Research Kubecost

[tannavee]

No description provided.

[rkevin-arch]

https://github.com/kubecost/cost-model

https://www.xenonstack.com/insights/cost-analysis-kubernetes/

[moorepants]

I added low priority, as getting galaxy working first is the higher priority. This is something that will be needed down the line.

[sandertyu]

https://docs.kubecost.com/getting-started.html (most useful link)
https://docs.kubecost.com/cost-allocation.html
http://blog.kubecost.com/blog/cost-monitoring/

Installation looks fairly simple, and the interface is accessible at http://localhost:9090 after running kubectl port-forward --namespace kubecost deployment/kubecost-cost-analyzer 9090. Full installation instructions are here, but you may need to put in an email before the page is accessible.

Kubecost also works via a modified sort of Prometheus metric collection, and it is possible (although nuanced) to integrate Kubecost with existing Prometheus/Grafana deployments. More info on that here. Full integration with existing Prometheus is only officialy supported with the paid version of Kubecost. We'd have to try and do the integration ourselves to see how difficult it is, but if we can get it to work, then this is another reason to use Prometheus rather than InfluxDB.

[mandeepika]

General: Kubecost provides cost monitoring and capacity management solutions (writing it for myself because I didn't know what Kubecost even is). It lets you do cost monitoring, cost allocation, manage out of cluster costs, and send notifications in cases of potential problems if detected. It determines the cost from: time in the running state, resources consumed or reserved, and prices of resources consumed and reserved.

Cost/Plan: The Kubecost plan will be free for 30 days and then have to buy a plan as mentioned here. The plan goes something like $199/month for 50 nodes with unlimited clusters and increases if we want more nodes. It will let us save metric reports and data for 30 days.

Installation: As mentioned here, the requirements of the tool are: Kubernetes 1.8 or higher, Kube-state-metrices, Node exporter, Prometheus, Grafana. (I don't know what we have and not have). Prometheus will be used as a database for gathering the metrices and Grafana will be used as a visualization tool to display info. So, Noah is right we will want to use Prometheus. Also, on here, they mention they recommend using the bundled prometheus-server and grafana, and reusing the existing kube-state-metrices and node exporter deployments for easiest installation process and maintainance. However, if we do have existing Prometheus and Grafana, the process of integration is a little bit complicated and is given here. We will need to contact [email protected] if we choose the latter option and want further help with this process.