apps/tost-telegraf/templates/configmap-config.yaml

{{/*
# Copyright 2021-present Open Networking Foundation

# SPDX-License-Identifier: Apache-2.0
*/}}

apiVersion: v1
kind: ConfigMap
metadata:
  name: "tost-telegraf-config-script"
  labels:
    chart: "{{ .Chart.Name }}-{{ .Chart.Version }}"
    release: "{{ .Release.Name }}"
    app: tost-telegraf
data:
  sdfabric_telegraf.sh: |
    #!/bin/bash
    # Install kubectl and jq
    curl -sLO https://dl.k8s.io/release/v1.21.0/bin/linux/amd64/kubectl
    install -m 755 kubectl /usr/local/bin/kubectl
    apt update
    apt install -y jq

    # Utility for jq
    cat <<EOF > ~/.jq
    def count(s): reduce s as \$_ (0;.+1);
    EOF

    get-onos-server() {
        ONOS_SERVER={{ .Values.onos.server }}
        CONTROLLERS=( $(kubectl get -n tost --output json  pods | \
            jq '.items[] | select(.status.containerStatuses[].ready==true and .status.containerStatuses[].name=="onos-classic")' | \
            jq -r .status.podIP) )
        if (( ${#CONTROLLERS[@]} )); then
            CONTROLLERS=( $(shuf -e "${CONTROLLERS[@]}") )
            ONOS_SERVER=${CONTROLLERS[0]}
        fi
    }

    while IFS= read -r LINE; do
        # Topology matrix
        get-onos-server
        ACTIVE_LINKS=$(curl --fail -sSL --user {{ .Values.onos.username }}:{{ .Values.onos.password }} --noproxy $ONOS_SERVER:{{ .Values.onos.port }} -X GET -H 'Accept: application/json' \
        http://$ONOS_SERVER:{{ .Values.onos.port }}/onos/v1/links | \
        jq 'count(.links[]?.state | select(. == "ACTIVE"))')
        get-onos-server
        DEVICES=$(curl --fail -sSL --user {{ .Values.onos.username }}:{{ .Values.onos.password }} --noproxy $ONOS_SERVER:{{ .Values.onos.port }} -X GET -H 'Accept: application/json' \
            http://$ONOS_SERVER:{{ .Values.onos.port }}/onos/v1/devices | \
            jq 'count(.devices[]? | select(.available and .type=="SWITCH"))')
        get-onos-server
        ENABLE_DEVICE_PORTS=$(curl --fail -sSL --user {{ .Values.onos.username }}:{{ .Values.onos.password }} --noproxy $ONOS_SERVER:{{ .Values.onos.port }} -X GET -H 'Accept: application/json' \
            http://$ONOS_SERVER:{{ .Values.onos.port }}/onos/v1/devices/ports | \
            jq 'count(.ports[]?.isEnabled | select(.))')
        ACTIVE_LINKS=${ACTIVE_LINKS:-0}
        DEVICES=${DEVICES:-0}
        ENABLED_DEVICE_PORTS=${ENABLED_DEVICE_PORTS:-0}
        echo "onos_telegraf active_links=${ACTIVE_LINKS},enable_device_ports=${ENABLE_DEVICE_PORTS},devices=${DEVICES}"
        # Readiness for each ONOS instance and the config loader(overall readiness)
        kubectl get po -n {{ .Values.onos.namespace }} -l '{{ .Values.onos.onos_classic_label }}' -o json | \
            jq -r '.items[]? | "onos_telegraf,pod=" + (.metadata.name) + " ready=" + (count(select(.status.containerStatuses[].ready)) | tostring)'
        kubectl get po -n {{ .Values.onos.namespace }} -l '{{ .Values.onos.onos_config_loader_label }}' -o json | \
            jq -r '"onos_telegraf,pod=onos-config-loader ready=" + (count(select(.items[0].status.containerStatuses[].ready)) | tostring)'
    done

  sdfabric_health_indicator.sh: |
    #!/bin/bash
    {{ if .Values.health_indicator.enabled }}
    # Constant
    readonly HEALTH_UNKNOWN=0
    readonly HEALTH_UP=1
    readonly HEALTH_DEGRADED=2
    readonly HEALTH_DOWN=3
    readonly REASON_PKT_LOSS=1
    readonly REASON_RTT=2
    readonly REASON_ONOS_NOT_READY=3
    readonly REASON_ATOMIX_NOT_READY=4
    readonly REASON_LINK_DOWN=5
    readonly REASON_DEVICE_DOWN=6

