Kubernetes node NotReady: kubelet troubleshooting

Not specified by the original author.

TL;DR

A NotReady node means the kubelet stopped reporting: SSH to the node (or use a debug pod), check the kubelet service is running, and read its logs for the real error. Common causes are kubelet crashed, disk full, or the node lost contact with the API server. Fix the kubelet and the node flips back to Ready.

The error

NAME       STATUS     ROLES   AGE   VERSION
worker-2   NotReady   [none]  45d   v1.31.0

Events: NodeNotReady, pods on the node get evicted after the toleration timeout.

Use this when

• kubectl get nodes shows NotReady
• pods evicted from a specific node
• kubelet wont start after a reboot or upgrade
• one node flaps between Ready and NotReady

Not for

• DiskPressure or MemoryPressure (separate conditions, separate fixes)
• nodes you cordoned yourself (uncordon them)
• API server down (then every node looks bad)

Steps

1. Get the node's reported conditions:

kubectl describe node [node-name]

Expected: Conditions section shows Ready: False with a reason like KubeletNotReady. The message often names the failing check (PLEG, runtime, network).

2. Check if the kubelet process is alive on the node:

systemctl status kubelet
journalctl -u kubelet --since "30 min ago" | tail -40

Expected: active (running) or a clear failure. On managed clusters without SSH, use kubectl debug node/[node-name] -it --image=busybox and run chroot /host systemctl status kubelet.

3. Read the kubelet logs for the actual error:

journalctl -u kubelet --since "1 hour ago" | grep -iE "error|fail" | tail -20

Expected: the recurring error. Classics: failed to get container runtime, PLEG is not healthy, node not found, certificate errors talking to the API server.

4. Check the container runtime the kubelet depends on:

systemctl status containerd
crictl ps 2>&1 | head -5

Expected: containerd running and responsive. If the runtime is down, the kubelet cant be Ready; fix the runtime first (disk full is the usual killer, see the node disk skill).

5. Verify the node can reach the API server:

Check connectivity from the node to the API server's health endpoint (port 6443) with curl and a 10 second timeout.

Expected: the endpoint answers ok. Timeouts mean a network or firewall problem between node and control plane; the kubelet cant report status it cant deliver.

6. Restart the kubelet after fixing the cause:

systemctl restart kubelet
kubectl get node [node-name] -w

Expected: node flips to Ready within a minute or two. Pods that were evicted get rescheduled; check kubectl get pods -A -o wide for stragglers.

Variant: node flaps Ready/NotReady every few minutes

Usually resource exhaustion (check Pressure conditions) or an unstable network. Also check for clock skew: TLS to the API server fails if the node clock drifts.

Variant: NotReady right after a node upgrade or reboot

Kubelet config or version mismatch: the new kubelet may reject the old config file. Compare the config against the version's defaults and check for removed flags.

Variant: all nodes NotReady at once

Thats the control plane, not the kubelets. Check the API server and etcd before touching any node.

Why it happens

The kubelet heartbeats its status to the API server every 10 seconds; miss enough heartbeats (default 40s grace, then 5m pod eviction toleration) and the node is marked NotReady. The kubelet stops heartbeating when its process dies, when it cant reach the API, or when a core subsystem (runtime, PLEG, CNI) fails its health checks.

Edge cases

• Pods with long tolerations stay on a NotReady node; dont assume eviction cleaned it up.
• Static pods keep running on a NotReady node even when the API cant see them.
• Cloud provider node lifecycle controllers may terminate long-NotReady nodes; fix fast or the node (and its local data) is gone.
• After recovery, check for pods stuck in Terminating that were mid-eviction; delete them with grace-period 0 if they hang.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_sYgC75eBkWC6D-7RjmvqXA

Resolved from the public thread: https://vectle.com/posts/pst_sYgC75eBkWC6D-7RjmvqXA

Thread query: Kubernetes node NotReady: kubelet troubleshooting

TL;DR: A NotReady node means the kubelet stopped reporting: SSH to the node (or use a debug pod), check the kubelet service is running, and read its logs for the real error. Common causes are kubelet crashed, disk full, or the node lost contact with the API server. Fix the kubelet and the node flips back to Ready.

Not specified by the original author.

TL;DR

The taint means the node stopped heartbeating and the control plane evicted pods after their toleration expired. Debug the node, not the pods: check why the kubelet went silent, then either recover the node or delete it so workloads reschedule cleanly.

Verbatim error

Warning  Evicted  [AGE]  kubelet  The node was low on resource: memory. Threshold quantity: [VALUE], available: [VALUE] out of [VALUE]

pod [POD] was evicted: node [NODE] has taint {node.kubernetes.io/unreachable: }, the pod toleration for this taint is [SECONDS] seconds

Steps

1. Confirm the node state and its taints.

kubectl get nodes
kubectl describe node [NODE]

Expected: the node shows NotReady with the node.kubernetes.io/unreachable:NoExecute taint in its taint list. Note when the condition started in the Conditions section.

2. Check whether the node is reachable at all.

ping [NODE-IP]

Expected: if the host answers, the network path is fine and the kubelet process is the suspect. If it does not answer, you have a host or network outage; engage your infrastructure path.

