keptn Keptn Refinement Meetings, 25 Apr 2023

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Keptn Refinement Meeting - Apr 25, 2023

Description

Meeting notes: Meeting notes: https://docs.google.com/document/d/10Fig1eYFZ9iQFSYWkz0c4eTwzgJiPtQI5IsczbvLsuE/edit

Learn more: https://keptn.sh
Get started with tutorials: https://tutorials.keptn.sh
Join us in Slack: https://slack.keptn.sh
Star us on Github: https://github.com/keptn/keptn
Follow us on Twitter: https://twitter.com/keptnProject
Sign up to our newsletter: https://bit.ly/KeptnNews

A

B

And let me share oopsie the screen.

B

And I'll share also the meeting agenda.

B

Oh nice. Thank you much for the scrum poker.

A

C

Can somebody please share the scrapbooker.

B

Link it's a already in the.

C

ah Sorry thanks: okay.

B

It's almost second that I try to reorganize my screen, sorry for the preparation, but.

B

Nothing was working.

A

B

Dokie so, let's start with a list of ticket that we have open for today. The first one is from Florian about a small improvements for the validation web hook.

D

Exactly yeah, because in the white book we already have a logic that checks for the maximum length of a name for an app. But of course, it's also possible for users to create the captain application custom resource themselves.

D

And here we don't check for any name length, and this is a potential for for errors, because what the app controller will do is it will create a application version which, uh whose name is name, is based off concatenation of several properties of that Captain application.

D

And that means that if theoretically, the name of the captain app already is equal or at least close to the about 253 characters which are allowed for customer or for Q analysis resources in general, then we will have an error that the cabinet version cannot be created and yeah I think this is something that can be yeah concert or if we check for that restriction as early as possible.

D

So therefore, my suggestion for evaluating web.

B

Hook, but instead of pushing towards the user to make sure that the length of the name is a in a good range, can't we, when we create the captain app version stream, the name.

D

I also thought of that, but this might be tricky because in the workloads we also refer to an app and I think here we do look for for a matching um cabinet version to check whether a workload can enter its deployment phase.

D

So maybe this also works but uh would require more effort and.

E

D

Yeah should at least be considered as well yeah.

B

Because of a problem that we, if we require the user to know the right size for the name most likely, that we never discover the problem, maybe let's.

A

B

uh Instance or workload.

E

D

The workload instance controller I'm, currently looking forward to accept place where this will likely need to be adjusted.

D

Okay, so after you have saved that I will add the link to the piece of code that needs to be double checked.

B

Yes, thank you.

B

Okay, it's saved.

B

And thank you for adding the piece of code. Oh and I have an issue.

B

Let's estimate where we do the 31st, which is more user friendly, and if this doesn't work, then we do the length check.

B

This should be the complexity to estimate.

B

Or do you have any other questions for this topic team.

C

So we are estimating.

B

C

B

No, both estimate that you need to do the alternative and if it doesn't work, you do the first. The left check. Okay,.

F

Can we write them opposite I would try the first and then do the alternative.

A

B

I'm not do you need further input to estimate or.

F

I'm undecided I will.

A

B

Strong opinion for uh s Moritz.

G

D

And I saw that worksheet already uh announced its interest. So if it's fine for everyone, we can assign it to him or just decided. Okay, perfect. Thank you.

B

Are you happy with that rakshit or.

C

Yeah, thank you.

B

Thanks to you perfect, then remove the lip up. It thanks to remove also available now, the second one uh you fix the title.

B

Anna or which reason we want to have a cattle test for Jager.

F

F

So realized, while we were doing some verification of the examples that actually we had uh issue with the Matrix operator that at the restart was not working anymore. Basically and we realized we are not tests that has this kind of complex scenario. I now introduced a task that restarts the operator, but maybe nice. If we either have something that checks the traces or all of this server things we have going on in more complex scenario than just its own and it works.

A

B

Tough, it's a real implementation. Therefore, we need to configure The Collector.

F

Where we have expectation of how it has to look like.

B

It would be interesting to try Tracer test for this, but I think it's an Overkill.

F

Yeah, maybe phrased it wrong, I think the thing is we don't ever check any scenario apart from PLT is installed and everything works fine in the integration test and that's how we didn't figure out.

D

Let's say we're not very confident when it comes to updating dependencies for, like example, Jaeger or the open, Telemetry collector So. Currently, this requires a manual test to see if the traces are still working.

