kubectl-cost is a kubectl plugin
that provides easy CLI access to Kubernetes cost information, like historical cost
and predicted future cost, via Kubecost's APIs. It
allows developers, devops, and others to quickly determine the cost & efficiency
of any Kubernetes workload.
Note: If you use OpenCost, most of
kubectl costworks! See the OpenCost documentation for examples. Let us know how it goes, and open an issue if you encounter any problems!
kubectl cost can estimate the cost of undeployed changes! Try it
now with kubectl cost predict -f your-deployment.yaml:
OBJECT Δ QTY RESOURCE UNIT COST PER UNIT Δ COST/MO % CHANGE
────────────────────────────────────────────────────────────────────────────────────────────
kubecost1 deployment -92 CPU millicores 0.023 USD -2.14 USD -92.00%
kubecost1-cost-analyzer
-172.65 RAM MiB 0.0030 USD -0.53 USD -43.38%
-
Install Kubecost
This software requires that you have a running deployment of Kubecost in your cluster. The recommend path is to use Helm but there are alternative install options.
helm repo add kubecost https://kubecost.github.io/cost-analyzer/ helm upgrade -i --create-namespace kubecost kubecost/cost-analyzer --namespace kubecost --set kubecostToken="a3ViZWN0bEBrdWJlY29zdC5jb20=xm343yadf98" -
Install kubectl cost
If you have Krew, the
kubectlplugin manager, installed:kubectl krew install cost
The Krew manifest can be found here.
os=$(uname | tr '[:upper:]' '[:lower:]') && \ arch=$(uname -m | tr '[:upper:]' '[:lower:]' | sed -e s/x86_64/amd64/) && \ curl -s -L https://github.com/kubecost/kubectl-cost/releases/latest/download/kubectl-cost-$os-$arch.tar.gz | tar xz -C /tmp && \ chmod x /tmp/kubectl-cost && \ sudo mv /tmp/kubectl-cost /usr/local/bin/kubectl-cost
If you prefer to download from GitHub, or are on Windows, go to the releases page and download the appropriate binary for your system. Rename it to
kubectl-costand put it in yourPATH.As long as the binary is still named
kubectl-costand is somewhere in yourPATH, it will be usable. This is becausekubectlautomatically finds plugins by looking for executables prefixed withkubectl-in yourPATH.Alternatively, you can view DEVELOPMENT.md if you would like to build from source.
There are several subcommands that focused on aggregation-based cost monitoring:
namespace, deployment, controller, label, pod, and node. They display
cost information aggregated by the name of the subcommand (see Examples). Each
aggregation subcommand has two modes: rate and non-rate. Rate (the default) displays
the projected monthly cost based on the activity during the window. Non-rate
(--historical) displays the total cost for the duration of the window.
Starting with v1.100 installations of Kubecost, kubectl cost predict is
available. It uses historical resource cost information in your cluster to
predict the cost implications of undeployed changes. It currently supports
Pod, Deployment, and StatefulSet workloads, with more support on the way. It
accepts YAML-formatted data either in a file (-f your-file.yaml) or from
STDIN (-f -).
There is also kubectl cost tui, which displays a TUI and is currently limited to
monthly rate projections. It supports most of the above subcommands while in an
experimental status.
See the built-in usage info with --help to learn more about specific flags
available for each subcommand.
Show the projected monthly rate for each namespace with all cost components displayed.
