path: root/tools/helm-healer.sh

#!/bin/bash

PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin

#
# globals and defaults
#

NAMESPACE=
OVERRIDES=
HELM_CHART_RELEASE_NAME=
HELM_DELETE_ALL=
HELM_SKIP_DEPLOY=
VOLUME_STORAGE=
HELM_TIMEOUT=3600
RELEASE_PREFIX=onap

#
# control variables
#

CMD=$(basename "$0")
COLOR_ON_RED='\033[0;31;1m'
COLOR_ON_GREEN='\033[0;32;1m'
COLOR_OFF='\033[0m'


#
# functions
#

help()
{
cat <<EOF
${CMD} - simple tool for fixing onap helm deployment

DESCRIPTION
    This script does nothing smart or special it just tries to
    redeploy onap component. It can fix only problems related to
    race conditions or timeouts. Nothing else. It will not fix
    broken ONAP - there is no such ambition - that effort should
    be directed in the upstream.

USAGE
    ${CMD} -h|--help
        This help

    ${CMD} -n|--namespace <namespace>
           (-f|--file <override>)...
           (-s|--storage <directory>)|--no-storage-deletion
           [-p|--release-prefix <release prefix>]
           [-t|--timeout <secs>]
           [(-c|--component <component release name>)...|
            (-D|--delete-all)]
           [-C|--clean-only]

EXAMPLES

    Usage 1: (simple heuristics - redeploy failed components):
        ${CMD} -n onap -f /some/override1.yml -s /dockerdata-nfs

    Usage 2: (redeploy ONLY explicitly listed components):
        ${CMD} -n onap -f /some/override1.yml -s /dockerdata-nfs \\
               -c onap-aaf -c onap-sdc -c onap-portal

    Usage 3: (delete EVERYTHING and redeploy):
        ${CMD} -n onap -f /some/override1.yml -s /dockerdata-nfs --delete-all

    Usage 4: (delete EVERYTHING and DO NOT redeploy - clean env.)
        ${CMD} -n onap -s /dockerdata-nfs --delete-all --clean-only

NOTES

    Namespace argument (always) and at least one override file (if you don't
    use '--delete-all') are mandatory for this script to execute. Also you must
    provide path to the storage ('--storage') OR explicitly request to not
    delete file storage of the component ('--no-storage-deletion').

    The storage should be a directory where persistent volume resides. It will
    work only if the component created the persistent volume with the same
    filename as its release name. Otherwise no files are deleted. The exception
    is when '--delete-all' is used - in that case all content of the storage is
    deleted (because ONAP is not consistent with the volume directory names
    - e.g.: sdnc).

    '--file' can be used multiple of times and it is used for override files
    which are passed on to helm. The order is significant because if two
    override files modify one value the latest one is used. This option is
    ignored if '--clean-only' is used.

    CAUTION 1: filename of an override file cannot contain whitespace! This is
    actually helm/onap deploy plugin issue which does not handle such files. So
    I dropped the more complicated version of this script when there is no
    reason to support something on what will helm deploy choke anyway.

    '--prefix' option is helm release argument - it is actually prefix when you
    list the helm releases - helm is little confusing here.

    CAUTION 2: By default release prefix is 'onap' - if you deployed release
    'onap' and now run this script with different prefix then it will skip all
    'onap-*' components and will deploy a new release with new prefix - BEWARE
    TO USE PROPER RELEASE PREFIX!

    Timeout sets the waiting time for helm deploy per component.

    '--component' references to the release name of the chart which you want to
    redeploy excplicitly - otherwise 'ALL FAILED' components will be
    redeployed. You can target more than one component at once - just use the
    argument multiple times.

    Component option is mutually exclusive with the '--delete-all' which will
    delete all components - healthy or not. Actually it will delete the whole
    NAMESPACE and everything in it. Also to be sure it will cleanup all
    orphaned images and volumes on all kubernetes nodes.