F

E

A

G

So I introduce a new test suite for E3 tests.

B

Yes, I would say.

B

Which is run on Main nightly, build something like that, because I think this might be heavy, so at the moment I will keep it minor run and afterwards after we see how this end-to-end test looks like we can Define. The pipeline now uh take a details here.

A

A

A

Produce new suite at very far is: let me correctly exposed just.

B

Metrics nah yeah, nah yeah just traces for first.

F

We have some information smallness.

B

There is Tracer test marks this.

A

A

B

We don't have a research label.

B

I will keep this as a research, so no estimation required, since it's pretty blurry what should be done.

B

And I will suggest that the goal of this ticket.

C

And you maybe add research to The Heading or something that it's clear.

B

And also at the moment, the goal should be just to document how the end-to-end test should should look like, and then we can create a flop ticket to implement that. So we can proper estimate.

A

uh Right or updated.

B

Why I'm saying all good with that or do you have a different opinion.

A

B

Okie dokie.

B

Then last ticket of today.

B

G

This is a bit older already, but I dug it up recently when I went through all our issues do we have and edited really for refinement level. I think this would be valuable to implement.

G

So basically, what Thomas wants here is: um yes,.

B

Not just the lock line.

G

Yeah something to make it easy to debug um metrics that are not configured correctly.

D

Instead of in event, would it be more fitting to put something in the status of the resource.

B

Exactly and my issue with event is, this is a periodic task so periodically we will emit the same event.

B

I see why once an event to integrate better with a or is it called? Okay, it's GPT, but I think an event is a bit of an Overkill here, since we present logs.

B

Which are events basically.

A

B

Maybe we can introduce a feature flag for this.

B

Report metrics error as events.

C

B

A good idea, I see one good, Morris and flow are more towards not really I'd rather have it just out.

G

Of the box, honestly, we.

F

Have a retry count, we could add something more okay, failure, also in the state, okay, daily resolution.

C

But the retry count is on evaluations. This is a little bit different. It's Captain metrics, it's fetching, um basically the data from the from videos, for example forever.

C

If you have a fetch interval five seconds every five seconds in case of a failure, it will send an event.

F

Yeah, that's why I would prefer to store it in the state of the resource.

E

C

Exactly well enabling an event um as a future flag. Why not? If the user wishes to do it, but.

F

Because this would stream events every few seconds producing polluting.

G

The only reason why we don't provide a lock message for this is because that's already pollute the breath.

B

G

A long message.

G

Maybe we do, but nobody will find it because that's already polluted. So maybe that's not blue. The next thing.

F

In my mind, the event stream should be for anything that crashes, the controllers. Not if the Matrix is not there, then it's not something that should be propagated.

C

Yeah, it's Captain metric controller, basically.

F

And no because the controller is not failing, it's just the metric. It's yeah, it's not to control your problem.

B

So kubernetes API Concepts event: where is it?

B

B

Do they have any recommendation for when an event should exist.

B

A third point, because there are.

C

State of the change of the state.

B

Limited retention, so there's no issue with having a too many events since we may discuss the handling by itself. I will say.

B

We try to change the system because we want to still something in the status field of the CRT, the he failed. So I see the point.

G

But I think usually you would then have a backup interval like if you think of odds not being able to be scheduled, you would have a you, would have a back off yeah so that it doesn't actually produce periodically.

B

But the kubernetes does filtering and merging of events.

B

So if we emit the same event over and over and over it would be just a single event and the count would be increased and.

C

Last scene, information when the event is that.

B

Let's go with the event and then, if it's too much, we can always remove it.

B

Contrary opinion for this.

B

Now it's your time to speak.

F

My understanding is kubernetes API releases, an event. There is a change in the state of the pods, so this is once it's failed. It's not that you have to be against time. God.

B

F

What is crashing.

B

You see exactly the count of that event. Increasing though.

E

B

F

That be the same case for us if.

B

We emit the same event over and over. Let's say that the first interval here is one second, so every second, we emit the same exact event. Every.

F

Second, the.

B

Counter will increase.

B

Then, let's restart okie dokie, no things here.

A

B

From the team or from the chat.

B

Don't forget to add your name here to the list.

B

I also thought of that.

B

Okie dokie, then, if there are no more topics, I will say thank everyone and I have a great rest of your day. Meeting.

E