GitLab Static Site Editor, 29 Sep 2020

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Issue refinement Session: Usage Ping tracking for Static Site Editor

Description

In this session, we refine the issue asking for usage ping data to be added to the SSE

A

Hello, everybody: this is the static site, editor issue, refinement session and in today's session, we're going to be talking about uh an issue that asks us to ensure that the static site editor is instrumented for usage ping data and for those that aren't sure what that is so for self-hosted, gitlab customers.

A

There's a usage ping that sends back aggregated data to gitlab that we use to help understand the usage of our various product features.

A

It's an opt-in feature that customers can opt out of so there's no privacy concerns and we'd like to make sure that we get some data back um from uh from self-hosted customers to identity, to be able to understand whether um we only have customers on gitlab.com on our sas platform or where the self-hosted customers are also using the static site editor, and that will be valuable information for us as we make plans for the future.

A

I'm going to share my screen.

A

All right check: can you see my screen? Fine.

B

A

uh Before I get started, are you familiar with the usage thing.

B

uh I know nothing about this.

A

All right, um I know very little about it as well, which is why I put it on for uh refinement, because I I know we've implemented, snowplow tracking, um which.

A

um But I I don't know with you. My understanding is that this does not relate to to usage, um and I need to actually understand a little bit more about this um to understand.

A

You know I can know we would capture.

B

Snow with the snowplow date, but.

B

There's a page on it in the darks that I sent what is usage ping.

A

Yeah right so snowplows and enterprise-grade marketing product energy's platform, blah blah blah. Okay. So that's fine, um but I want to know whether this. What I really want to know is where the snow cloud is enabled for self-hosted customers um or, if it's only in gitlab.com, because if we.

B

See the the limitations link on that page, I sent it.

B

Says of these limitations, we recommend instrumenting your products with snowplow for analytics on gitlab.com and use usage ping to track aggregated backend. Events on self-managed.

B

Does that mean snowplow only works like your lapcom and then self-managed? You can only get back-end events.

A

Usage ping does not track fronting deadlines, like page views, link, clicks or user sessions that only focus on aggregated backing events. Because of these occasions, where you recognize instrumenting, instrumenting product with snow platform or detailed, analytics and use all right. So that almost implies that snowplow is not available on self managed. But it.

B

Seems like it and we can only get back end on there.

A

Yes- and you can only get back-end aggregated data with the usage ping.

B

So there's developing and testing usage ping uh section at the bottom of that page too.

A

um Yeah, so why should we enable since a weekly, payload.

B

Yeah, it looks like this.

A

B

Developing and testing section has fairly detailed instructions of what we need to do all right.

A

It still doesn't tell you, it still doesn't tell me that snow with a snow plow, it is disabled tracking can be enabled at the instance level, which enables tracking both the front, and so I I assume one of the users.

A

One of the um one of the issues with with snowplow is twofold: um it it obviously it respects to do not track standard in browsers, so it doesn't report everything.

A

And so, if you want to have a more true database level, I guess uh analytics of events, then yes, it makes sense to have to use user string, but also the fact that some self-hosted customers are in air-gapped environments, where there's no ability to make calls out to the internet so um which is what snowplow would require, so that would that would basically make the snowplow unreliable as a true.

A

Reflection of data.

A

Would you agree with that.

B

E! That's what it seems like.

A

Yeah, so you can do it in the back end as well. It seems implementing snowblower are really back in tracking gitlab, provides tracking the interface that wraps a snowblower ruby tracker for custom tracking legends.

B

A

I found something.

B

uh Yeah, uh let me send you this link there. It is so this is the uh the direction page for telemetry build collection framework, current state snow, plow, eventually available on.

A

Self-Managed all right, so it's planned, but it's not uh right now: okay, well, that that makes it very clear why we need usage okay, so.

B

Okay, I think that other page is pretty detailed on what we need to do for usage ping. The developer guide.

A

Yeah, uh so that's on this one again, but now what I want to do is just quickly just talk about what it is. There's issues asking of us. So it's that excited is returning usage analytics, but it appears to only be for sas.com users, which is correct, because snow cloud's not available for self-managed uh the state we need to have visibility into our self-hosted users are taking advantage of the feature.

A

um All right proposal determine first, whether or not we are getting data. Okay, so yeah we can start editing so proposal. We know we are not getting data from usage, if not determine if we can get data. Yes, we know we can't, if so instrument for usage, okay and update dashboards. So proposal is that we need to instrument uh the static slider, editor or usage ping data uh update dashboards- and this would be specifically this one.

A

Now I can, let me just get into.

A

Quickly and get to that dashboard.

