How to Restart All Pods in a Kubernetes Namespace

Where I work, we use a repo-per-namespace setup and so it often happens that I want to restart all pods and deployments in a single Kubernetes namespace. Maybe I want to see the startup logs, or maybe I want to shut down production for a few seconds. Don’t question my motives.

Anyway, what matters is that bouncing all deployments one after another is inconvenient and I don’t like typing.

The best way to bounce (kubectl >= 1.15)

I recently found out from a friend that there is an easier way as of kubectl 1.15+. Restarting all the pods in a namespace is as easy as running the following kubectl command.

kubectl -n {NAMESPACE} rollout restart deploy

The old way (kubectl <= 1.14)

In older versions of kubectl you needed to run a command for each deployment in the namespace. In true lazy developer fashion I wrote a little script that will do it for me:

deploys=`kubectl -n $1 get deployments | tail -n +2 | cut -d ' ' -f 1`
for deploy in $deploys; do
  kubectl -n $1 rollout restart deployments/$deploy
done

It’s fairly simple to use. Assuming I named the script kubebounce.sh:

./kubebounce.sh {NAMESPACE}

I made a little open-source repo with installation instructions if you want to add it to your $PATH. Be sure to star the repo if you find it useful.

How It Works

Bash isn’t exactly the easiest language to read. Let’s go over each portion of the script.

deploys=`kubectl -n $1 get deployments | tail -n +2 | cut -d ' ' -f 1`

In bash, $1 refers to the first command-line argument, the namespace in our case. In essence, this line gets all the deployments in the target namespaces and saves them into a deploys variable. We pipe the output of the kubectl get deployments command into a tail -n +2 command, which just strips off the first line of the output. Then we run that output through a cut command which leaves us with a nice list of all the deployment names.

That’s the trickier part, next we just loop over all the deployments and restart them one by one:

for deploy in $deploys; do
  kubectl -n $1 rollout restart deployments/$deploy
done

PS: I’m the author of a fully interactive Kubernetes course on Boot.dev if you’d like to learn more about Kubernetes.