Delete any orphaned CNS volumes found in vCenter without a corresponding Kubernetes PV

remove-orphaned-cns-volumes.sh

#!/bin/bash
# Delete any orphaned CNS volumes found in vCenter without a corresponding Kubernetes PV.
#
# Usage:
#   ./remove-orphaned-cns-volumes.sh 
#   KUBECONFIG=whatever.kubeconfig FORCE=true ./remove-orphaned-cns-volumes.sh 

set -euo pipefail

delete_volume() {
    echo "Deleting volume: $1"
    govc volume.rm "$1"
}

temp_dir=$(mktemp -d)
govc volume.ls | sort > "$temp_dir/vcenter.txt"
# Retreive the PV list after CNS volumes to avoid race condition: new volumes may appear in this list, but no new volumes should be accidentally deleted.
kubectl get pv -o jsonpath='{range .items[*]}{.spec.csi.volumeHandle}{"\t"}{.metadata.name}{"\n"}{end}' | sort > "$temp_dir/kubernetes.txt"
orphaned=$(comm -23 "$temp_dir/vcenter.txt" "$temp_dir/kubernetes.txt")
rm -rf "$temp_dir"
if [ -z "$orphaned" ]
then
    echo "No orphaned CNS volumes found"
    exit
fi

echo "Found potentially orphaned CNS volumes:"
echo "$orphaned"

if [ -z "${FORCE-}" ]
then
    read -p "Delete CNS volumes? [y/N]" -n 1 -r
    echo
    if [[ ! $REPLY =~ ^[Yy]$ ]]
    then
        exit 1
    fi
fi

# Some careful wordsplitting: split the loop on lines, and then extract the first tab delimeted argument
IFS=$'\n'
for volume in $orphaned
do
  IFS=$'\t'
  # shellcheck disable=SC2086
  delete_volume $volume
done

hi @bburky,
thank you very much for your script, it saved me a lot of time.
May I suggest you a modification taking care the possibility to have several K8s clusters hosted on vSphere.

> "$temp_dir/kubernetes-temp.txt"
for context in $(kubectl config view -o jsonpath='{.contexts[*].name}')
do
        echo checking context : $context
        kubectl get pv -o jsonpath='{range .items[*]}{.spec.csi.volumeHandle}{"\t"}{.metadata.name}{"\n"}{end}' --context $context  >> "$temp_dir/kubernetes-temp.txt"
done
sort $temp_dir/kubernetes-temp.txt > $temp_dir/kubernetes.txt

It could include Kubernetes cluster not on vSphere but the probability to have same PV name is zero.
Thanks again
Farid

I personally don't end up using the contexts feature too often because my clusters have the same admin username and contexts don't work then. For example, deploying multiple identical test clusters, then tearing them down. Instead I specify the KUBECONFIG environment variable to select the right file.

Anyone seeing this error?

> govc volume.ls
16583672-ef49-4fb7-8153-b1a06b6f0c85	pvc-c485e476-3a96-4736-a646-845bf3075aed
> govc volume.rm pvc-c485e476-3a96-4736-a646-845bf3075aed
govc: ServerFaultCode: Received SOAP response fault from [<cs p:00007f0010029530, TCP:localhost:443>]: retrieveVStorageObject
InvalidArgument: pvc-c485e476-3a96-4736-a646-845bf3075aed is incorrectly formatted.