    PACKET_LOSS_PERCENT=0
    AVG_RTT=0

    check_host() {
        PING_RESULT=$(ping -i 0.1 -W 1 -c 10 $1)
        PACKET_LOSS_PERCENT=$(echo $PING_RESULT | grep -P -o '\d+% packet loss' | awk -F '%' '{print $1}')
        AVG_RTT=$(echo $PING_RESULT | grep -P -o '(\d+\.\d+)/(\d+\.\d+)/(\d+\.\d+)/(\d+\.\d+)' | awk -F'/' '{print $2}')
        AVG_RTT=${AVG_RTT%.*} # default bash can only compare integer
        if [[ $PACKET_LOSS_PERCENT -ge {{ .Values.health_indicator.packet_loss_threshold }} ]]; then
            echo "sdfabric_telegraf health=$HEALTH_DOWN,reason=$REASON_PKT_LOSS"
            return 1
        elif [[ $AVG_RTT -gt {{ $.Values.health_indicator.rtt_threshold_ms }} ]]; then
            echo "sdfabric_telegraf health=$HEALTH_DEGRADED,reson=$REASON_RTT,rtt=$AVG_RTT,expected_rtt={{ $.Values.health_indicator.rtt_threshold_ms }}"
            return 1
        elif [[ $PACKET_LOSS_PERCENT -gt 0 ]]; then
            echo "sdfabric_telegraf health=$HEALTH_DEGRADED,reason=$REASON_PKT_LOSS,percent=$PACKET_LOSS_PERCENT"
            return 1
        fi
        return 0
    }

    # Wait until jq and kubectl are installed
    while ! (which jq && which kubectl) > /dev/null ; do
        sleep 1
    done

    MAX_NUM_LINKS=0

    while IFS= read -r LINE; do
        {{ range .Values.health_indicator.expected_hosts }}
        check_host {{ . }} || continue
        {{ end }}

        NUM_ATOMIX_NOT_READY=$(kubectl get po -n {{ .Values.onos.namespace }} -l '{{ .Values.onos.atomix_label }}' -o json | \
            jq -r 'count(select(.items[].status.containerStatuses[].ready == false))')
        if [[ $NUM_ATOMIX_NOT_READY -gt 0 ]]; then
            echo "sdfabric_telegraf health=$HEALTH_DEGRADED,reason=$REASON_ATOMIX_NOT_READY,num_atomix_not_ready=$NUM_ATOMIX_NOT_READY"
            continue
        fi

        NUM_ONOS_NOT_READY=$(kubectl get po -n {{ .Values.onos.namespace }} -l '{{ .Values.onos.onos_classic_label }}' -o json | \
            jq -r 'count(select(.items[].status.containerStatuses[].ready == false))')
        if [[ $NUM_ONOS_NOT_READY -gt 0 ]]; then
            echo "sdfabric_telegraf health=$HEALTH_DEGRADED,reason=$REASON_ONOS_NOT_READY,num_onos_not_ready=$NUM_ONOS_NOT_READY"
            continue
        fi

        ACTIVE_LINKS=$(curl --fail -sSL --user {{ .Values.onos.username }}:{{ .Values.onos.password }} --noproxy {{ .Values.onos.server }}:{{ .Values.onos.port }} -X GET -H 'Accept: application/json' \
            http://{{ .Values.onos.server }}:{{ .Values.onos.port }}/onos/v1/links | \
            jq 'count(.links[]?.state | select(. == "ACTIVE"))')
        if [[ -z $ACTIVE_LINKS ]] || [[ ! $ACTIVE_LINKS -eq {{ .Values.health_indicator.expected_num_links }} ]]; then
            echo "sdfabric_telegraf health=$HEALTH_DEGRADED,reason=$REASON_LINK_DOWN,active_links=$ACTIVE_LINKS,expected_links={{ .Values.expected_num_links }}"
            continue
        fi

        UNAVAILABLE_DEVICES=$(curl --fail -sSL --user {{ .Values.onos.username }}:{{ .Values.onos.password }} --noproxy {{ .Values.onos.server }}:{{ .Values.onos.port }} -X GET -H 'Accept: application/json' \
            http://{{ .Values.onos.server }}:{{ .Values.onos.port }}/onos/v1/devices | \
            jq 'count(.devices[]? | select(.type=="SWITCH" and (.available == false)))')
        if [[ $UNAVAILABLE_DEVICES -gt 0 ]]; then
            echo "sdfabric_telegraf health=$HEALTH_DEGRADED,reason=$REASON_DEVICE_DOWN,num_device_down=$UNAVAILABLE_DEVICES"
            continue
        fi

        echo "sdfabric_telegraf health=$HEALTH_UP"
    done
    {{ else }}
    # Health indicator is off
    while IFS= read -r LINE; do
      echo "sdfabric_telegraf health=$HEALTH_UNKNOWN"
    done
    {{ end }}