Get a shell in a workload

Running a command inside something usually means finding a pod first: list them, read a generated name off the table, then type it. kx exec takes the index of the workload itself.

kx get deploy
kx exec 1              # a shell in one of that Deployment's pods
kx exec 1 -- ps aux    # a command instead

With no command it tries each configured shell in turn — bash, then sh, unless the shells key says otherwise. That matters for distroless and Alpine images, where the shell you assume is not the one that’s there.

What it accepts

Pods, Deployments, ReplicaSets, StatefulSets and DaemonSets.

Services are refused. kubectl exec does not accept one, and kx says so itself rather than letting kubectl produce a worse message about it.

kx does not pick the pod — kubectl exec resolves the workload, the same way kx port-forward leaves that choice to kubectl. Which pod you get is therefore not guaranteed to be the same one across the shell probe and the session that follows. For a Deployment whose replicas are interchangeable that is what you want; when it isn’t, address a pod directly.

Exit codes come back

The command’s own exit code is forwarded, so kx exec works inside a script:

kx exec 1 -- test -f /etc/config.yaml || echo "config missing"

kubectl’s own stderr is suppressed here, because a failing command inside a container otherwise prints “command terminated with exit code N” on top of whatever the command already said.

Reports in the browser Take a node out of service