GitLab Incubation Engineering, 23 Dec 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Incubation Engineering Server Runtime - Weekly Demo 2022-12-23

Description

This is the weekly update for the Server Runtime SEG for the week of the 23rd of December 2022

This week I've continued on build on the excellent GitLab agent for Kubernetes. Updates include:

1. Ability to change workspace state
Ability to start, stop and terminate a workspace

2. Reflect state from Kubernetes
Server Runtime will reflect the state of the workspace in Kubernetes

3. Retroactively fix state
Server Runtime will now refresh the state of the workspaces it is managing when the agent first starts up.

Chapters
0:00 Overview
0:49 Demo - Changing workspace state
1:49 Demo - Reflect state from Kubernetes
2:42 Demo - Retroactively fix state

A

Hi, everyone welcome to another update from the server runtime sag for the week of the 23rd of December 2022.. So this week, I've been really focused on state changes. The ability to start stop a Terminator workspace. Also, what I've been working on is reflecting changes in kubernetes. So, for example, when uh um workspace gets deprovisioned because of a node being unavailable or any changes in the kubernetes in the kubernetes namespace, those will be reflected automatically in rails.

A

uh What I'm also doing is making sure that in case the agent is down, for example, our gitlab agent isn't working the cluster when it does come up. It retrospectively fixes the state, so that rails always reflects the right state. So, with that I'll jump straight into the demo.

A

Okay, so here you can see, I have two workspaces running so workspace, 36 and 37.. If I go into kubernetes I can see that workspace. 36 is currently in a running State. I can go ahead and change that state by clicking on stop and when I click the stop button, you should see instantaneously. It starts terminating uh that that status reflects from kubernetes and so I can see. The status is now stopped, so I could go ahead and also terminate this workspace, so it is currently stopped.

A

So it's not consuming any CPU, um but we can go ahead and also completely terminate it. So this will completely kill the isolation. All the namespaces everything all the resources related to this workspace will be killed. So if you see that it's now in a terminating state, if you see in kubernetes, if I look at the name spaces, uh you can see that the namespace uh workspace 36 is in a terminating State as well. So if I refresh the screen, you should soon see it is terminated.

A

So once the computer is killed and we go ahead and terminate that workspace. um So now you can change the status of the workspace.

A

uh The next thing I want to quickly show you is that, if that, if kubernetes changes the status, that is you know something goes wrong in kubernetes and kubernetes, just not able to bring that pod up uh we're able to reflect that status in in um in real as well so I'm going to directly go into a different workspace, workspace, 37 I'm, going to go ahead and edit that uh deployment um for that workspace and I'm going to stop that deployment.

A

Essentially so right now you see the rails shows the status of of running so I'm going to go ahead and make the replica 0, which will go ahead to this deployment, and so, if you see almost immediately, if I refresh rails, you'll see that it shows the status has stopped. uh So as things change in kubernetes you're. Listening to all these events and then, as soon as these events occur, we immediately reflect the status of the events in in rails as well.

A

uh The other other great thing is: um we can even do this uh when the agent is down so brought the agent down. uh So agents are running, let's say in between uh the workspace started running automatically uh so agent's not aware of this. So agent isn't uh at all aware of the changes, and so let's say the workspace started running uh at this point in time.

A

A

So the the workspace is currently starting up in running status. um So if I go uh right now to kubernetes kubernetes sorry to rails, you can see real still things that stopped. What I can go ahead and do is if I start up the agent now, uh the agent will start up and it'll reflect the status automatically, and this happens because I've now introduced a new flag in the configuration called resync on Startup and what this flag does. Is it automatically resyncs the status on when the uh when the agent starts up?

A

So now, if I refresh the screen, you can see that the workspace is up and running uh the next thing. I want to quickly show you is even if the workspace is completely killed. We do a full reconciliation. So if I was to go ahead and completely delete this namespace so that there was no trace of this uh workspace at all in kubernetes.

A

So that is now uh almost terminated um so now there's no. The agent doesn't know that this um that this uh this has been terminated. If I go ahead and actually refresh here, you can go and see that as soon as I start the agent uh the status is reflected. So we do a quick, full reconciliation. So even when the agent is down um and the workspace workspace is terminated, we can we can see those events, we do a comparison and we delete it.

A

um So so that's it for this update. uh Thank you for listening. uh This is the last update for the year, uh so have a great year ahead. Thank you.