check-routing-engine-temperature.rule

/*
 * Monitors routing-engine temperature and notifies when anomalies are found.
 * 
 * Three inputs control detection
 *
 *   1) "re-slot-no" is a regular expression that matches the routing engine
 *      that you would like to monitor.  By default it's '0-1', which matches
 *      both the routing engines. Use something like '0' to match only
 *      routing engine 0.
 *
 *   2) "re-temperature-low-threshold" is the threshold that causes the
 *      rule to report an anomaly. By default it's 45 degree Celcius of 
 *      RE temperature. This rule will set a dashboard color to green
 *      when temperature is below low threshold.
 * 
 *   3) "re-temperature-high-threshold" is the threshold that causes the
 *      rule to report an anomaly. By default it's 55 degree Celcius of RE 
 *      temperature.
 *
 * Requirement 14 (https://paragon-automation.atlassian.net/browse/REQ-14)
 *
 */
healthbot {
    topic hardware.chassis {
        rule check-routing-engine-temperature {
            /*
             * Monitors the routing-engine temperature and notifies via dashboard, red color 
             * when threshold is exceeded else green. 
			 *
             * Use routing-engine as key for rule.
             */		
            keys routing-engine;
            synopsis "Routing engine temperature";
            description "Collects system RE Temperature statistics periodically and notifies anomalies";
            /*
             * Sensor configuration to get data from network devices.
             */			 
            sensor components-oc {
                synopsis "routing-engine temperature KPI";
                description "open-config sensor to collect telemetry data from network device";
                open-config {
                    sensor-name /components/component;
                    frequency 60s;
                }
            }
            sensor components-oc-junos {
                open-config {
                    sensor-name /components/component;
                    frequency 60s;
                }
            }			
            /*
             * Fields defined using sensor path. Map the longer sensor names
             * to the shorter field names used in the rules.
             */				 
            field routing-engine-cpu-temperature {
                sensor components-oc {
                    where "/components/component/properties/property/@name == 'CPU Temperature'";
                    path /components/component/properties/property/state/value;
                }
                sensor components-oc-junos {
                    where "/components/component/properties/property/@name == 'temperature-cpu'";
                    path /components/component/properties/property/state/value;
                }				
                type integer;
                description "CPU temperature";
                plots re-cpu-temperature-plots {
                    unit celsius;
                    triggers routing-engine-cpu-temperature;
                    thresholds high-threshold {
                        field-name "$high-threshold";
                        severity critical;
                    }
                    thresholds low-threshold {
                        field-name "$low-threshold";
                        severity minor;
                    }
                    thresholds routing-engine-cpu-temperature-anomaly {
                        field-name "$routing-engine-cpu-temperature-anomaly";
                        severity minor;
                        type jaiml-threshold;
                    }					
                }				
            }
            field routing-engine-temperature {
                sensor components-oc {
                    where "/components/component/properties/property/@name == 'temperature'";
                    path /components/component/properties/property/state/value;
                }
                sensor components-oc-junos {
                    where "/components/component/properties/property/@name == 'temperature'";
                    path /components/component/properties/property/state/value;
                }				
                type integer;
                description "RE temperature";
                plots re-temperature-plots {
                    unit celsius;
                    triggers routing-engine-temperature;
                    thresholds high-threshold {
                        field-name "$high-threshold";
                        severity critical;
                    }
                    thresholds low-threshold {
                        field-name "$low-threshold";
                        severity minor;
                    }
                    thresholds routing-engine-temperature-anomaly {
                        field-name "$routing-engine-temperature-anomaly";
                        severity minor;
                        type jaiml-threshold;
                    }					
                }				
            }
            field routing-engine {
                sensor components-oc {
                    where "/components/component/@name =~ /^Routing Engine[{{re-slot-no}}]$|^RE[{{re-slot-no}}]$|^CHASSIS[{{re-slot-no}}]:RE[{{re-slot-no}}]$/";
                    path "/components/component/@name";
                }
                sensor components-oc-junos {
                    where "/components/component/@name =~ /^Routing Engine[{{re-slot-no}}]$|^RE[{{re-slot-no}}]$|^CHASSIS[{{re-slot-no}}]:RE[{{re-slot-no}}]$/";
                    path "/components/component/@name";
                }				
                description "RE name to monitor";
            }
            field high-threshold {
                constant {
                    value "{{re-temperature-high-threshold}}";
                }
                type integer;
                description "RE temperature high threshold";
            }
            field low-threshold {
                constant {
                    value "{{re-temperature-low-threshold}}";
                }
                type integer;
                description "RE temperature low threshold";
            }
            field routing-engine-cpu-temperature-anomaly {
                formula {
                    jaiml-anomaly-detection {
                        field-name "$routing-engine-cpu-temperature";
                    }
                }
                type integer;
                description "JAIML anomaly detection on routing engine CPU temperature field";
            }
            field routing-engine-temperature-anomaly {
                formula {
                    jaiml-anomaly-detection {
                        field-name "$routing-engine-temperature";
                    }
                }
                type integer;
                description "JAIML anomaly detection on routing engine temperature field";
            }
            /*
             * Anomaly detection logic.
