简体中文 | English

Most of this article is translated from README_CN.md by ChatGPT 3.5. :)

kubectl-cache is a plugin for kubectl (the Kubernetes CLI client) used to get or list Kubernetes resources with local cache.

Traditionally, kubectl get ... get or list Kubernetes resources by sending a get or list request to the Kubernetes APIServer. However, processing list requests can be inefficient when there are a large number of objects. Frequent list requests can stress the APIServer and its storage backend (etcd), potentially affecting APIServer stability.

A common solution to this problem is to build a local cache of resources that need to be queried and detect changes in the cluster using the watch interface (to update the cache), as described in Efficient detection of changes. Kubernetes' Go client library, k8s.io/client-go, includes tools such as Informers (k8s.io/client-go/informers), which implement this access pattern.

Informer is great, but can it be used with kubectl commands?

The answer is kubectl-cache, which provides a way to use Informers with kubectl commands. kubectl-cache efficiently queries Kubernetes resources based on a local cache without the need for frequent list operations, transparently to the user. It provides a get subcommand that operates almost identically to kubectl get.

The usage is almost the same as kubectl get ...:

Comparison with kubectl get ... response speed:

Using the get subcommand (kubectl cache get or kubectl-cache get), you can get or list Kubernetes resources (similar to kubectl get). See Get or List Kubernetes Resources.

When you execute kubectl-cache's get subcommand, it checks if there is a running proxy service in the background (and starts a new one if not), then queries Kubernetes resources through this proxy. The proxy manages and queries the local cache using Informers. If there are no requests for 10 minutes, the proxy will exit automatically.

In addition to transparently starting a proxy through the get subcommand, you can also explicitly run a proxy for the Kubernetes APIServer locally using the proxy subcommand (kubectl cache proxy or kubectl-cache proxy), similar to kubectl proxy. See Running a Proxy.

For read requests (get and list), the proxy queries and returns results from the local cache (based on Informers). For write requests (create, update, patch, delete, deletecollection) and watch requests, the proxy forwards the requests directly to the Kubernetes APIServer.

kubectl-cache is a plugin for kubectl (see Extending kubectl with plugins), so it is recommended to install kubectl before installing kubectl-cache (see Install kubectl).

Krew is a plugin manager for kubectl. If you have Krew installed, you can install kubectl-cache with the following command:

kubectl krew install cache

Download the executable binary from the Releases page, extract it, and place the kubectl-cache file into any $PATH directory.

Requires Go 1.22. Execute the following command to download the source code and build it:

go install github.com/yhlooo/kubectl-cache/cmd/kubectl-cache@latest

The built binary will be located in ${GOPATH}/bin by default. Make sure this directory is included in your $PATH.

kubectl-cache can be used as a standalone CLI tool (via the kubectl-cache command) or as a plugin for kubectl (via the kubectl cache command).

You can get or list Kubernetes resources using the get subcommand (kubectl cache get or kubectl-cache get).

Some examples:

# List all Pods in the default namespace
kubectl cache get pod

# List all Pods in the default namespace with label app=test
kubectl cache get pod -l 'app=test'

# List all Pods in the default namespace in the Pending phase
kubectl cache get pod --field-selector 'status.phase=Pending'

# Get a Pod named test-pod in the default namespace
kubectl cache get pod test-pod

# Get a Pod named test-pod in the default namespace and output in YAML format
kubectl cache get pod test-pod -o yaml

# List all Pods in all namespaces
kubectl cache get pod -A

The kubectl cache get command behaves almost identically to kubectl get, with cache added before get.

For more options and usage, refer to kubectl cache get --help.

You can run a proxy for the Kubernetes APIServer locally using the proxy subcommand (kubectl cache proxy or kubectl-cache proxy):

# Run the proxy on port 8001
kubectl cache proxy --port 8001

In another terminal:

# Get or list Kubernetes resources via the proxy
kubectl --server https://127.0.0.1:8001 get pod

The kubectl cache proxy command behaves almost identically to kubectl proxy, with cache added before proxy.

For more options and usage, refer to kubectl cache proxy --help.

• The Field Selector only supports the = and == operators; the != operator is not supported
• For Custom Resources, the field selector only supports the metadata.name and metadata.namespace fields, even when implemented using the Aggregated API
• When printing custom resources implemented using the Aggregated API in the default table format, only the Name (metadata.name) and Age (metadata.creationTimestamp) columns are available
• When printing APIService and CustomResourceDefinition resources in the default table format, only the Name (metadata.name) and Age (metadata.creationTimestamp) columns are available
• In the list API, the limit, continue and resourceVersion queries do not work