Merge branch 'main' into k8s-node-nrql

Author: zstix

alert-policies/adobe-commerce-business-insights/5xxErrors.yml

@@ -0,0 +1,35 @@
+name: 5xx Server Errors
+
+description: |+
+  This alert is triggered if the customer faces 5xx server errors more than 5 times in 5 minutes.
+type: STATIC
+nrql:
+  query: "SELECT count(*) as '5xx Server Errors' from Transaction WHERE httpResponseCode LIKE '5%'"
+
+# 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: ABOVE
+    # Value that triggers a violation
+    threshold: 10
+    # 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: ABOVE
+    # 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
+
+# Duration after which a violation automatically closes
+# Time in seconds; 300 - 2592000 (Default: 86400 [1 day])
+violationTimeLimitSeconds: 86400

alert-policies/adobe-commerce-business-insights/cpuUsage.yml

@@ -0,0 +1,35 @@
+name: CPU Usage (%)
+
+description: |+
+  This alert is triggered if CPU usage exceeds 90% for 5 minutes.
+type: STATIC
+nrql:
+  query: "SELECT latest(host.cpuPercent) AS 'CPU Used %' FROM Metric"
+
+# 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: ABOVE
+    # Value that triggers a violation
+    threshold: 90
+    # 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: ABOVE
+    # Value that triggers a violation
+    threshold: 80
+    # Time in seconds; 120 - 3600
+    thresholdDuration: 300
+    # How many data points must be in violation for the duration
+    thresholdOccurrences: ALL
+
+# Duration after which a violation automatically closes
+# Time in seconds; 300 - 2592000 (Default: 86400 [1 day])
+violationTimeLimitSeconds: 86400

alert-policies/adobe-commerce-business-insights/downtime.yml

@@ -0,0 +1,35 @@
+name: Downtime (%)
+
+description: |+
+  This alert is triggered if Downtime is more than 1% for 2 minutes.
+type: STATIC
+nrql:
+  query: "SELECT percentage(count(result), where result = 'FAILED') as 'Downtime (%)' from SyntheticCheck"
+
+# 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: ABOVE
+    # Value that triggers a violation
+    threshold: 1
+    # Time in seconds; 120 - 3600
+    thresholdDuration: 120
+    # How many data points must be in violation for the duration
+    thresholdOccurrences: ALL
+  - priority: WARNING
+    # Operator used to compare against the threshold.
+    operator: ABOVE
+    # Value that triggers a violation
+    threshold: 0.5
+    # Time in seconds; 120 - 3600
+    thresholdDuration: 120
+    # How many data points must be in violation for the duration
+    thresholdOccurrences: ALL
+
+# Duration after which a violation automatically closes
+# Time in seconds; 300 - 2592000 (Default: 86400 [1 day])
+violationTimeLimitSeconds: 86400

alert-policies/adobe-commerce-business-insights/memoryUsage.yml

@@ -0,0 +1,35 @@
+name: Memory Usage (%)
+
+description: |+
+  This alert is triggered if Memory usage exceeds 90% for 5 minutes.
+type: STATIC
+nrql:
+  query: "SELECT latest(host.memoryUsedPercent) as 'Memory Used %' FROM Metric"
+
+# 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: ABOVE
+    # Value that triggers a violation
+    threshold: 90
+    # 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: ABOVE
+    # Value that triggers a violation
+    threshold: 80
+    # Time in seconds; 120 - 3600
+    thresholdDuration: 300
+    # How many data points must be in violation for the duration
+    thresholdOccurrences: ALL
+
+# Duration after which a violation automatically closes
+# Time in seconds; 300 - 2592000 (Default: 86400 [1 day])
+violationTimeLimitSeconds: 86400

alert-policies/amazon-cloudsearch/HighIndexUtilization.yml

@@ -0,0 +1,27 @@
+name: High Index Utilization
+
+description: |+
+  This alert is triggered when the Index Utilization is above 90%.
+
+type: STATIC
+nrql:
+  query: "SELECT average(`aws.cloudsearch.IndexUtilization`) as 'Query' FROM Metric"
+
+# 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: ABOVE
+    # Value that triggers a violation
+    threshold: 90
+    # Time in seconds; 120 - 3600
+    thresholdDuration: 300
+    # How many data points must be in violation for the duration
+    thresholdOccurrences: ALL
+
+# Duration after which a violation automatically closes
+# Time in seconds; 300 - 2592000 (Default: 86400 [1 day])
+violationTimeLimitSeconds: 86400