             */
            trigger routing-engine-temperature {
                alert-type hardware.temperature.routing_engine.routing-engine-temperature;			
                synopsis "RE temperature KPI";
                description "Sets health based increase in RE temperature";
                frequency 1offset;
                /*
                 * Sets color to green when RE temperature
                 * re-temperature is below low threshold ($low-threshold).
                 */ 
                term is-re-temperature-normal {
                    when {
                        less-than "$routing-engine-temperature" "$low-threshold";
                        less-than-or-equal-to "$routing-engine-temperature-anomaly" 0;
                    }
                    then {
                        status {
                            color green;
                            message "$routing-engine temperature($routing-engine-temperature degree C) is normal";
                        }
                    }
                }
                /*
                 * Sets color to yellow when RE temperature re-temperature is below low threshold ($low-threshold)
                 * and the temperature value is anomalous.
                 */ 				
                term is-re-temperature-anomalous {
                    when {
                        less-than "$routing-engine-temperature" "$low-threshold";
                        equal-to "$routing-engine-temperature-anomaly" 1;
                    }
                    then {
                        status {
                            color yellow;
                            message "$routing-engine temperature is anomalous";
                        }
                    }
                }
                /*
                 * Sets color to green when RE temperature is below low threshold ($low-threshold)
                 * 
                 */				
                term is-temperature-less-than-low-threshold {
                    when {
                        less-than "$routing-engine-temperature" "$low-threshold";
                    }
                    then {
                        status {
                            color green;
                            message "$routing-engine temperature($routing-engine-temperature degree celsius) is normal";
                        }
                    }
                }				
                /*
                 * Sets color to yellow and notify anomaly when RE temperature 
                 * ($routing-engine-temperature) is below high threshold ($high-threshold)
                 */  
                term is-re-temperature-median {
                    when {
                        greater-than-or-equal-to "$routing-engine-temperature" "$low-threshold";					
                        less-than "$routing-engine-temperature" "$high-threshold";
                    }
                    then {
                        status {
                            color yellow;
                            message "$routing-engine temperature($routing-engine-temperature degree C) is-equal-to or exceeds high threshold";
                        }
                    }
                }
                /*
                 * Defaults color to red and notify anomaly.
                 */ 
                term re-temperature-abnormal {
                    when {
                        greater-than-or-equal-to "$routing-engine-temperature" "$high-threshold";
                    }
                    then {
                        status {
                            color red;
                            message "$routing-engine temperature($routing-engine-temperature degree C) is-equal-to or exceeds critical threshold";
                        }
                    }
                }
            }
            trigger routing-engine-cpu-temperature {
                alert-type hardware.temperature.routing_engine.cpu.routing-engine-cpu-temperature;			
                synopsis "RE CPU temperature KPI";
                description "Sets health based increase in RE CPU temperature";
                frequency 1offset;
                /*
                 * Sets color to green when RE CPU temperature
                 * re-cpu-temperature is below low threshold ($low-threshold).
                 */ 
                term is-temperature-normal {
                    when {
                        less-than "$routing-engine-cpu-temperature" "$low-threshold";
                        less-than-or-equal-to "$routing-engine-cpu-temperature-anomaly" 0;
                    }
                    then {
                        status {
                            color green;
                            message "$routing-engine CPU temperature($routing-engine-cpu-temperature degree C) is normal";
                        }
                    }
                }
                /*
                 * Sets color to yellow when RE temperature re-temperature is below low threshold ($low-threshold)
                 * and the temperature value is anomalous.