    '--clean-only' can be used with any usage: heuristics, explicit component
    list or with '--delete-all'. It basically just skips the last step - the
    actual redeploy.
EOF
}

use_help()
{
    printf "Try help: ${CMD} --help\n"
}

msg()
{
    printf "${COLOR_ON_GREEN}INFO: $@ ${COLOR_OFF}\n"
}

error()
{
    printf "${COLOR_ON_RED}ERROR: $@ ${COLOR_OFF}\n"
}

on_exit()
{
    printf "$COLOR_OFF"
}

# remove all successfully completed jobs
clean_jobs()
{
    kubectl get jobs -n ${NAMESPACE} \
        --ignore-not-found=true \
        --no-headers=true | \
        while read -r _job _completion _duration _age ; do
            _done=$(echo ${_completion} | awk 'BEGIN {FS="/";} {print $1;}')
            _desired=$(echo ${_completion} | awk 'BEGIN {FS="/";} {print $2;}')
            if [ "$_desired" -eq "$_done" ] ; then
                delete_job "$_job"
            fi
        done
}

get_failed_labels()
{
    get_labels 'status.phase==Failed'
}

# arg: [optional: selector]
get_labels()
{
    if [ -n "$1" ] ; then
        _selector="--field-selector=${1}"
    else
        _selector=
    fi

    kubectl get pods -n ${NAMESPACE} \
        --show-labels=true \
        ${_selector} \
        --ignore-not-found=true \
        --no-headers=true | \
        while read -r _pod _ready _status _restart _age _labels ; do
            [ -z "$_labels" ] && break
            for _label in $(echo "$_labels" | tr ',' ' ') ; do
                case "$_label" in
                    release=*)
                        _label=$(echo "$_label" | sed 's/release=//')
                        echo "$_label"
                        ;;
                esac
            done
        done | sort -u
}

# arg: <release name>
helm_undeploy()
{
    msg "Undeploy helm release name: ${1}"
    helm undeploy ${1} --purge
    sleep 15s
}

# arg: <job name>
delete_job()
{
    kubectl delete job -n ${NAMESPACE} \
        --cascade=true \
        --now=true \
        --wait=true \
        ${1}

    # wait for job to be deleted
    _output=start
    while [ -n "$_output" ] && sleep 1 ; do
        _output=$(kubectl get pods -n ${NAMESPACE} \
            --ignore-not-found=true \
            --no-headers=true \
            --selector="job-name=${1}")
    done
}

#arg: <component>
get_resources_for_component()
{

helm status $1 | awk -f <(cat - <<-'EOD'
BEGIN {
  work="no"
  kind=""
  a["dummy"]=""
}

$1 ~ ":" {
  if ( $1 == "RESOURCES:" ) {
          work="yes"
} else {
  work="no"
}

}

$1 == "==>" {
  split($2, a, "[/(]")
  kind=a[2]
}

$1 != "NAME" && $1 != "==>" && work == "yes" && $1 !~ ":" && $1 != "" {
  printf "%s/%s\n", kind, $1
}

EOD
)
}

# arg: <resource>
delete_resource()
{
    local _resource="$1"
    local _kind="${_resource%/*}"
    local _name="${_resource#*/}"


    if kubectl get ${_resource} >/dev/null 2>&1; then
        msg "${_resource} has not been removed with helm undeploy, manual removal is required. Proceeding"
        kubectl delete ${_resource} -n ${NAMESPACE} \
            --cascade=true \
            --now=true \
            --wait=true \
            2>&1 | grep -iv 'not[[:space:]]*found'

        # wait for resource to be deleted
        _output=start
        while [ -n "$_output" ] && sleep 1 ; do
            _output=$(kubectl get ${_kind} ${_name} -n ${NAMESPACE} \
                --ignore-not-found=true \
                --no-headers=true )
        done
        msg "Done"
    fi
}

delete_namespace()
{
    msg "Delete the whole namespace: ${NAMESPACE}"
    kubectl delete namespace \
        --cascade=true \
        --now=true \
        --wait=true \
        "$NAMESPACE"

