Find one unused disk on each OCP node supporting storage for use in e. g. creating a LocalVolumeSet

find-unused-disk

#!/bin/bash

declare args=()
declare nodes=
declare -i verbose=0
declare label=cluster.ocs.openshift.io/openshift-storage

read -r -d '' script <<'EOF'
#!/bin/bash
declare -a sata_drives
declare -a nvme_drives
declare -i minsize=1500
declare -i maxsize=0
declare -i doit=0
declare -i verbose=0
declare devdir=
declare -i all_drives=0
declare -i status=0
declare -i found=0

help() {
    if [[ -z "$*" ]] ; then echo "Usage: $0 args"; fi
    cat <<EOD
    -a          Check for all matching drives
    -t          Print the by-path name of the drive
    -i          Print a by-id name of the drive
    -s          Check for matching SATA drives
    -S          Don't check for SATA drives
    -p          Check for matching NVMe drives
    -P          Don't check for NVMe drives
    -z          Zap the found drive
    -Z          Print the command, but don't zap the drive
    -n          Only print the name of the found drive
    -v          Print verbose debugging messages
    -m minsize  Only look at drives at least as big as minsize GB
    -M maxsize  Only look at drives no bigger than maxsize GB
EOD
exit 1
}

if [[ -n "${___HELP___}" ]] ; then
    help 1
fi

while getopts 'sSpPvzZnm:M:tiah' opt ; do
    case "$opt" in
        s) sata_drives=(/dev/sd?)     ;;
        S) sata_drives=()             ;;
        p) nvme_drives=(/dev/nvme?n1) ;;
        P) nvme_drives=()             ;;
        m) minsize=$OPTARG            ;;
        M) maxsize=$OPTARG            ;;
        z) doit=2                     ;;
        Z) doit=1                     ;;
        n) doit=0                     ;;
        t) devdir=by-path             ;;
        i) devdir=by-id               ;;
        a) all_drives=1               ;;
        v) verbose=1                  ;;
        h) help                       ;;
        *)
          echo "Unknown option $opt"
          help
          ;;
    esac
done
shift $((OPTIND-1))

docmd() {
    echo "$*" 1>&2
    if ((doit > 1)) ; then
        #"$@"
        return $!
    fi
}

print_drive() {
    local dev=$1
    if [[ -n "${devdir:-}" ]] ; then
        local path
        for path in /dev/disk/"$devdir"/* ; do
            if [[ $(readlink -f "$path") = "$dev" ]] ; then
                echo "$path"
                break
            fi
        done
    else
        echo "$dev"
    fi
}

doit() {
    local dev="${1:-}"
    if [[ ! -b "$dev" ]] ; then
        echo "$dev: not block device"
        exit 1
    fi
    case "$((doit))" in
        0) print_drive "$f" ;;
        *) docmd podman run -authfile /var/lib/kubelet --rm --privileged --device "$f" --entrypoint ceph-bluestore-tool quay.io/ceph/ceph:v19 zap-device --dev "$f" --yes-i-really-really-mean-it
    esac
}