kubectl cost namespace --show-all-resourcesExample output:
------------------- ----------- ---------- ---------- ------------- ---------- ---------- ---------- ------------- --------------------
| NAMESPACE | CPU | CPU EFF. | MEMORY | MEMORY EFF. | GPU | PV | NETWORK | SHARED COST | MONTHLY RATE (ALL) |
------------------- ----------- ---------- ---------- ------------- ---------- ---------- ---------- ------------- --------------------
| kube-system | 29.366083 | 0.066780 | 5.226317 | 0.928257 | 0.000000 | 0.000000 | 0.000000 | 137.142857 | 171.735257 |
| kubecost-stage | 6.602761 | 0.158069 | 1.824703 | 1.863699 | 0.000000 | 2.569600 | 0.000000 | 137.142857 | 148.139922 |
| kubecost | 6.499445 | 0.116629 | 1.442334 | 1.461370 | 0.000000 | 2.569600 | 0.000000 | 137.142857 | 147.654236 |
| default | 3.929377 | 0.000457 | 0.237937 | 0.283941 | 0.000000 | 0.000000 | 0.000000 | 137.142857 | 141.310171 |
| logging | 0.770976 | 0.003419 | 0.645843 | 0.260154 | 0.000000 | 0.000000 | 0.000000 | 137.142857 | 138.559676 |
| frontend-services | 0.710425 | 0.003660 | 0.595008 | 0.244802 | 0.000000 | 0.000000 | 0.000000 | 137.142857 | 138.448290 |
| data-science | 0.000284 | 2.000000 | 0.009500 | 2.000000 | 0.000000 | 0.000000 | 0.000000 | 137.142857 | 137.152641 |
------------------- ----------- ---------- ---------- ------------- ---------- ---------- ---------- ------------- --------------------
| SUMMED | 47.879350 | | 9.981644 | | 0.000000 | 5.139200 | 0.000000 | 960.000000 | 1023.000194 |
------------------- ----------- ---------- ---------- ------------- ---------- ---------- ---------- ------------- --------------------
Predict the cost impact of a YAML spec based on its requests:
read -r -d '' DEF << EndOfMessage
apiVersion: apps/v1
kind: Deployment
metadata:
name: nginx-deployment
namespace: default
labels:
app: nginx
spec:
replicas: 3
selector:
matchLabels:
app: nginx
template:
metadata:
labels:
app: nginx
spec:
containers:
- name: nginx
image: nginx:1.14.2
resources:
requests:
cpu: "3"
memory: "2Gi"
EndOfMessage
echo "$DEF" | kubectl cost predict -f -Example output:
OBJECT Δ QTY RESOURCE UNIT COST PER UNIT Δ COST/MO % CHANGE
────────────────────────────────────────────────────────────────────────────────────────
default deployment 9 CPU cores 23.27 USD 209.47 USD
nginx-deployment
6 RAM GiB 3.12 USD 18.72 USD
────────────────────────────────────────────────────────────────────────────────────────
TOTAL MONTHLY COST CHANGE 228.18 USD
Show how much each namespace cost over the past 5 days with additional CPU and memory cost and without efficiency.
kubectl cost namespace \
--historical \
--window 5d \
--show-cpu \
--show-memory \
--show-efficiency=falsePredict the cost of the Deployment defined in k8s-deployment.yaml.
kubectl cost predict -f 'k8s-deployment.yaml' \
--show-cost-per-resource-hrShow the projected monthly rate for each controller based on the last 5 days of activity with PV (persistent volume) cost breakdown.
kubectl cost controller --window 5d --show-pvShow costs over the past 5 days broken down by the value
of the app label:
kubectl cost label --historical -l appShow the projected monthly rate for each deployment based on the last month of activity with CPU, memory, GPU, PV, and network cost breakdown.
kubectl cost deployment --window month -AShow the projected monthly rate for each deployment
in the kubecost namespace based on the last 3 days
of activity with CPU cost breakdown.
kubectl cost deployment \
--window 3d \
--show-cpu \
-n kubecostThe same, but with a non-standard Kubecost deployment
in the namespace kubecost-staging with the cost
analyzer service called kubecost-staging-cost-analyzer.
kubectl cost deployment \
--window 3d \
--show-cpu \
-n kubecost \
-N kubecost-staging \
--service-name kubecost-staging-cost-analyzerShow how much each pod in the "kube-system" namespace cost yesterday, including CPU-specific cost.
kubectl cost pod \
--historical \
--window yesterday \
--show-cpu \
-n kube-systemAlternatively, kubectl cost can show cost by the asset type. To view node cost with breakdowns of RAM and CPU cost for a window of 7 days.
kubectl cost node \
--historical \
--window 7d \
--show-cpu \
--show-memoryWhich yields an output with this format:
------------- --------------------------------------------- --------------- -------------- ---------------
| CLUSTER | NAME | CPU COST | RAM COST | TOTAL COST |
------------- --------------------------------------------- --------------- -------------- ---------------
| cluster-one | gke-test-cluster-default-pool-d6266c7c-dqms | 4.128570 | 2.128920 | 6.257491 |
| | gke-test-cluster-pool-1-9bb98ef8-3w6g | 4.128570 | 2.128920 | 6.257491 |
| | gke-test-cluster-pool-1-9bb98ef8-cf3j | 4.128570 | 2.128924 | 6.257495 |
| | gke-test-cluster-pool-1-9bb98ef8-kdsf | 4.128570 | 2.128924 | 6.257495 |
------------- --------------------------------------------- --------------- -------------- ---------------
| SUMMED | | USD 16.514280 | USD 8.515688 | USD 25.029972 |
------------- --------------------------------------------- --------------- -------------- ---------------
See kubectl cost [subcommand] --help for the full set of flags. Each
subcommand has its own set of flags for adjusting query behavior and output.