    # wait for namespace to be deleted
    _output=start
    while [ -n "$_output" ] && sleep 1 ; do
        _output=$(kubectl get all -n ${NAMESPACE} \
            --ignore-not-found=true \
            --no-headers=true)
    done
}

delete_persistent_volume()
{
    _persistent_volume=$1
     if kubectl get ${_persistent_volume} >/dev/null 2>&1; then
          msg "${_persistent_volume} has not been removed with helm undeploy, manual removal is required. Proceeding"
          #very often k8s hangs on Terminating state for pv due to  still active pvc. It is better to delete pvc directly
         _claim=$(kubectl get ${_persistent_volume} -o jsonpath='{ .spec.claimRef.name}')
         delete_resource PersistentVolumeClaim/${_claim}
     fi
}

# arg: [optional: directory]
delete_storage()
{
    _node=$(kubectl get nodes \
        --selector=node-role.kubernetes.io/worker \
        -o wide \
        --no-headers=true | \
        awk '{print $6}' | head -n 1)

    if [ -z "$_node" ] ; then
        error "Could not list kubernetes nodes - SKIPPING DELETION"
    else
        if [ -n "$1" ] ; then
            msg "Delete directory '${1}' on $_node"
            ssh $_node "rm -rf '${1}'"
        else
            msg "Delete directories '${VOLUME_STORAGE}/*' on $_node"
            ssh $_node "find '${VOLUME_STORAGE}' -maxdepth 1 -mindepth 1 -exec rm -rf '{}' \;"
        fi
    fi
}

docker_cleanup()
{
    _nodes=$(kubectl get nodes \
        --selector=node-role.kubernetes.io/worker \
        -o wide \
        --no-headers=true | \
        awk '{print $6}')

    if [ -z "$_nodes" ] ; then
        error "Could not list kubernetes nodes - SKIPPING docker cleanup"
        return
    fi

    for _node in $_nodes ; do
        msg "Docker cleanup on $_node"
        ssh $_node "docker system prune --force --all --volumes" >/dev/null &
    done

    msg "We are waiting now for docker cleanup to finish on all nodes..."
    wait
}

is_helm_serve_running()
{
    # healthy result: HTTP/1.1 200 OK
    _helm_serve_result=$(\
        curl --head --silent --connect-timeout 3 http://127.0.0.1:8879 | \
        head -n 1 | cut -d" " -f 3 | tr '[:upper:]' '[:lower:]' | tr -d '\r' )

    if [ "$_helm_serve_result" == ok ] ; then
        return 0
    else
        return 1
    fi
}

# arg: <release name>
undeploy_component()
{
    local _component=$1

    #Because Helm undeploy is not reliable: Gathering resources assigned to componen to track and remove orphans later
    _component_resources=($(get_resources_for_component ${_component}))

    declare -a _persistent_volumes
    declare -a _standard
    declare -a _unknown_kinds

    for resource in ${_component_resources[@]}; do
        case $resource in
            CronJob/* | Job/* | Secret/* | ConfigMap/* | Pod/* | Service/* | Deployment/* | StatefulSet/*)
                _standard+=(${resource});;
            #Ignoring PVC, they will be handled along with PV as 'helm' status does not return them for some components
            PersistentVolumeClaim/*)
                ;;
            PersistentVolume/*)
                _persistent_volumes+=(${resource});;
            *)
                _unknown_kinds+=(${resource})
        esac
    done


    #Gathering physical location of directories for persistent volumes to delete them after undeploy
    declare -a _physical_locations
    for volume in ${_persistent_volumes[@]}; do
        _physical_locations+=($(kubectl get ${volume} -o jsonpath='{ .spec.hostPath.path}' ))
    done

    helm_undeploy ${_component}

    #Manual items removal
    for resource in ${_standard[@]}; do
        delete_resource ${resource}
    done

    for volume in ${_persistent_volumes[@]}; do
        delete_persistent_volume ${volume}
    done

    for subdir in ${_physical_locations[@]}; do
        delete_storage ${subdir}
    done

    if [ "${#_unknown_kinds[@]}" -ne 0 ] ; then
        for resource in ${_unknown_kinds[@]}; do
            error "Untracked resource kind present: ${resource}, attempting to delete it..."
            delete_resource ${resource}
        done
        return
    fi
}

