[GKE] [Cloud/managed K8S] Scaphandre pod is always pending in Kubernetes cluster #277
-
Bug descriptionI tried to install Scaphandre on a basic Kubernetes cluster hosted on GKE but the Scaphandre node have been pending for more than 24 hours since installation. The status of the pods shown by To Reproduce1 - Setup a basic Kubernetes cluster on GKE Expected behaviorThe pods should be running like Grafana and Prometheus related pods ScreenshotsEnvironment
Additional contextAlso, no metrics |
Beta Was this translation helpful? Give feedback.
Replies: 5 comments
-
Hello, thanks for opening the issue. I believe that GKE (and other flavor of public cloud provider managed kubernetes) may not provide access to RAPL. I believe you need to deploy Scaphandre to a physical/bare metal server (like an on premise kubernetes cluster) that do not restrict access to RAPL. Maybe this explains why the pod does not starts up well (although in my opinion, it should start but still report empty of zero metrics). |
Beta Was this translation helpful? Give feedback.
-
Thank you, that seems the problem indeed, GCP does not give access to the RAPL metrics. Do you know any other way to get the power consumption in a managed cloud without relying on RAPL ? |
Beta Was this translation helpful? Give feedback.
-
Hi,
|
Beta Was this translation helpful? Give feedback.
-
Hi @bpetit, maybe we could close the issue (or even better move it to a discussion in github ?). |
Beta Was this translation helpful? Give feedback.
-
Also to mention https://github.com/Boavizta/Energizta/ that should allow to build accurate models to do so at some point ! Yes @demeringo converting as a discussion is a good idea ! Thanks |
Beta Was this translation helpful? Give feedback.
Hello, thanks for opening the issue.
I believe that GKE (and other flavor of public cloud provider managed kubernetes) may not provide access to RAPL.
This may explain the lack of metrics.
I believe you need to deploy Scaphandre to a physical/bare metal server (like an on premise kubernetes cluster) that do not restrict access to RAPL.
Maybe this explains why the pod does not starts up well (although in my opinion, it should start but still report empty of zero metrics).