, execute this one and tell me how it looks: pull the docker image and and do an echo. um This could be an interesting idea um if we maybe provide it as an experiment and also add a telemetry tracking to it and see if users, actually, you want to like, go further.

C

Similar to simulator similar to how you like run, queries like you have like many queries. You just highlight the specific way you place on person play and it runs only that specific line, so maybe just like run that specific job from the entire pipeline.

A

Or maybe maybe just run the job. The problem is dependencies and and artifacts and things you.

C

Need to know about something it's out of our hand, but like it's, it's actually it's it's! It's. A cool idea like like highlighting, highlight the job and say: okay, run this job like in a void like if it has.

A

The problem, if it's, if it's too, if it's too complicated from a ui perspective, just create a form empty like pasting, the yammer text in there and say run it run, it run, it run it on the runner, and maybe you can select the runner where you want to run it on as a second iteration.

A

But I think it's like the the the real-time feedback, the real-time validation of testing, your job or testing your entire pipeline. If it runs and throwing away the results throwing away the locks, I don't care throwing away the artifacts even um and maybe maybe keeping I don't know it's it's a desire. You have because my boss gets notified because the pipeline is failing. um I don't want that. I just want to test it um because otherwise I get ping and say why is the pipeline failing yeah? I detest something.

A

um This is like um a feeling of safety. I would say or like I want to be in my own space and simulation is important on the other side, and I know we are going a little over time.

A

I have some minutes left. The thing is, if I'm a manager, if I like I'm looking at things, um there is something like business process modeling this isn't coming from the monitoring side. um Maybe I want to like change some input parameters and test what the output is. So, if I'm changing the build type from debug to release, um I just want to see what's going on in there. I don't care about the results. I don't want any graphs or metrics or things around this. I just want to see the simulation and say.

A

Oh, this looks good enough. I approve and when I say I approve the change, I'm like applying it to production, so it commits the pipeline uh ci config or I said no, that's not a good change uh to get reset, okay, checkout or whatever in the background um and it's gone so I think it's a business business level desire, but also from a developer's point of view like when you're working in teams you just want to test it out your teammates, don't care, they don't need to be notified.

C

B

Yeah, it's very interesting thanks for explaining kind of the drive behind this. It makes all sense.

A

Context, the last sentence today for context. I have been asked that very often, with a singer and executing things on the server and simulating something I was like. I have no idea how to implement that, um but where the request still stands so it's it's often asked.

B

Yeah, uh so there's only one point uh left. uh I was hoping to get some feedback on uh this mvc implementation for the visual builder, but um I figured we can do they. Sync, you can review the description. The issue description is quite detailed and I linked the prototype here and there's also a mock-up. So you can have a look when you have the time and just let me know what you think, because it has to do with onboarding and we want to provide some existing resources to new ci cd users.

B

So as they're filling out the ammo like what kind of resources would be the most helpful. So I thought you would be a good person to have a look at that, but whenever you have time it's not an urgent issue, so.

A

It's it's good when it's not urgent, because today is a little busy um I'll put it on my on my list, hopefully tomorrow and and give you give you asking feedback on that I'll ping. You when I have questions.

B

Awesome thanks a lot all right. Thank you. Both thanks.

C

B

C