Upgrading etcd

Introduction

Etcd is the most crucial component of a Kubernetes cluster. It stores the cluster state.

This document will provide step by step guide on upgrading etcd in Lokomotive.

Steps

Repeat the following steps on all the controller node one node at a time.

Step 1: SSH into the controller node

Find the IP of the controller node by visiting the cloud provider dashboard and ssh into it.

ssh [email protected]<IP Address>

Step 2: Create necessary directories with correct permissions

Latest etcd (>= v3.4.10) necessitates the data directory permissions to be 0700, accordingly change the permissions. Verify the permissions are changed to rwx------. More information on the changes in upstream release notes can be found here .

NOTE: This step is needed only for the Lokomotive deployment done using lokoctl version < 0.4.0.

sudo chmod 0700 /var/lib/etcd/
sudo ls -ld /var/lib/etcd/

If the node reboots, we need the right settings in place so that systemd-tmpfile service does not alter the permissions of the data directory. To make the changes persistent run the following command:

echo "d    /var/lib/etcd 0700 etcd etcd - -" | sudo tee /etc/tmpfiles.d/etcd-wrapper.conf

Step 3: Upgrade etcd

Run the following commands:

NOTE: Before proceeding to other commands, set the etcd_version variable to the latest etcd version.

export etcd_version=<latest etcd version e.g. v3.4.10>

sudo sed -i "s,ETCD_IMAGE_TAG=.*,ETCD_IMAGE_TAG=${etcd_version}\"," \
        /etc/systemd/system/etcd-member.service.d/40-etcd-cluster.conf
sudo systemctl daemon-reload
sudo systemctl restart etcd-member

Step 4: Verify upgrade

Verify that the etcd service is in active (running) state:

sudo systemctl status --no-pager etcd-member

Run the following command to see logs of the process since the last restart:

sudo journalctl _SYSTEMD_INVOCATION_ID=$(sudo systemctl \
              show -p InvocationID --value etcd-member.service)

NOTE: Do not proceed with the upgrade of the rest of the cluster if you encounter any errors.

Once you see the following log line, you can discern that the etcd daemon has come up without errors:

etcdserver: starting server... [version: 3.4.10, cluster version: to_be_decided]

Once you see the following log line, you can discern that the etcd has rejoined the cluster without issues:

embed: serving client requests on 10.88.81.1:2379

Step 5: Verify using etcdctl

We can use etcdctl client to verify the state of etcd cluster.

# Find the endpoint of this node's etcd:
export endpoint=$(grep ETCD_ADVERTISE_CLIENT_URLS /etc/kubernetes/etcd.env | cut -d"=" -f2)
export flags="--cacert=/etc/ssl/etcd/etcd-client-ca.crt \
              --cert=/etc/ssl/etcd/etcd-client.crt \
              --key=/etc/ssl/etcd/etcd-client.key"
endpoints=$(sudo ETCDCTL_API=3 etcdctl member list $flags --endpoints=${endpoint} \
            --write-out=json | jq -r '.members | map(.clientURLs) | add | join(",")')

# Verify:
sudo ETCDCTL_API=3 etcdctl member list $flags --endpoints=${endpoint}
sudo ETCDCTL_API=3 etcdctl endpoint health $flags --endpoints=${endpoints}

The last command should report that nodes are healthy. If it indicates otherwise then try commands from Step 4 to see what’s wrong. If the nodes are healthy, it is safe to move forward with the next controller node.