Quick reference for navigating and operating a Kubernetes cluster with k9s. Everything is driven from command mode (​:​) to switch resource views; single letters then act on the highlighted row.

k9s                          # default context, all namespaces
k9s -n my-namespace          # start in a namespace
k9s --context staging        # start in a specific context
k9s --readonly               # disable all mutating ops (prod/demos)
k9s -c pods                  # jump straight to a resource view
k9s --kubeconfig ~/.kube/x   # explicit kubeconfig

Open with ​:​ then type a name or alias.

​:ctx switches context, ​:ns switches namespace, ​:q (or repeated Esc) quits. For CRDs use the kind, or the fully-qualified kind.group form if the short alias is ambiguous.

Highlight a row, then:

From a pod or container, l opens the log view:

For multi-container pods, Enter into the pod first, then l on the specific container.

• /text — fuzzy filter rows by name
• /-l app=foo — filter by label selector
• /!text — inverse filter (hide matches)
• Enter keeps the filter; Esc clears it
• Sort columns: Shift-c (CPU), Shift-m (mem), Shift-n (name), Shift-a (age), Shift-s (status), Shift-r (restarts) — press again to reverse

:ctx                  # interactive context switcher
:ctx staging          # jump directly
:ns                   # namespace switcher

Ctrl-a lists every resource alias the current cluster exposes (including CRDs) — fastest way to explore an unfamiliar cluster.

There is no single “network” view — networking spans several resource types plus the port-forward machinery.

If your CNI exposes CRDs they're reachable too — e.g. ​:ippools, ​:ciliumendpoints, ​:ciliumnetworkpolicies. Use Ctrl-a to list what the cluster actually has.

1. Highlight a pod or service → Shift-f → pick container port, local port, bind address → confirm.
2. ​:pf shows all active forwards. Ctrl-d cancels one. Forwards die when k9s exits.
3. From a forward, Ctrl-b runs k9s's built-in HTTP load test; results land in ​:benchmarks (​:be), where you can view or delete them.

​:svc → d to confirm the selector and ports → Enter to verify it has endpoints (empty endpoints = selector mismatch or no ready pods) → ​:po, find a backing pod, s to shell in and test connectivity from inside the cluster. ​:xray svc gives a tree of service → endpoints → pods for topology at a glance.

• –readonly (or readOnly: true in config) is worth defaulting on for any prod-touching context — greys out every mutating key.
• Per-context skins are the best safeguard against acting on the wrong cluster — make prod red.
• Ctrl-z toggles “wait” mode if the screen looks frozen during heavy refreshes; lowering refreshRate helps on large clusters.
• If a CRD doesn't show under its short alias, use kind.group or find it via ​:crd.
• ​:pulse is a cluster health dashboard; ​:xray <resource> shows a dependency tree.
• ? in any view lists the keys valid right there.