# arg: <release name>
deploy_component()
{
    # TODO: until I can verify that this does the same for this component as helm deploy
    #msg "Redeployment of the component ${1}..."
    #helm install "local/${_chart}" --name ${1} --namespace ${NAMESPACE} --wait --timeout ${HELM_TIMEOUT}
    error "NOT IMPLEMENTED"
}


#
# arguments
#

state=nil
arg_namespace=
arg_overrides=
arg_timeout=
arg_storage=
arg_nostorage=
arg_components=
arg_prefix=
arg_deleteall=
arg_cleanonly=
while [ -n "$1" ] ; do
    case $state in
        nil)
            case "$1" in
                -h|--help)
                    help
                    exit 0
                    ;;
                -n|--namespace)
                    state=namespace
                    ;;
                -f|--file)
                    state=override
                    ;;
                -t|--timeout)
                    state=timeout
                    ;;
                -s|--storage)
                    state=storage
                    ;;
                --no-storage-deletion)
                    if [ -n "$arg_storage" ] ; then
                        error "Usage of storage argument together with no storage deletion option!"
                        use_help
                        exit 1
                    elif [ -z "$arg_nostorage" ] ; then
                        arg_nostorage=nostorage
                    else
                        error "Duplicit argument for no storage option! (IGNORING)"
                    fi
                    ;;
                -c|--component)
                    if [ -n "$arg_deleteall" ] ; then
                        error "'Delete all components' used already - argument mismatch"
                        use_help
                        exit 1
                    fi
                    state=component
                    ;;
                -D|--delete-all)
                    if [ -n "$arg_components" ] ; then
                        error "Explicit component(s) provided already - argument mismatch"
                        use_help
                        exit 1
                    elif [ -z "$arg_deleteall" ] ; then
                        arg_deleteall=deleteall
                    else
                        error "Duplicit argument for 'delete all' option! (IGNORING)"
                    fi
                    ;;
                -p|--prefix)
                    state=prefix
                    ;;
                -C|--clean-only)
                    if [ -z "$arg_cleanonly" ] ; then
                        arg_cleanonly=cleanonly
                    else
                        error "Duplicit argument for 'clean only' option! (IGNORING)"
                    fi
                    ;;
                *)
                    error "Unknown parameter: $1"
                    use_help
                    exit 1
                    ;;
            esac
            ;;
        namespace)
            if [ -z "$arg_namespace" ] ; then
                arg_namespace="$1"
                state=nil
            else
                error "Duplicit argument for namespace!"
                use_help
                exit 1
            fi
            ;;
        override)
            if ! [ -f "$1" ] ; then
                error "Wrong filename for override file: $1"
                use_help
                exit 1
            fi
            arg_overrides="${arg_overrides} -f $1"
            state=nil
            ;;
        component)
            arg_components="${arg_components} $1"
            state=nil
            ;;
        prefix)
            if [ -z "$arg_prefix" ] ; then
                arg_prefix="$1"
                state=nil
            else
                error "Duplicit argument for release prefix!"
                use_help
                exit 1
            fi
            ;;
        timeout)
            if [ -z "$arg_timeout" ] ; then
                if ! echo "$1" | grep -q '^[0-9]\+$' ; then
                    error "Timeout must be an integer: $1"
                    use_help
                    exit 1
                fi
                arg_timeout="$1"
                state=nil
            else
                error "Duplicit argument for timeout!"
                use_help
                exit 1
            fi
            ;;
        storage)
            if [ -n "$arg_nostorage" ] ; then
                error "Usage of storage argument together with no storage deletion option!"
                use_help
                exit 1
            elif [ -z "$arg_storage" ] ; then
                arg_storage="$1"
                state=nil
            else
                error "Duplicit argument for storage!"
                use_help
                exit 1
            fi
            ;;
    esac
    shift
done