alert-policies/azure-machine-learning/ModelDeployFailed.yml

@@ -0,0 +1,33 @@
+name: Model Deployment Failed
+
+description: |+
+  This alert is triggered if the number of Failure exceeds 20 within 10 minutes.
+type: STATIC
+nrql:
+  query: "FROM Metric SELECT sum(azure.machinelearningservices.workspaces.ModelDeployFailed) AS 'ModelDeployFailed'"
+
+# 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: ABOVE
+    # Value that triggers a violation
+    threshold: 20
+    # Time in seconds; 120 - 3600
+    thresholdDuration: 600
+    # How many data points must be in violation for the duration
+    thresholdOccurrences: ALL
+
+  # Adding a Warning threshold is optional
+  - priority: WARNING
+    operator: ABOVE
+    threshold: 10
+    thresholdDuration: 600
+    thresholdOccurrences: ALL
+
+# Duration after which a violation automatically closes
+# Time in seconds; 300 - 2592000 (Default: 86400 [1 day])
+violationTimeLimitSeconds: 86400

alert-policies/openstack-controller/ImageStatus.yml

@@ -0,0 +1,32 @@
+# Name of the alert
+name: Image Status
+
+# Description and details
+description: |+
+  This alert is triggered when the image status is inactive for 5 minutes.
+# Type of alert
+type: STATIC
+
+# NRQL query
+nrql:
+
+  query: "SELECT count(*)FROM OSImageSample where openstack.glance.image.status != 'active'"
+
+# 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: ABOVE
+    # 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
+
+# Duration after which a violation automatically closes
+# Time in seconds; 300 - 2592000 (Default: 86400 [1 day])
+violationTimeLimitSeconds: 86400

alert-policies/openstack-controller/MemoryUsage.yml

@@ -0,0 +1,40 @@
+# Name of the alert
+name: Memory Usage Percent
+
+# Description and details
+description: |+
+  This alert is triggered when the memory usage exceeds 90% for 5 minutes
+# Type of alert
+type: STATIC
+
+# NRQL query
+nrql:
+
+  query: "SELECT average(`memoryUsedBytes`/`memoryTotalBytes`*100) FROM SystemSample facet entityName"
+
+# 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: ABOVE
+    # Value that triggers a violation
+    threshold: 90
+    # 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: ABOVE
+    # Value that triggers a violation
+    threshold: 85
+    # Time in seconds; 120 - 3600, must be a multiple of 60 for Baseline conditions
+    thresholdDuration: 300
+    # How many data points must be in violation for the duration
+    thresholdOccurrences: ALL  
+# Duration after which a violation automatically closes
+# Time in seconds; 300 - 2592000 (Default: 86400 [1 day])
+violationTimeLimitSeconds: 86400

alert-policies/openstack-controller/ServerFreeMemoryLow.yml

@@ -0,0 +1,32 @@
+# Name of the alert
+name: Server Free Memory Low(%)
+
+# Description and details
+description: |+
+  This alert is triggered when server free memory is less than 10% for 5 minutes.
+# Type of alert
+type: STATIC
+
+# NRQL query
+nrql:
+
+  query: "SELECT average(`memoryFreeBytes`/`memoryTotalBytes`*100) FROM SystemSample facet entityName"
+
+# 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: 10
+    # Time in seconds; 120 - 3600
+    thresholdDuration: 300
+    # How many data points must be in violation for the duration
+    thresholdOccurrences: ALL
+
+# Duration after which a violation automatically closes
+# Time in seconds; 300 - 2592000 (Default: 86400 [1 day])
+violationTimeLimitSeconds: 86400

alert-policies/stripe/clientErrors.yml

@@ -0,0 +1,35 @@
+name: 4xx client errors
+
+description: |+
+  This alert is triggered if customer faces 4xx errors more than 5 times for 5 minutes.
+type: STATIC
+nrql:
+  query: "SELECT count(*) from Transaction WHERE httpResponseCode LIKE '4%'"
+
+# 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: ABOVE
+    # 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
+  - priority: WARNING
+    # Operator used to compare against the threshold.
+    operator: ABOVE
+    # 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
+
+# Duration after which a violation automatically closes
+# Time in seconds; 300 - 2592000 (Default: 86400 [1 day])
+violationTimeLimitSeconds: 86400

alert-policies/stripe/downtime.yml

@@ -0,0 +1,35 @@
+name: Downtime (%)
+
+description: |+
+  This alert is triggered if Downtime is more than 1% for 2 minutes.
+type: STATIC
+nrql:
+  query: "SELECT percentage(count(result), where result = 'FAILED') as 'Downtime (%)' from SyntheticCheck"
+
+# 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: ABOVE
+    # Value that triggers a violation
+    threshold: 1
+    # Time in seconds; 120 - 3600
+    thresholdDuration: 120
+    # How many data points must be in violation for the duration
+    thresholdOccurrences: ALL
+  - priority: WARNING
+    # Operator used to compare against the threshold.
+    operator: ABOVE
+    # Value that triggers a violation
+    threshold: 0.5
+    # Time in seconds; 120 - 3600
+    thresholdDuration: 120
+    # How many data points must be in violation for the duration
+    thresholdOccurrences: ALL
+
+# Duration after which a violation automatically closes
+# Time in seconds; 300 - 2592000 (Default: 86400 [1 day])
+violationTimeLimitSeconds: 86400