Statefulset vs. Deployment + PVC?

[chrisbsmith]

I'm just getting started with Uptime Kuma and was looking for a helm chart for it (thanks btw!) and came across your repo. However, I was wondering why you chose to use a deployment + pvc instead of a statefulset to deploy the pod? I've had issues in the past where a new pod was unable to start because the terminating pod was still running and hanging onto the PVC.

My understanding is that this is the exact situation where a Statefulset is useful. Maybe I don't understand the uptime-kuma infrastructure requirements well enough though.

[chrisbsmith]

If interested, I have a successful PR into my fork that I could merge upstream.
chrisbsmith#2

[dirsigler]

Hey @chrisbsmith
the reason behind using a Deployment is just I didn't know better :D
Created the Chart via the common Helm commands which just provide basic skeletons.

I would really be interested your change proposals and I think you are complelety right that the Statefulset fits the application.

Please open a new Pull Request with your changes 🚀

[chrisbsmith]

Totally get it! I only knew about this from struggling with issues in the past.

Alright, submitted #11 #12. Let me know if you have any questions on it.

Thanks!!

[maxirus]

FYI: You had the right approach in the beginning with Deployment + PVC. StatefulSets are used to scale-out stateful workloads since you define volumeClaimTemplates which allow the K8s Controller to replicate PVCs as you increase replicas. Very useful for HA workloads.

Since uptime-kuma is using SQLite, it only allows for a single instance. Once support for PgSQL (#959) or MySQL (#953) is added, StatefulSets would make sense. But you'd also want to add the PgSQL/MySQL DB as a dependency.

[chrisbsmith]

Thanks for this explanation @maxirus. So in the current state, things are really a wash? Is there a benefit to reverting to Deployment + PVC? I should've done more research on the uptime-kuma model, but I don't see a harm to leaving it as a StatefulSet, especially if a team is working to add SQL support.

[maxirus]

@chrisbsmith Yes. "Weird" things will start to happen. I say weird in the sense that it's not what a typical user would expect. For example, you'd want Deployment to fail if the PVC is already attached in a scaling event. With a StatefulSet, it'd just create a new Volume, with new data, and in the event of a scale-in; you'd potentially lose the data.

[chrisbsmith]

@maxirus, got it. These make sense now that you describe them. These events you outlined aren't critical when the data is being persisted to a remote service. But when using a remote storage solution like SQL, you wouldn't need to put uptime-kuma in a STS anyways; instead you would do that with the backend.

@dirsigler apologies for push this change. At least we got to learn a little something here.

[maxirus]

But when using a remote storage solution like SQL, you wouldn't need to put uptime-kuma in a STS anyways; instead you would do that with the backend

Correct; unless for some reason they decide to persist config data, cache, etc. to disk. I'm not too familiar with the project as I don't intend to use it until a proper DB is supported. IMHO, it's important that your tool doing the monitoring be HA. In a world with Docker compose and K8s, this becomes trivial to use MySQL/PgSQL.

[dirsigler]

@chrisbsmith no worries!
I didn't know either, but I am happy to learn new things.

Would then say we revert the changes later on to again spin up a Deployment and PVC.

EDIT: and big thanks @maxirus for the insights ❤️

[beatkind]

@dirsigler my suggestion on this would be not to revert the changes but add a switch that changes the deployment from deploy to sts with a default value on deploy. So it is easy for the future to change that, as soon as Uptime Kuma is able to handle it and if someone wants to use a sts before that there is an easy way of doing that.

Of course, a breaking change in changing a default value must be communicated, but that's a bridge we'll cross when we come to it.

[dirsigler]

Interesting approach, indeed.
Already reviewed your changes and approved them.

With these changes we default to Deployments but keep the Statefulset "design" for the future.
As maxirus already mentioned here: #10 (comment) we then can also add later MySQL/Postgres as a dependency when all the changes were made on the upstream App.