There are several flags that modify the behavior of queries to the backing Kubecost/OpenCost APIs:
-r, --release-name string The name of the Helm release, used to template service names if they are unset. For example, if Kubecost is installed with 'helm install kubecost2 kubecost/cost-analyzer', then this should be set to 'kubecost2'. (default "kubecost")
--service-name string The name of the Kubecost cost analyzer service. By default, it is derived from the Helm release name and should not need to be overridden.
--service-port int The port of the service at which the APIs are running. If using OpenCost, you may want to set this to 9003. (default 9090)
-N, --kubecost-namespace string The namespace that Kubecost is deployed in. Requests to the API will be directed to this namespace. Defaults to the Helm release name.
--use-proxy Instead of temporarily port-forwarding, proxy a request to Kubecost through the Kubernetes API server.
--allocation-path string URL path at which Allocation queries can be served from the configured service. If using OpenCost, you may want to set this to '/allocation/compute' (default "/model/allocation")
--predict-speccost-path string URL path at which Prediction queries can be served from the configured service. (default "/model/prediction/speccost")
--no-usage Set true ignore historical usage data (if any exists) when performing cost prediction.
--opencost Set true to configure Kubecost parameters according to the OpenCost default specification. It is equivalent to providing the options '--service-port 9003 --service-name opencost --kubecost-namespace opencost --allocation-path /allocation/compute'
--only-after Set true to only show the overall predicted cost of the workload.
--only-diff Set true to only show the cost difference (cost "impact") instead of the overall cost plus diff. (default true)
kubectl cost has to interact with the Kubernetes API server. It tries to use your existing kubeconfig. These flags are common to kubectl and allow you to customize this behavior:
--as string Username to impersonate for the operation
--as-group stringArray Group to impersonate for the operation, this flag can be repeated to specify multiple groups.
--cache-dir string Default cache directory (default "/home/delta/.kube/cache")
--certificate-authority string Path to a cert file for the certificate authority
--client-certificate string Path to a client certificate file for TLS
--client-key string Path to a client key file for TLS
--cluster string The name of the kubeconfig cluster to use
--context string The name of the kubeconfig context to use
-h, --help help for cost
--insecure-skip-tls-verify If true, the server's certificate will not be checked for validity. This will make your HTTPS connections insecure
--kubeconfig string Path to the kubeconfig file to use for CLI requests.
--request-timeout string The length of time to wait before giving up on a single server request. Non-zero values should contain a corresponding time unit (e.g. 1s, 2m, 3h). A value of zero means don't timeout requests. (default "0")
-s, --server string The address and port of the Kubernetes API server
--tls-server-name string Server name to use for server certificate validation. If it is not provided, the hostname used to contact the server is used
--token string Bearer token for authentication to the API server
--user string The name of the kubeconfig user to usekubectl cost logs some of its behavior at the debug log level. If something isn't working as you'd expect, try setting --log-level debug before opening a bug report.
In order to provide a seamless experience for standard Kubernetes
configurations, kubectl-cost temporarily forwards a port on your system to a
Kubecost pod and uses that port to proxy requests. The port will only be bound
to localhost and will only be open for the duration of the kubectl cost run.
Due to Linux default conventions, the port may appear as held for a little while
after the run (see TCP's TIME_WAIT).
If you don't want a port to be temporarily forwarded, there is legacy behavior
exposed with the flag --use-proxy or using environment
KUBECTL_COST_USE_PROXY that will instead use the Kubernetes API server to
proxy requests to Kubecost. This behavior has its own pitfalls, especially with
security policies that would prevent the API server from communicating with
services. If you'd like to test this behavior, to make sure it will work with
your cluster:
kubectl proxy --port 8080curl -G 'http://localhost:8080/api/v1/namespaces/kubecost/services/kubecost-cost-analyzer:tcp-model/proxy/getConfigs'If you are running an old version of Kubecost, you may have to replace
tcp-modelwithmodel
If that curl succeeds, --use-proxy flag in CLI or setting up environment variable KUBECTL_COST_USE_PROXY should work for you.
Otherwise:
- There may be an underlying problem with your Kubecost install, try
kubectl port-forwarding thekubecost-cost-analyzerservice, port 9090, and querying one of our APIs. - Your problem could be a security configuration that is preventing the API server communicating with certain namespaces or proxying requests in general.
- If you're still having problems, hit us up on Slack (see below) or open an issue on this repo.
A cluster running Kubernetes version 1.8 or higher
Have questions? Join our Slack community or contact us via email at [email protected]!