@hiiretail/gcp-infra-cli 0.77.1 → 0.78.0

package/generators/resources/monitoring/index.js

@@ -6,10 +6,12 @@ const BaseGenerator = require('../../../src/BaseGenerator');
 const { required } = require('../../../src/validators');
 const validate = require('./validate');
 const { handleSlos, handleAlerts, handleUptimeChecks } = require('./handle-yaml');
+const { getProjectId } = require('../pubsub/get-gcp-projects');
 
 const uptimeCheckTemplates = yaml.load(fs.readFileSync(`${__dirname}/templates/uptime-checks/uptime-checks.yaml`));
 const alertTemplates = yaml.load(fs.readFileSync(`${__dirname}/templates/alerts/alerts.yaml`));
 const sloTemplates = yaml.load(fs.readFileSync(`${__dirname}/templates/slos/slos.yaml`));
+const projectId = getProjectId('prod');
 
 module.exports = class extends BaseGenerator {
   async prompting() {
@@ -54,32 +56,11 @@ module.exports = class extends BaseGenerator {
         validate: required && validate.confluenceUrl,
       },
       {
-        when: (response) => response.alertResource === 'cloud_scheduler',
-        type: 'input',
-        name: 'jobId',
-        message: 'Please provide the "job id"',
-        validate: required,
-      },
-      {
-        when: (response) => response.alertResource === 'cloud_sql',
-        type: 'input',
-        name: 'databaseId',
-        message: 'Please provide the "database id"',
-        validate: required && validate.databaseId,
-      },
-      {
-        when: (response) => response.alertResource === 'memorystore',
-        type: 'input',
-        name: 'instanceId',
-        message: 'Please provide the "instance id"',
-        validate: required && validate.instanceID,
-      },
-      {
-        when: (response) => response.alertResource === 'pub_sub',
-        type: 'input',
-        name: 'subscriptionId',
-        message: 'Please provide the "subscription id"',
-        validate: required && validate.pubSubSubscription,
+        when: (response) => response.monitoringResource === 'alerts',
+        type: 'list',
+        name: 'projectId',
+        message: 'Please select clan project id',
+        choices: [`${projectId}`],
       },
       {
         when: (response) => response.monitoringResource === 'uptime-checks',
@@ -165,7 +146,6 @@ module.exports = class extends BaseGenerator {
       const yamlPath = `${resourceDir}/alerts.yaml`;
 
       copyTemplate('alerts', resourceDir, yamlPath);
-
       const oldYaml = yaml.load(fs.readFileSync(yamlPath, 'utf8')) || [];
       const newYaml = await handleAlerts(oldYaml, alertTemplates, this.answers);
 