A

A

So this is the dashboard that eric is using for his uh overall usage determination so that so update the update. um I'm recording this set excited usage, matrix dashboard.

A

All right right, so instrument, aesthetic site editor. You should know details for the details.

A

um I think we need to refer back to the the issue where we implemented the original trackings to see in order to track on those every commit generating aesthetic site generator. So that's that's a key for the details. Okay, so.

A

Event, data we want to track is.

A

Irregular generated from the sonic sign to editor.

A

We also session data and aesthetic side. Editors.

A

A

Grace that's fine. First, we should track how many projects get created um right now.

A

We only record commits initialized editor and create merge requests. Okay, so those are the the three even though create merge. Requests right now uh is part of the create, commit and create metrics part of the same flow. We do record it as separate events, let's just check here, so that click drop-downs.

A

Create commands and create metric quiz, initialize editor.

A

A

So the other one we want to capture the conversation. It's initialized.

A

Every time a limitation quest is created, which one is that exciting all right, so that's the that's the event data we want to track. Do you.

B

Want to throw a link to that developer guide in there.

A

Yeah, I definitely want to do that.

A

Okay for the details.

A

The next thing I wanted to to look at is so the way I understand it is that these are all individual events that that we record now. My understanding from the usage thing is that it's aggregated counts. It's not necessarily individual events that will come back. If I understand it,.

A

And also does this show up in sizes.

A

B

A

So we can query this data in sizes. The usage being chrome drop is set inside hector directly when the cron job runs. It calls gitlab usage data, github usage, cascades down to 400, plus other counting method calls the response of all myth calls are merged together in a single json payload judicial payload is posted to the versions application.

A

There are several types of counters which are all found in usage data ordering you back down just so. Some kind of a given active record relation extinct by distinct count of a given item for a column is for enumerate, account use for settings and sum the values oven.

A

So why batch counting for large tables pressure can take a long time to count rows? Batch counting is a counting method where a single large quest, broken into smaller queries.

A

A

So it almost seems like we're: gonna get aggregated counts from an instance.

A

Again, we can get aggregated accounts from an instance against you know these three events.

B

We can get roll counts of uh yes,.

A

Yes, so table so okay, so we we create, we record each time these. These events occur, which creates a row entry, but in terms of the usage data that gets sent back, it's not the individual row entries. It's the the aggregate accounts.

A

Is that also your understanding.

B

Yes, uh I guess what I'm unclear on is we need to make a new table to capture these things to actually commits, or could we use an existing table.

A

I think there's existing ones, but, let's so, let's quickly just go through so use your rails console to manage generate the sql query that uh optimize queries with database library. Add the metric difference when adding changing or updating metrics. Please update the event. Dictionaries, usage ping date table.

A

A

A

So I guess this.

A

Issues created from others see this is this is where we would want to add things, and we can look at like web ide commits whereby the emerge requests uh web ide views the this is essentially what we, what we want to replicate count up views of the web id kind of commits from the way by the account of merge, request creator from the web id, essentially that you know this is exactly what we want just for the static site, editor.

B

B

So we need to add an event.

A

Events, um do you want me to capture.

A

Refer to even dictionary.

B

There's also a link uh event dictionary for usage ping, there's an issue about it.

B

uh Last sentence in that paragraph, or in that section giving dictionary, we currently have an initiative across the entire to completely benefit.

A

Masters created a month ago,.

B

A

Sharon jerome, so jerome runs the er telemetry owner.

B

Create static site editor.

A

Thanks for now, changed it to product analytics.

B

So if you click through that issue, we have a a line in the table there that has metric to sign zero yeah. There.

A

B

So I assume this is gonna. Give us a one there.

A

Well, we're gonna want three matrix.

B

Okay, however, many we have.

A

A

A

Highlighted in yellow, so the this is a neolithic is orange.

A

Who calls this yellow um sorry, peace, peach, no chat.

A

That's peach, that's orange.

B

All right, it's orange salmon.

A

B

A

uh uh These are this: these aren't steps right, I mean add new metric to versions. Applications, see this as well check. If the new matrix needs to be added in the see usage, data schema and usage data parameters accepted any metric on the account keys. I say I change like ask for it to be able to review this with one of those tests would give out.

A

Okay, so this is an example: payload.

A

Let's see this web idea.

A

That's not what we really need to care about. What we really want to developing and testing is the future. I wonder if there's a an mr that we can use specialized indexes, I mean there's, they seem to have examples for every step, but really nice. So I know I mentioned in this here an example where, when a1 added a whole bunch of metrics to the music thing, which might be useful to reference, let's see if we can find an mr that added something click, the timer click status.

