kubectl-fzf provides a fast and powerful fzf autocompletion for kubectl.



  • Seamless integration with kubectl autocompletion
  • Sub second completion
  • Label autocompletion
  • Automatic namespace switch


  • go (minimum version 1.12)
  • fzf


Cache builder

Local installation

Install cache_builder:

# Mac
# Linux

cd /tmp
wget "https://github.com/bonnefoa/kubectl-fzf/releases/latest/download/$FILE"
tar -xf $FILE
install cache_builder ~/local/bin/cache_builder

As a kubernetes deployment

You can deploy the cache builder as a pod in your cluster.

helm template --namespace myns --set image.cache_builder.tag=1.0.11 --set toleration=aToleration . | kubectl apply -f -

You can check the latest image version here.

Shell autocompletion

Source the autocompletion functions:

# kubectl_fzf.sh needs to be sourced after kubectl completion.

# bash version
wget https://raw.githubusercontent.com/bonnefoa/kubectl-fzf/master/kubectl_fzf.sh -O ~/.kubectl_fzf.sh
echo "source <(kubectl completion bash)" >> ~/.bashrc
echo "source ~/.kubectl_fzf.sh" >> ~/.bashrc

# zsh version
wget https://raw.githubusercontent.com/bonnefoa/kubectl-fzf/master/kubectl_fzf.plugin.zsh -O ~/.kubectl_fzf.plugin.zsh
echo "source <(kubectl completion zsh)" >> ~/.zshrc
echo "source ~/.kubectl_fzf.plugin.zsh" >> ~/.zshrc

Using zplug

You can use zplug to install the autocompletion functions

zplug "plugins/kubectl", from:oh-my-zsh, defer:2
zplug "bonnefoa/kubectl-fzf", defer:3


cache_builder: local version

cache_builder will watch cluster resources and keep the current state of the cluster in local files.
By default, files are written in /tmp/kubectl_fzf_cache (defined by KUBECTL_FZF_CACHE)

To create cache files necessary for kubectl_fzf, just run in a tmux or a screen


It will watch the cluster in the current context. If you switch context, cache_builder will detect and start watching the new cluster.
The initial resource listing can be long on big clusters and autocompletion might need 30s+.

connect: connection refused or similar messages are expected if there’s network issues/interruptions and cache_builder will automatically reconnect.


You can configure cache_builder with the configuration file $HOME/.kubectl_fzf.yaml

# Role to hide from the role list in node completion
  - common-node
# Namespaces to exclude for configmap and pod listing
# Regexps are accepted
  - consul-agent
  - datadog-agent
  - coredns
  - kube-system
  - kube2iam
  - dev-.*


  • Minimal setup needed.


  • It can be CPU and memory intensive on big clusters
  • It also can be bandwidth intensive. The most expensive is the initial listing at startup and on error/disconnection. Big namespace can increase the probability of errors during initial listing.

cache_builder: pod version

If the pod is deployed in your cluster, the autocompletion fetch the cache files from the pod using rsync.

To check if you can reach the pod, try:

IP=$(k get endpoints -l app=kubectl-fzf --all-namespaces -o=jsonpath='{.items[*].subsets[*].addresses[*].ip}')
nc -v -z $IP 80

By default, it will use the port 80.
You can change it by deploying the chart with a different port value and using KUBECTL_FZF_RSYNC_PORT:

helm template --namespace myns --set port=873 . | kubectl apply -f -

If there’s no direct access, a port-forward will be opened.
Opening a port-forward can take several seconds and thus, slow the autocompletion.
If you want to avoid this, you can keep the open_port_forward.sh script running which will keep the port-forward opened.


  • No need to run a local cache_builder
  • Bandwidth usage is limited to the rsync transfert which is low.


  • Rsynced resources are cached for 30 seconds so the autocompletion can get outdated.


Once cache_builder is running, you will be able to use kubectl_fzf by calling the kubectl completion

kubectl get pod <TAB>


Environment variable Description Default
KUBECTL_FZF_CACHE Cache files location /tmp/kubectl_fzf_cache
KUBECTL_FZF_EXCLUDE Exclusion patterns passed to the autocompletion “”
KUBECTL_FZF_OPTIONS fzf parameters -1 --header-lines=2 --layout reverse -e --no-hscroll --no-sort

To turn down exact match in search:

export KUBECTL_FZF_OPTIONS=(-1 --header-lines=2 --layout reverse)

To enable hscroll

export KUBECTL_FZF_OPTIONS=(-1 --header-lines=2 --layout reverse -e)

To bind space to accept current result (You can check the list of available keys and actions in the fzf man)

export KUBECTL_FZF_OPTIONS=(-1 --header-lines=2 --layout reverse -e --no-hscroll --no-sort --bind space:accept)

To exclude all namespaces starting with “dev” and consul-agent resources:

export KUBECTL_FZF_EXCLUDE=("^dev" "consul-agent")


With zsh, if the suggested completion doesn’t match the start of the query, the completion will fail.

k get pod pr<TAB>
# result needs to start with `pr` otherwise autocompletion will fail

If you’re using an out-of-the-box oh-my-zsh configuration, the default matcher-list zstyle (zstyle ':completion:*' matcher-list 'r:|=*' 'l:|=* r:|=*') will interfere with the search. If fzf does not find any match, or if you interrupt it by pressing Esc or Ctrl-C/Cmd-C, zsh will see it as a failed completion and will restart it again.

Changing the zstyle to zstyle ':completion:*' matcher-list 'r:|=*' fixes the issue.


Debug logs

To launch cache_builder with debug logs

cache_builder -logtostderr -v 14

The normal autocompletion is used

First, check if cache files are correctly generated in /tmp/kubectl_fzf_cache.
The autocompletion will fallback to normal method if cache files are absent.

If the files are present, check that the __kubectl_get_containers is correctly overloaded

# Incorrect type
type __kubectl_get_containers
__kubectl_get_containers is a shell function from /dev/fd/15

# Expected output
type __kubectl_get_containers
__kubectl_get_containers is a shell function from .../kubectl-fzf/kubectl_fzf.plugin.zsh

Be sure that kubectl_fzf.plugin is loaded after kubectl completion zsh in your bashrc/zshrc.


View Github