package/generators/resources/monitoring/templates/alerts/alerts.yaml

@@ -8,6 +8,7 @@ cloud_run:
             resource.type="knative_revision"
             resource.labels.service_name="<%-serviceName%>"
             metric.type="knative.dev/serving/revision/request_latencies"
+            resource.labels.project_id="<%-projectId%>"
           threshold_value: 1000
           duration: 300s
           aggregations:
@@ -17,147 +18,210 @@ cloud_run:
       content: <% if (runbookLink) { %>[Runbook](<%-runbookLink%>)<%} else { %> <% } %>
 cloud_scheduler:
   failed_job:
-    display_name: "[P4] <%-systemName%> - Cloud Scheduler | <%-jobId%> - Job Failed"
+    display_name: "[P4] <%-systemName%> - Cloud Scheduler | Job Failed"
     conditions:
       - display_name: Cloud Scheduler Job - Log entries with SEVERITY=Error exceed threshold
         condition_threshold:
           filter: |
             resource.type="cloud_scheduler_job"
-            resource.labels.job_id="<%-jobId%>"
             metric.type="logging.googleapis.com/log_entry_count"
             metric.labels.severity="ERROR"
+            resource.labels.project_id="<%-projectId%>"
           threshold_value: 1
           aggregations:
             - alignment_period: 60s
               per_series_aligner: ALIGN_COUNT
+              group_by_fields: ["resource.label.job_id"]
     documentation:
       content: <% if (runbookLink) { %>[Runbook](<%-runbookLink%>)<%} else { %> <% } %>
 cloud_sql:
   cpu_over_65:
-    display_name: "[P3] <%-systemName%> - CloudSQL | <%-databaseId.substring(databaseId.lastIndexOf(':') + 1)%> - CPU over 65%"
+    display_name: "[P3] <%-systemName%> - CloudSQL | CPU over 65%"
     conditions:
       - display_name: Cloud SQL Database - CPU utilization above 65% over 5 min
         condition_threshold:
           filter: |
             resource.type="cloudsql_database"
-            resource.labels.database_id="<%-databaseId%>"
             metric.type="cloudsql.googleapis.com/database/cpu/utilization"
+            resource.labels.project_id="<%-projectId%>"
           threshold_value: 0.65
           duration: 300s
           aggregations:
             - alignment_period: 60s
               per_series_aligner: ALIGN_MAX
+              group_by_fields: ["resource.label.database_id"]
     documentation:
       content: <% if (runbookLink) { %>[Runbook](<%-runbookLink%>)<%} else { %> <% } %>
   cpu_over_85:
-    display_name: "[P3] <%-systemName%> - CloudSQL | <%-databaseId.substring(databaseId.lastIndexOf(':') + 1)%> - CPU over 85%"
+    display_name: "[P3] <%-systemName%> - CloudSQL | CPU over 85%"
     conditions:
       - display_name: "Cloud SQL Database - CPU-usage above 85% over 1 min"
         condition_threshold:
           filter: |
             resource.type="cloudsql_database"
-            resource.labels.database_id="<%-databaseId%>"
             metric.type="cloudsql.googleapis.com/database/cpu/utilization"
+            resource.labels.project_id="<%-projectId%>"
           threshold_value: 0.85
           duration: 60s
           aggregations:
             - alignment_period: 60s
               per_series_aligner: ALIGN_MAX
+              group_by_fields: ["resource.label.database_id"]
     documentation:
       content: <% if (runbookLink) { %>[Runbook](<%-runbookLink%>)<%} else { %> <% } %>
   cpu_over_90:
-    display_name: "[P3] <%-systemName%> - CloudSQL | <%-databaseId.substring(databaseId.lastIndexOf(':') + 1)%> - CPU over 90%"
+    display_name: "[P3] <%-systemName%> - CloudSQL | CPU over 90%"
     conditions:
       - display_name: Cloud SQL Database - CPU-usage above 90%
         condition_threshold:
           filter: |
             resource.type="cloudsql_database"
-            resource.labels.database_id="<%-databaseId%>"
             metric.type="cloudsql.googleapis.com/database/cpu/utilization"
+            resource.labels.project_id="<%-projectId%>"
           threshold_value: 0.9
           aggregations:
             - alignment_period: 60s
               per_series_aligner: ALIGN_MAX
+              group_by_fields: ["resource.label.database_id"]
     documentation:
       content: <% if (runbookLink) { %>[Runbook](<%-runbookLink%>)<%} else { %> <% } %>
   query_over_1s:
-    display_name: "[P4] <%-systemName%> - CloudSQL | <%-databaseId.substring(databaseId.lastIndexOf(':') + 1)%> - Query resolve time"
+    display_name: "[P4] <%-systemName%> - CloudSQL | Query resolve time"
     conditions:
       - display_name: Cloud SQL Instance Database - Per query execution times above 1000 ms
         condition_threshold:
           filter: |
             resource.type="cloudsql_instance_database"
-            resource.labels.resource_id="<%-databaseId%>"
             metric.type="cloudsql.googleapis.com/database/postgresql/insights/perquery/execution_time"
+            resource.labels.project_id="<%-projectId%>"
           threshold_value: 1000000
           aggregations:
             - alignment_period: 60s
               per_series_aligner: ALIGN_DELTA
+              group_by_fields: ["resource.label.resource_id"]
     documentation:
       content: <% if (runbookLink) { %>[Runbook](<%-runbookLink%>)<%} else { %> <% } %>
 memorystore:
   memory_over_50:
-    display_name: "[P4] <%-systemName%> - Memorystore | <%-instanceId.substring(instanceId.lastIndexOf('/') + 1)%> - Memory over 50%"
+    display_name: "[P4] <%-systemName%> - Memorystore | Memory over 50%"
     conditions:
       - display_name: Memorystore Redis Instance - Memory Usage above 50% over 5 min
         condition_threshold:
           filter: |
             resource.type="redis_instance"
-            resource.labels.instance_id="<%-instanceId%>"
             metric.type="redis.googleapis.com/stats/memory/usage_ratio"
+            resource.labels.project_id="<%-projectId%>"
           threshold_value: 0.5
