SSH into nodes

SSH into nodes#

Sometimes, you need to directly SSH into a kubernetes node to troubleshoot an issue. This document describes how to do that on various cloud providers.

Kubernetes ephemeral containers#

Kubernetes offers a feature to use debug pods to get a node shell session.

This should work across cloud providers.

kubectl debug node/mynode -it --image=ubuntu

This creates a debug pod in the default namespace and drops you into a root shell with the node filesystem mounted on /host. When finished debugging delete the create debugger pod.


  1. Make sure you are authenticated with gcloud

  2. Set the project we are operating on, so gcloud knows where to look:

    gcloud config set project <name-of-project>


    You can find the name of the project under gcp.project in the cluster.yaml file for the cluster.

  3. Find the name of the node you want to login to, usually via kubectl get node. You can also find out the node a specific pod is on by kubectl get node -o wide.

  4. SSH into the node with gcloud compute ssh <node-name>. This will set you up with a user who has sudo permissions on the node, so you can poke around!


  1. Make sure you are logged in to the aws commandline tool, and authenticated as yourself to have access to AWS organization under which this cluster lives. You can validate that with aws sts get-caller-identity - the output should include your personal username, not that of the hub deployer!

  2. You also need the AWS Session Manager installed.

  3. Get the instance id of the node. Unlike with GCP, on AWS the instance id is not the same as the node-name reported by kubectl get node or kubectl get pod -o wide. The instance name is on the kubernetes node object as a label with name You can get the entire object’s definition with kubectl get node <node-name> -o yaml, and pick out the from there. This is of the form i-<some-string>.

  4. Get the region of the node. From the output you got in step 3, you can look at the label to get the region. For us, it’s often us-west-2 (as that is where a lot of scientific data is stored)

  5. You can now ssh with:

    aws ssm start-session --target <instance-id> --region <region>
  6. This will put you on the node with the sh shell, which is missing a lot of the features we expect from interactive shells today. You can get on bash with bash -l.

  7. You will be a user with full sudo access, so you can troubleshoot to your heart’s content.