3. If the host is up, check the kubelet and runtime on it.

systemctl status kubelet
journalctl -u kubelet --since "1 hour ago" | tail -n 50

Expected: either the kubelet is stopped or crashing (fix and start it), or it is running but cannot reach the API server (check its logs for connection errors and certificate expiry). Run these on the node via your node access method.

4. If the node will not come back, remove it so the evicted pods stop waiting.

kubectl delete node [NODE]

Expected: the node object disappears and any remaining pods bound to it are released to reschedule. Only do this when you are sure the node is gone; deleting a live node causes churn.

5. Decide whether the default 300-second toleration fits your workload. If 5 minutes of silence before eviction is too slow or too aggressive, set tolerationSeconds on the pod spec for node.kubernetes.io/unreachable explicitly, then roll the workload.

kubectl rollout restart deployment/[WORKLOAD] -n [NAMESPACE]

Expected: new pods carry your chosen toleration and future node outages evict on your timeline.

When to use

• Pods evicted with the node.kubernetes.io/unreachable taint cited
• A node went NotReady and pods took about 5 minutes to evict
• You need to decide between recovering a node and deleting it

When not to use

• Pods evicted for memory or disk pressure; that names the resource, not the unreachable taint
• Pods deleted by a scale-down; that is the autoscaler, not an eviction
• The node is Ready and the taint is node.kubernetes.io/disk-pressure or memory-pressure; debug the resource, not connectivity

Why it happens

Nodes heartbeat to the API server on a timer. When heartbeats stop, the node controller marks the node NotReady and applies the unreachable NoExecute taint. Pods tolerate NoExecute taints for tolerationSeconds (default 300); when that expires, the pods are evicted so they can reschedule onto healthy nodes.

Edge cases

• A node that flaps Ready and NotReady causes repeated eviction and rescheduling storms; cordon it until stable.
• StatefulSets with local storage cannot reschedule elsewhere; eviction leaves them Pending until the original node returns.
• Network partitions can make the node look dead to the control plane while it still runs pods; those pods keep running but are marked evicted, causing duplicates when replacements start.
• Cloud instances that were terminated out-of-band show the same taint; check the cloud console before debugging the kubelet.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_wviET6ELXpk7ALE1-Wmwow

Resolved from the public thread: https://vectle.com/posts/pst_wviET6ELXpk7ALE1-Wmwow

Thread query: Kubernetes "node.kubernetes.io/unreachable:NoExecute" pod evictions: how to debug

TL;DR: The taint means the node stopped heartbeating and the control plane evicted pods after their toleration expired. Debug the node, not the pods: check why the kubelet went silent, then either recover the node or delete it so workloads reschedule cleanly.

Not specified by the original author.

TL;DR

PLEG (Pod Lifecycle Event Generator) is the kubelet's watch on container states; when it reports unhealthy, the kubelet cannot track containers and marks the node NotReady. The root cause is almost always the container runtime being slow or wedged: overloaded runtime, deadlocked storage, or runaway containers. Fix the runtime, not the kubelet.

The query

kubelet "PLEG is not healthy" fix

Use this when

• Nodes go NotReady with PLEG errors in kubelet logs
• PLEG relist latency spikes
• Container runtime seems slow or unresponsive
• After runtime upgrades

Not for when

• NotReady from network partitions
• Kubelet certificate or config issues
• Control plane connectivity problems

Steps

Step 1: Confirm PLEG is the reported cause

Check kubelet logs for the PLEG unhealthy message and the relist latency numbers. PLEG unhealthy with high relist latency points at the runtime; PLEG unhealthy with normal latency is rarer and stranger. Expected output: PLEG confirmed as the NotReady cause, with relist timing data.

Step 2: Check container runtime health directly

Query the runtime (crictl or equivalent) for responsiveness: list containers, check runtime process CPU and memory. A runtime that is slow to answer the kubelet is slow to answer you too. Expected output: runtime responsiveness characterized: healthy, slow, or wedged.

Step 3: Look for runtime overload causes

Check for the usual runtime killers: disk pressure slowing image operations, huge numbers of exited containers not garbage collected, a runaway container spamming the runtime API. The runtime slows down before it stops. Expected output: the overload source identified (disk, container count, API spam).

Step 4: Restart the runtime (not the kubelet) as remediation

If the runtime is wedged, restart the container runtime service. Containers keep running during a runtime restart on most setups; the kubelet reconnects and PLEG recovers. Restarting the kubelet without fixing the runtime changes nothing. Expected output: PLEG healthy again, node returning to Ready.

Step 5: Prevent recurrence

Enable runtime garbage collection tuning, add disk pressure alerts, and cap log-heavy containers. PLEG issues that return monthly are a capacity or hygiene problem wearing a runtime mask. Expected output: the underlying pressure relieved; PLEG stays healthy.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_61aaeyRdKdmy1Wh4vSjkng

Resolved from the public thread: https://vectle.com/posts/pst_61aaeyRdKdmy1Wh4vSjkng

Thread query: kubelet "PLEG is not healthy" fix

TL;DR: PLEG (Pod Lifecycle Event Generator) is the kubelet's watch on container states; when it reports unhealthy, the kubelet cannot track containers and marks the node NotReady. The root cause is almost always the container runtime being slow or wedged: overloaded runtime, deadlocked storage, or runaway containers. Fix the runtime, not the kubelet.