                 */ 
                term is-re-cpu-temperature-anomalous {
                    when {
                        less-than "$routing-engine-cpu-temperature" "$low-threshold";					
                        equal-to "$routing-engine-cpu-temperature-anomaly" 1;
                    }
                    then {
                        status {
                            color yellow;
                            message "$routing-engine CPU temperature is anomalous";
                        }
                    }
                }
                /*
                 * Sets color to green when RE CPU temperature is below low threshold ($low-threshold)
                 * 
                 */				
                term is-temperature-less-than-low-threshold {
                    when {
                        less-than "$routing-engine-cpu-temperature" "$low-threshold";
                    }
                    then {
                        status {
                            color green;
                            message "$routing-engine CPU temperature($routing-engine-cpu-temperature degree celsius) is normal";
                        }
                    }
                }				
                /*
                 * Sets color to yellow and notify anomaly when RE CPU
                 * temperature ($routing-engine-cpu-temperature) is below high threshold
                 * ($high-threshold).
                 */  
                term is-temperature-median {
                    when {
                       greater-than-or-equal-to "$routing-engine-cpu-temperature" "$low-threshold";					
                       less-than "$routing-engine-cpu-temperature" "$high-threshold";
                    }
                    then {
                        status {
                            color yellow;
                            message "$routing-engine CPU temperature($routing-engine-cpu-temperature degree C) is-equal-to or exceeds high threshold";
                        }
                    }
                }
                /*
                 * Defaults color to red and notify anomaly.
                 */ 
                term temperature-abnormal {
                    when {
                        greater-than-or-equal-to "$routing-engine-cpu-temperature" "$high-threshold";
                    }				
                    then {
                        status {
                            color red;
                            message "$routing-engine CPU temperature($routing-engine-cpu-temperature degree C) is-equal-to or exceeds critical threshold";
                        }
                    }
                }
            }			
            /*
             * Variables with default values. Default values can be changed
             * while deploying playbook instance.
             */			
            variable re-slot-no {
                value 0-1;
                description "Routing engine numbers to monitor, regular expression, e.g. '0'";
                type string;
            }
            variable re-temperature-high-threshold {
                value 55;
                description "RE temperature high threshold: Utilization increase between metrics, before we report anomaly";
                type int;
            }
            variable re-temperature-low-threshold {
                value 45;
                description "RE temperature low threshold: Utilization increase between metrics, before we report anomaly";
                type int;
            }		
            rule-properties {
                version 1;
                domain-name chassis;
                contributor juniper;
                category advanced;
                supported-healthbot-version 4.2.0;
                supported-devices {
                    sensors components-oc;				    
                    juniper {
                        operating-system junosEvolved {
                            products ACX {
                                sensors components-oc;							
                                platforms All {
                                    releases 22.3R1 {
                                        release-support min-supported-release;
                                    }
                                }
                            }
                            products PTX {
                                sensors components-oc-junos;
                                platforms PTX10008 {
                                    releases 22.2R3 {
                                        release-support min-supported-release;
                                    }
                                }								
                            }							
                        }							
                        operating-system junos {							
                            products MX {
                                sensors components-oc-junos;
                                platforms MX204 {
                                    releases 22.2R3 {
                                        release-support min-supported-release;
                                    }
                                }								
                                platforms MX304 {
                                    releases 22.2R3 {
                                        release-support min-supported-release;
                                    }
                                }
                                platforms MX480 {
                                    releases 23.2R1 {
                                        release-support min-supported-release;
                                    }
                                }
                                platforms MX960 {
                                    releases 23.2R1 {
                                        release-support min-supported-release;
                                    }
                                }
                                platforms MX10004 {
                                    releases 22.3R1 {
                                        release-support min-supported-release;
                                    }
                                }
                                platforms MX10008 {
                                    releases 22.3R1 {
                                        release-support min-supported-release;
                                    }
                                }
                                platforms MX10016 {
                                    releases 22.3R1 {
                                        release-support min-supported-release;
                                    }
                                }								
                            }
                            products JNP {
                                sensors components-oc-junos;
                                platforms JNP10004 {
                                    releases 22.3R1 {
                                        release-support min-supported-release;
                                    }
                                }
                                platforms JNP10008 {
                                    releases 22.3R1 {
                                        release-support min-supported-release;
                                    }
                                }
                                platforms JNP10016 {
                                    releases 22.3R1 {
                                        release-support min-supported-release;
                                    }
                                }								
                            }
                            products EX {
                                sensors components-oc-junos;
                                platforms EX4300-48MP {
                                    releases 22.4R2.8 {
                                        release-support min-supported-release;
                                    }
                                }
                            }
                            products QFX {
                                sensors components-oc-junos;
                                platforms QFX5110-48S {
                                    releases 22.3R1 {
                                        release-support min-supported-release;
                                    }
                                }
                            }
                        }
                    }
                }
            }
        }
    }	
}