vecho() {
    if ((verbose)) ; then echo "$*" 1>&2 ; fi
}

vecho "> Starting search on $(hostname)"
for f in "${sata_drives[@]}" "${nvme_drives[@]}" ; do
    vecho ">>  Looking at $f"
    vecho ">>>   Checking for mounted drive"
    mountpoints="$(lsblk -J "$f" | jq -r '.blockdevices[0]?.mountpoints[]? | select(. != null)')"
    if [[ -n "$mountpoints" ]] ; then
        # shellcheck disable=SC2086
        vecho ">>>>    SKIPPING: $f already mounted: " $mountpoints
        continue
    fi
    vecho ">>>   Checking for mounted partitions"
    cmountpoints="$(lsblk -J "$f" | jq -r '[.blockdevices[0]?.children[]?.mountpoints[]?] | flatten[] | select(. != null)')"
    if [[ -n "$cmountpoints" ]] ; then
        # shellcheck disable=SC2086
        vecho ">>>>    SKIPPING: $f has mounted partitions: " $cmountpoints
        continue
    fi
    vecho ">>>   Checking for unmounted partitions"
    children="$(lsblk -J "$f" | jq -r '.blockdevices[0]?.children[]?.name | select (. != null)')"
    if [[ -n "$children" ]] ; then
        # shellcheck disable=SC2086
        vecho ">>>>    SKIPPING: $f has unmounted partitions: " $children
        continue
    fi
    # If we immediately continued after finding a used disk
    # the find would get a broken pipe, which is ugly
    # albeit harmless
    failed=0
    if [[ -d /mnt/local-storage ]] ; then
        while read -r l ; do
            vecho ">>>   Looking at local storage entry $l ($(readlink -f "$l"))"
            if [[ $(readlink -f "$l") = "$f" ]] ; then
                vecho ">>>>    SKIPPING: $f is target of $l ($(readlink -f "$l"))"
                failed=1
            fi
        done <<< "$(sudo find /mnt/local-storage -type l -print)"
        if ((failed)) ; then continue; fi
    fi
    size="$(lsblk -J "$f" | jq -r '.blockdevices[0].size')"
    if [[ $size =~ ([0-9]+(\.[0-9]+)?)([KMGTP])? ]] ; then
        dsize=${BASH_REMATCH[1]}
        suffix=${BASH_REMATCH[3]}
        case "$suffix" in
            K) scale=1000 ;;
            M) scale=$((1000 * 1000)) ;;
            G) scale=$((1000 * 1000 * 1000)) ;;
            T) scale=$((1000 * 1000 * 1000 * 1000)) ;;
            P) scale=$((1000 * 1000 * 1000 * 1000 * 1000)) ;;
            *) continue ;;
        esac
        bytes=$(python3 <<< "print('%d' % ($dsize * $scale))")
        vecho ">>>   Checking size"
        if ((bytes < minsize * 1000 * 1000 * 1000 || (maxsize > 0 && bytes > maxsize * 1000 * 1000 * 1000))) ; then
            vecho ">>>>    SKIPPING: $dsize$suffix does not match $minsize < $bytes <= $maxsize"
            continue
        fi
        vecho ">>  $f passes all checks"
        if ((all_drives)) ; then
            doit "$f"
            case "$!" in
                0) found=$((found+1)) ;;
                *) status=1           ;;
            esac
        else
            doit "$f"
            exit $!
        fi
    fi
done
exit "$status"
EOF


help() {
    cat <<EOD
Usage: $0 <args> <nodes>
    Nodes default to all nodes labeled 'cluster.ocs.openshift.io/openshift-storage'
$(___HELP___=1; eval "$script")
    -l label    Use only nodes with the specified label.
EOD
    exit
}

while getopts 'm:M:l:abcdefghijknopqrstuvwxyzABCDEFGHIJKLNOPQRSTUVWXYZ' opt ; do
    case "$opt" in
        m|M)   args+=("-$opt" "$OPTARG") ;;
        l)     label=$OPTARG             ;;
        v)     verbose=1; args+=("-$opt");;
        h|H)   help                      ;;
        *)     args+=("-$opt")           ;;
    esac
done

doit() {
    if ((verbose)) ; then echo "$*" 1>&2; fi
    "$@"
}

shift $((OPTIND-1))
if [[ -n "$*" ]] ; then
    nodes=("$@")
else
    readarray -t nodes <<< "$(doit oc get node --no-headers -l "$label" | awk '{print $1}')"
fi

for node in "${nodes[@]}" ; do
    # shellcheck disable=SC2086
    # shellcheck disable=SC2029
    if ((verbose)) ; then echo 1>&2; fi
    disk=$(doit ssh -o StrictHostKeyChecking=no "core@$node" "cat > /tmp/diskfinder && chmod +x /tmp/diskfinder && /tmp/diskfinder ${args[*]}" <<< "$script")
    if [[ -n "$disk" ]] ; then
        printf "%s %s\n" "$node" "$disk"
    else
        echo "Can't find usable disk on $node" 1>&2
    fi
done