           duration: 300s
           aggregations:
             - alignment_period: 60s
               per_series_aligner: ALIGN_MAX
+              group_by_fields: ["resource.label.instance_id"]
     documentation:
       content: <% if (runbookLink) { %>[Runbook](<%-runbookLink%>)<%} else { %> <% } %>
   memory_over_75:
-    display_name: "[P4] <%-systemName%> - Memorystore | <%-instanceId.substring(instanceId.lastIndexOf('/') + 1)%> - Memory over 75%"
+    display_name: "[P4] <%-systemName%> - Memorystore | Memory over 75%"
     conditions:
       - display_name: Memorystore Redis Instance - Memory Usage above 75% for 5min
         condition_threshold:
           filter: |
             resource.type="redis_instance"
-            resource.labels.instance_id="<%-instanceId%>"
             metric.type="redis.googleapis.com/stats/memory/usage_ratio"
+            resource.labels.project_id="<%-projectId%>"
           threshold_value: 0.75
           duration: 300s
           aggregations:
             - alignment_period: 60s
               per_series_aligner: ALIGN_MAX
+              group_by_fields: ["resource.label.instance_id"]
     documentation:
       content: <% if (runbookLink) { %>[Runbook](<%-runbookLink%>)<%} else { %> <% } %>
   memory_over_90:
-    display_name: "[P2] <%-systemName%> - Memorystore | <%-instanceId.substring(instanceId.lastIndexOf('/') + 1)%> - Memory over 90%"
+    display_name: "[P2] <%-systemName%> - Memorystore | Memory over 90%"
     conditions:
       - display_name: Memorystore Redis Instance - Memory Usage above 90%
         condition_threshold:
           filter: |
             resource.type="redis_instance"
-            resource.labels.instance_id="<%-instanceId%>"
             metric.type="redis.googleapis.com/stats/memory/usage_ratio"
+            resource.labels.project_id="<%-projectId%>"
           threshold_value: 0.90
           duration: 60s
           aggregations:
             - alignment_period: 60s
               per_series_aligner: ALIGN_MAX
+              group_by_fields: ["resource.label.instance_id"]
     documentation:
       content: <% if (runbookLink) { %>[Runbook](<%-runbookLink%>)<%} else { %> <% } %>
 pub_sub:
   unacknowledged_messages:
-    display_name: "[P4] <%-systemName%> - Pub/Sub | <%-subscriptionId.substring(subscriptionId.lastIndexOf('/') + 1)%> - Undelivered message(s)"
+    display_name: "[P3] <%-systemName%> - Pub/Sub | Undelivered message(s)"
     conditions:
       - display_name: Cloud Pub/Sub Subscription - Undelivered messages above 1 for 5 min
         condition_threshold:
           filter: |
             resource.type="pubsub_subscription"
-            resource.labels.subscription_id="<%-subscriptionId%>"
             metric.type="pubsub.googleapis.com/subscription/num_undelivered_messages"
+            resource.labels.project_id="<%-projectId%>"
           threshold_value: 1
           duration: 300s
           aggregations:
             - alignment_period: 60s
               per_series_aligner: ALIGN_MEAN
+              group_by_fields: ["resource.label.subscription_id"]
+    documentation:
+      content: <% if (runbookLink) { %>[Runbook](<%-runbookLink%>)<%} else { %> <% } %>
+  messages_in_dlq:
+    display_name: "[P3] <%-systemName%> - Pub/Sub | Message(s) in DLQ"
+    conditions:
+      - display_name: Cloud Pub/Sub Subscription - Number of undelivered message(s) forwarded to DLQ
+        condition_threshold:
+          filter: |
+            resource.type="pubsub_subscription"
+            metric.type="pubsub.googleapis.com/subscription/dead_letter_message_count"
+            resource.labels.project_id="<%-projectId%>"
+          threshold_value: 0
+          duration: 60s
+          aggregations:
+            - alignment_period: 60s
+              per_series_aligner: ALIGN_COUNT
+              group_by_fields: ["resource.label.subscription_id"]
+    documentation:
+      content: <% if (runbookLink) { %>[Runbook](<%-runbookLink%>)<%} else { %> <% } %>
+  latency:
+    display_name: "[P3] <%-systemName%> - Pub/Sub | Response latency distribution"
+    conditions:
+      - display_name: Cloud Pub/Sub Subscription - Latency above 3s
+        condition_threshold:
+          filter: |
+            resource.type="pubsub_subscription"
+            metric.type="pubsub.googleapis.com/subscription/push_request_latencies"
+            resource.labels.project_id="<%-projectId%>"
+          threshold_value: 3000000
+          duration: 120s
+          aggregations:
+            - alignment_period: 60s
+              per_series_aligner: ALIGN_DELTA
+              cross_series_reducer: REDUCE_PERCENTILE_95
+              group_by_fields: ["resource.label.subscription_id"]
+    documentation:
+      content: <% if (runbookLink) { %>[Runbook](<%-runbookLink%>)<%} else { %> <% } %>
+cloud_function:
+  failed_execution:
+    display_name: "[P2] <%-systemName%> - Cloud Function | Failed job execution"
+    conditions:
+      - display_name: Cloud Function - Execution error count
+        condition_threshold:
+          filter: |
+            resource.type="cloud_function"
+            metric.type="cloudfunctions.googleapis.com/function/execution_count"
+            metric.label.status!="ok"
+            resource.labels.project_id="<%-projectId%>"
+          threshold_value: 0
+          duration: 60s
+          aggregations:
+            - alignment_period: 60s
+              per_series_aligner: ALIGN_COUNT
+              group_by_fields: ["metric.label.status", "resource.label.function_name"]
     documentation:
       content: <% if (runbookLink) { %>[Runbook](<%-runbookLink%>)<%} else { %> <% } %>

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@hiiretail/gcp-infra-cli",
-  "version": "0.77.1",
+  "version": "0.78.0",
   "description": "Infrastructure as code generator for GCP.",
   "main": "src/cli.js",
   "bin": {