Update prometheus-operator deployments about pangeo-binder HOT 5 CLOSED

Go ahead whenever!

from pangeo-binder.

Cool, the kube-prometheus-stack readme has a section on migrating from prometheus-operator.

Looks like the prometheus-operator chart got taken out recently: 95f5fba#diff-06e805e5c3973f9c240e7dcb407c3d58L18-L20. I think this was an accident @TomAugspurger , but hey, it gives me a clean slate for this.

I'm deleting all the CRDs.

I'm having some CRD installation problems. If I just put the new helm chart into the dependencies of pangeo-binder and try the helm install command, I get several lines like

unable to recognize "": no matches for kind "Alertmanager" in version "monitoring.coreos.com/v1"

If I try to install the CRDs as listed in the readme, I get errors such as

error: unable to recognize "<filepath-omitted>": no matches for kind "CustomResourceDefinition" in version "apiextensions.k8s.io/v1"

The versions listed there are for prometheus-operator v0.42. We used to be using v0.38, which have apiVersion: apiextensions.k8s.io/v1beta1. There is also another resource listed, probes.monitoring.coreos.com, but this resource isn't present in v0.38, and the other CRDs do not have a cooperative apiVersion after v0.38.

There are also CRDs listed in the crds/ folder of the repo, so I'm guessing that they haven't gotten around to putting those in the readme. Trying to install these charts (from the raw.githubusercontent links) shows errors:

no matches for kind "CustomResourceDefinition" in version "apiextensions.k8s.io/v1"

as with the v0.42 prometheus-operator CRDs.

This latest version of kube-prometheus-stack is v10.0.1, which was released 2 hours ago. I'll take a break on this and see if the helm chart and it's instructions become more up-to-date later.

A final thought: kubectl version says my local machine is on v1.17.4 but the AWS pangeo-binder is on v1.15.11. I'm wondering if the apiVersions aren't working because of that.

from pangeo-binder.

@TomAugspurger we don't need to make a decision on this quickly (JupyterCon and ADSA 2020) will keep me busy this week), but if this isn't resolved next week, I'm thinking of just going back to Prometheus and Grafana as individual charts. I'm more confident that this will work at a basic level (especially since it's easier to pin to older versions that we know work). Let me know if you have any thoughts on anything here.

from pangeo-binder.

+100 to simplification. I didn't realize that prometheus-operator was so complicated when I picked it.

from pangeo-binder.

Checked the kube-prometheus-stack repo again today, no significant changes, and the readme didn't have different instructions, so I went and manually tried to deploy Prometheus and Grafana on the AWS staging binder.

@TomAugspurger I've got a configuration working for the two individual charts. I see you have an open PR, should I wait on that or just open up my PR and let you take a look?

from pangeo-binder.