Merge branch 'release' into trm/azure-dashboards

alert-policies/hardware-sentry/battery-charge.yml

@@ -0,0 +1,32 @@
+name: Battery Charge
+
+description: |+
+  This alert is triggered when a host battery charge falls too low.
+
+type: STATIC
+nrql:
+  query: "FROM Metric SELECT latest(hw.battery.charge) FACET site, host.name"
+
+# Function used to aggregate the NRQL query value(s) for comparison to the terms.threshold (Default: SINGLE_VALUE)
+valueFunction: SINGLE_VALUE
+
+# List of Critical and Warning thresholds for the condition
+terms:
+  - priority: CRITICAL
+    # Operator used to compare against the threshold.
+    operator: BELOW
+    # Value that triggers a violation
+    threshold: .3
+    # Time in seconds; 120 - 3600
+    thresholdDuration: 300
+    # How many data points must be in violation for the duration
+    thresholdOccurrences: ALL
+  - priority: WARNING
+    # Operator used to compare against the threshold.
+    operator: BELOW
+    # Value that triggers a violation
+    threshold: .5
+    # Time in seconds; 120 - 3600
+    thresholdDuration: 300
+    # How many data points must be in violation for the duration
+    thresholdOccurrences: ALL

alert-policies/hardware-sentry/connector-status.yml

@@ -0,0 +1,22 @@
+name: Connector Status
+description: |+
+  This alert is triggered when a connector becomes degraded.
+
+type: STATIC
+nrql:
+  query: "FROM Metric SELECT latest(hardware_sentry.connector.status) WHERE state = 'degraded' FACET site, host.name"
+
+# Function used to aggregate the NRQL query value(s) for comparison to the terms.threshold (Default: SINGLE_VALUE)
+valueFunction: SINGLE_VALUE
+
+# List of Critical and Warning thresholds for the condition
+terms:
+  - priority: CRITICAL
+    # Operator used to compare against the threshold.
+    operator: EQUALS
+    # Value that triggers a violation
+    threshold: 1
+    # Time in seconds; 120 - 3600
+    thresholdDuration: 300
+    # How many data points must be in violation for the duration
+    thresholdOccurrences: ALL

alert-policies/hardware-sentry/device-status(crit).yml

@@ -0,0 +1,23 @@
+name: Device Status (critical)
+
+description: |+
+  This alert is triggered when a device is failing.
+
+type: STATIC
+nrql:
+  query: "FROM Metric SELECT latest(hw.status) WHERE  state IN ('failed') OR battery.state = 'failed'  FACET site, host.name"
+
+# Function used to aggregate the NRQL query value(s) for comparison to the terms.threshold (Default: SINGLE_VALUE)
+valueFunction: SINGLE_VALUE
+
+# List of Critical and Warning thresholds for the condition
+terms:
+  - priority: CRITICAL
+    # Operator used to compare against the threshold.
+    operator: EQUALS
+    # Value that triggers a violation
+    threshold: 1
+    # Time in seconds; 120 - 3600
+    thresholdDuration: 300
+    # How many data points must be in violation for the duration
+    thresholdOccurrences: ALL

alert-policies/hardware-sentry/device-status(warn).yml

@@ -0,0 +1,23 @@
+name: Device Status (warning)
+
+description: |+
+  This alert is triggered when a device is in a warning state.
+
+type: STATIC
+nrql:
+  query: "FROM Metric SELECT latest(hw.battery.charge) FACET site, host.name"
+
+# Function used to aggregate the NRQL query value(s) for comparison to the terms.threshold (Default: SINGLE_VALUE)
+valueFunction: SINGLE_VALUE
+
+# List of Critical and Warning thresholds for the condition
+terms:
+  - priority: WARNING
+    # Operator used to compare against the threshold.
+    operator: EQUALS
+    # Value that triggers a violation
+    threshold: 1
+    # Time in seconds; 120 - 3600
+    thresholdDuration: 300
+    # How many data points must be in violation for the duration
+    thresholdOccurrences: ALL

alert-policies/hardware-sentry/host-nonresponsive.yml

@@ -0,0 +1,22 @@
+name: Host not responding
+description: |+
+  This alert is triggered when a host is not responding.
+
+type: STATIC
+nrql:
+  query: "SELECT latest(hardware_sentry.host.up) FROM Metric FACET host.id"
+
+# Function used to aggregate the NRQL query value(s) for comparison to the terms.threshold (Default: SINGLE_VALUE)
+valueFunction: SINGLE_VALUE
+
+# List of Critical and Warning thresholds for the condition
+terms:
+  - priority: CRITICAL
+    # Operator used to compare against the threshold.
+    operator: EQUALS
+    # Value that triggers a violation
+    threshold: 0
+    # Time in seconds; 120 - 3600
+    thresholdDuration: 300
+    # How many data points must be in violation for the duration
+    thresholdOccurrences: ALL

alert-policies/hardware-sentry/missing-devices.yml

@@ -0,0 +1,22 @@
+name: Physical disk endurance capacity
+description: |+
+  This alert is triggered when a disk endurance utilization is very low.
+
+type: STATIC
+nrql:
+  query: "SELECT latest(hw.status) FROM Metric WHERE state = 'present' FACET host.name"
+
+# Function used to aggregate the NRQL query value(s) for comparison to the terms.threshold (Default: SINGLE_VALUE)
+valueFunction: SINGLE_VALUE
+
+# List of Critical and Warning thresholds for the condition
+terms:
+  - priority: CRITICAL
+    # Operator used to compare against the threshold.
+    operator: EQUALS
+    # Value that triggers a violation
+    threshold: 0
+    # Time in seconds; 120 - 3600
+    thresholdDuration: 300
+    # How many data points must be in violation for the duration
+    thresholdOccurrences: ALL

alert-policies/hardware-sentry/physical-disk-cap.yml

@@ -0,0 +1,31 @@
+name: Physical disk endurance capacity
+description: |+
+  This alert is triggered when a disk endurance utilization is very low.
+
+type: STATIC
+nrql:
+  query: "SELECT latest(hw.physical_disk.endurance_utilization) FROM Metric FACET host.id"
+
+# Function used to aggregate the NRQL query value(s) for comparison to the terms.threshold (Default: SINGLE_VALUE)
+valueFunction: SINGLE_VALUE
+
+# List of Critical and Warning thresholds for the condition
+terms:
+  - priority: CRITICAL
+    # Operator used to compare against the threshold.
+    operator: BELOW
+    # Value that triggers a violation
+    threshold: .02
+    # Time in seconds; 120 - 3600
+    thresholdDuration: 300
+    # How many data points must be in violation for the duration
+    thresholdOccurrences: ALL
+  - priority: WARNING
+    # Operator used to compare against the threshold.
+    operator: BELOW
+    # Value that triggers a violation
+    threshold: .05
+    # Time in seconds; 120 - 3600
+    thresholdDuration: 300
+    # How many data points must be in violation for the duration
+    thresholdOccurrences: ALL