# sanity checks

if [ -z "$arg_namespace" ] ; then
    error "Missing namespace"
    use_help
    exit 1
else
    NAMESPACE="$arg_namespace"
fi

if [ -z "$arg_overrides" ] && [ -z "$arg_cleanonly" ] ; then
    error "Missing override file(s) or use '--clean-only'"
    use_help
    exit 1
else
    OVERRIDES="$arg_overrides"
fi

if [ -n "$arg_prefix" ] ; then
    RELEASE_PREFIX="$arg_prefix"
fi

if [ -n "$arg_timeout" ] ; then
    HELM_TIMEOUT="$arg_timeout"
fi

if [ -n "$arg_storage" ] ; then
    VOLUME_STORAGE="$arg_storage"
elif [ -z "$arg_nostorage" ] ; then
    error "Missing storage argument! If it is intended then use '--no-storage-deletion' option"
    use_help
    exit 1
fi

if [ -n "$arg_components" ] ; then
    HELM_CHART_RELEASE_NAME="$arg_components"
fi

if [ -n "$arg_deleteall" ] ; then
    HELM_DELETE_ALL=yes
fi

if [ -n "$arg_cleanonly" ] ; then
    HELM_SKIP_DEPLOY=yes
fi


#
# main
#

# set trap for this script cleanup
trap on_exit INT QUIT TERM EXIT

# another sanity checks
for tool in helm kubectl curl ; do
    if ! which "$tool" >/dev/null 2>&1 ; then
        error "Missing '${tool}' command"
        exit 1
    fi
done

if ! is_helm_serve_running ; then
    error "'helm serve' is not running (http://localhost:8879)"
    exit 1
fi

# if --delete-all is used then redeploy all components (the current namespace is deleted)
if [ -n "$HELM_DELETE_ALL" ] ; then
    # undeploy helm release (prefix)
    helm_undeploy "$RELEASE_PREFIX"

    # we will delete the whole namespace
    delete_namespace

    # we will cleanup docker on each node
    docker_cleanup

    # we will delete the content of storage (volumes)
    if [ -n "$VOLUME_STORAGE" ] ; then
        delete_storage
    fi
# delete and redeploy explicit or failed components...
else
    # if a helm chart release name was given then just redeploy said component and quit
    if [ -n "$HELM_CHART_RELEASE_NAME" ] ; then
        msg "Explicitly asked for component redeploy: ${HELM_CHART_RELEASE_NAME}"
        _COMPONENTS="$HELM_CHART_RELEASE_NAME"
    # simple heuristics: redeploy only failed components
    else
        msg "Delete successfully completed jobs..."
        clean_jobs

        msg "Find failed components..."
        _COMPONENTS=$(get_failed_labels)
    fi

    for _component in ${_COMPONENTS} ; do
        if echo "$_component" | grep -q "^${RELEASE_PREFIX}-" ; then
            msg "Redeploy component: ${_component}"
            undeploy_component ${_component}
        else
            error "Component release name '${_component}' does not match release prefix: ${RELEASE_PREFIX} (SKIP)"
        fi
    done
fi

if [ -z "$HELM_SKIP_DEPLOY" ] ; then
    # TODO: this is suboptimal - find a way how to deploy only the affected component...
    msg "Redeploy onap..."
    msg helm deploy ${RELEASE_PREFIX} local/onap --namespace ${NAMESPACE} ${OVERRIDES} --timeout ${HELM_TIMEOUT}
    helm deploy ${RELEASE_PREFIX} local/onap --namespace ${NAMESPACE} ${OVERRIDES} --timeout ${HELM_TIMEOUT}
else
    msg "Clean only option used: Skipping redeploy..."
fi

msg DONE

exit $?