A

A

Architecture, instrumentation of.

A

Let's see exam.

A

B

And we need to do it via radius.

B

A

Maybe preferred.

A

B

So if you look at the reddit section, it does have some links to uh examples like wiki page counter.

B

Pointers to code.

A

B

I don't know if this is to only to track stuff that is otherwise in redis or if redis is an acceptable.

A

B

Do it without having to create a table.

A

Trucker, you said she went for insulin, creating the graft get out.

A

B

Wondering that seems like a way to do it.

B

Issues update service.

A

See the problem is right now, where our calls to create uh the merge request. The and the commit is all done by the the the rest api and we we we hit that via the the front end. So we don't. We don't have a controller that we can.

B

But we can create one, we could add a new, we can add a new action or a new controller which is solely to populate this redis uh event, okay, showing so.

A

B

So do that before or after that, whatever the event points that we are at.

A

So my my question is: do we create a um a controller action that acts as a proxy for creating the commit and the merge request and and call that and have that then call the relevant service, or do we still call do we call it in addition to calling the api.

B

Do we need to figure that out any issue, or do we just need to call it out as a decision and let the mr implement it decide.

A

B

A

B

A

Oh chad, I am not um considerations.

A

Currently, it's not exciting.

B

My preference on it, though, would be to um keep the back end as simple as possible, and let the front end be the more complex part that handles coordinating this stuff.

A

I would always so currently the fronting does it in a promise chain, so I would, in the success call of each of those promises, just make an additional call to an action in our controller. That then, will take care of adding the event for us.

B

Right and when you do it on the front end also, it lets us if we need to have a lot more granule, granular control over progress and stuff, like that, not not blocking on back end requests that may be doing a lot of stuff.

B

B

B

I have to say I, like the simplicity of the back end, like um the fact that we don't have a lot of backing code, because back in development is it's a lot slower running. The test is slower. Everything about it is slower than doing the front. End work.

A

Would you would you extend our current static site, editor controller with just with another action, or have a specific controller that just deals with event tracking.

B

A

Implementation detail.

B

Either of those are acceptable approaches, I just had a friend tweet that entrepreneur's rails developer. I've worked with who said that he likes to make a lot of different controllers.

B

Personally, I think that's kind of okay, because then you have smaller specs. You don't have this giant spec, that's dealing with a bunch of non-cohesive things in the same controller, yeah.

A

You're, making arrangements to consider adding a controller action.

A

A

A

Currently, the fountain already makes calls for smaller cloud tracking and, like I almost think there should be an abstraction on the front end where the static side, energy from you know, from the components it just says. You know tracking me and then the abstraction goes and makes the necessary call for snowplow and for uh usage pings so that we don't have to keep maintaining two things all at the same time,.

B

Yeah when I like, when I do the config file, I think all of these should start with uh just reviewing a couple three instances of how other areas are doing it, uh especially more recent ones, to look at the best practices and it's great to get your head around it and then also like for the config file. I was able to just reuse a lot of those tests.

B

Some of them needed to be cleaned up a little bit, and I I cargo called it a bit, but it made it much faster than just tried to write all the tests and scratch myself to just use something that was really similar and change. The names.

A

So one of the things is, we need to have front, be able to just be sure, because it's definitely going to require some front end work. Not that you see you can't do it, I'm just adding an indication.

A

A

uh Okay, let's see what we have, we define the problem. The proposal is instrumental search for use, explain data update the static side reduces to dashboard. This, for me, is actually not part of the uh updating. The dashboard is not the job of the engineer that would be for products to do.

A

And I know eric has been working with uh someone from the data team already to um so, I'm not gonna make the uh the requirement. You know like the pro instrument is that excited for usage ping data for the details. The main data we want to track is is or not.

A

Let's just ignore it, um the main data we want to track every time the static site is initialized. Every time I communicate every time link to usage ping developer guide firearm, the web id already tracks similar events to what we want to capture refer to the ink victory.

A

We want to consider reviewing their implementation of.

A

Considerations currently, the statics are okay. This feels better defined for me now.

B

Yeah not bad starting from something we both knew nothing about.

A

B

This actually, this documentation is really good in git lab. It's pretty amazing yeah. I.

A

Mean previously, you know, I didn't know where to start with an issue like this. No, actually you know this, they know there's another piece of the application. That does what we want to do. I know there's developers guides. uh I think this. This sets uh in this case, basically up for a better start to this issue.

A

All right, um I'm gonna, I'm gonna call this ready for.

A

A

um All right- and that is all she wrote for this facial refinement session. I will stop the recording and we.

A