feat: Export Trino metrics to Prometheus and enable autoscalling with KEDA

distributed-databases/trino/addons.tf

@@ -153,6 +153,17 @@ module "eks_blueprints_addons" {
     ],
   }
 
+  helm_releases = {
+    keda = {
+      chart            = "keda"
+      chart_version    = "2.16.0"
+      repository       = "https://kedacore.github.io/charts"
+      description      = "Keda helm Chart deployment"
+      namespace        = "keda"
+      create_namespace = true
+    }
+  }
+
   tags = local.tags
 }
 

distributed-databases/trino/examples/hive-setup.sh

@@ -10,7 +10,7 @@ echo "The name of your bucket is: ${BUCKET}"
 echo "Now copying the 2022 NY Taxi data into the S3 bucket..."
 
 ## Copy the 2022 NY Taxi data into the S3 bucket
-aws s3 cp "s3://nyc-tlc/trip data/" s3://$BUCKET/hive/  --exclude "*"  --include  "yellow_tripdata_2022*" --recursive
+aws s3 cp "s3://aws-data-analytics-workshops/shared_datasets/tripdata/" s3://$BUCKET/hive/ --recursive
 
 sleep 2
 echo "Now we create the Glue Database..."

distributed-databases/trino/helm-values/kube-prometheus-amp-enable.yaml

@@ -0,0 +1,53 @@
+prometheus:
+  serviceAccount:
+    create: true
+    name: ${amp_sa}
+    annotations:
+      eks.amazonaws.com/role-arn: ${amp_irsa}
+  prometheusSpec:
+    remoteWrite:
+      - url: ${amp_remotewrite_url}
+        sigv4:
+          region: ${region}
+        queue_config:
+          max_samples_per_send: 1000
+          max_shards: 200
+          capacity: 2500
+    retention: 5h
+    scrapeInterval: 30s
+    evaluationInterval: 30s
+    scrapeTimeout: 10s
+    storageSpec:
+      volumeClaimTemplate:
+        metadata:
+          name: data
+        spec:
+          storageClassName: gp2
+          accessModes:
+            - ReadWriteOnce
+          resources:
+            requests:
+              storage: 50Gi
+    serviceMonitorSelectorNilUsesHelmValues: false
+alertmanager:
+  enabled: false
+
+grafana:
+  enabled: true
+  defaultDashboardsEnabled: true
+# Adding AMP datasource to Grafana config
+  serviceAccount:
+    create: false
+    name: ${amp_sa}
+  grafana.ini:
+    auth:
+      sigv4_auth_enabled: true
+  additionalDataSources:
+    - name: AMP
+      editable: true
+      jsonData:
+        sigV4Auth: true
+        sigV4Region: ${region}
+      type: prometheus
+      isDefault: false
+      url: ${amp_url}

distributed-databases/trino/helm-values/kube-prometheus.yaml

@@ -15,6 +15,7 @@ prometheus:
           resources:
             requests:
               storage: 50Gi
+    serviceMonitorSelectorNilUsesHelmValues: false
 alertmanager:
   enabled: false
 

distributed-databases/trino/helm-values/trino.yaml

@@ -11,13 +11,17 @@
 # └── Total: 80.6GB < 89GB ✓
 ---
 image:
-  tag: "427"
+  repository: trinodb/trino
+  tag: 447
+  pullPolicy: IfNotPresent
 server:
   workers: 3
   exchangeManager:
+    name: filesystem
     baseDir: "s3://${exchange_bucket_id}"
   autoscaling:
-    enabled: true
+    enabled: false
+    minReplicas: 1
     maxReplicas: 20
     targetCPUUtilizationPercentage: 75
     targetMemoryUtilizationPercentage: 80
@@ -106,7 +110,7 @@
   - "query.remote-task.max-error-duration=1m"
   - "query.max-hash-partition-count=100"   # Updated from query.hash-partition-count
   - "spill-enabled=true"                   # Updated from experimental.spill-enabled
   - "spiller-spill-path=/tmp/spill"        # Chagne this to SSD mount for faster
   - "memory.heap-headroom-per-node=9.6GB"
   - "optimizer.join-reordering-strategy=AUTOMATIC"  # Updated from join-reordering-strategy
   - "query.max-history=100"
@@ -146,3 +150,50 @@
   name: ${sa}
 ingress:
   enabled: false
+jmx:
+  enabled: true
+  registryPort: 9080
+  serverPort: 9081
+  exporter:
+    # jmx.exporter.enabled -- Set to true to export JMX Metrics via HTTP for [Prometheus](https://github.com/prometheus/jmx_exporter) consumption
+    enabled: true
+    image: bitnami/jmx-exporter:latest
+    pullPolicy: Always
+    port: 5556
+    configProperties: |-
+      hostPort: localhost:{{- .Values.jmx.registryPort }}
+      startDelaySeconds: 0
+      ssl: false
+      lowercaseOutputName: false
+      lowercaseOutputLabelNames: false
+      whitelistObjectNames: ["trino.execution:name=QueryManager","trino.execution:name=SqlTaskManager","trino.execution.executor:name=TaskExecutor","trino.memory:name=ClusterMemoryManager","java.lang:type=Runtime","trino.memory:type=ClusterMemoryPool,name=general","java.lang:type=Memory","trino.memory:type=MemoryPool,name=general"]
+      autoExcludeObjectNameAttributes: true
+      excludeObjectNameAttributes:
+        "java.lang:type=OperatingSystem":
+          - "ObjectName"
+        "java.lang:type=Runtime":
+          - "ClassPath"
+          - "SystemProperties"
+      rules:
+      - pattern: ".*"
+    resources:
+      limits:
+        cpu: 200m
+        memory: 512Mi
+      requests:
+        cpu: 200m
+        memory: 512Mi
+
+serviceMonitor:
+  enabled: true
+  labels:
+    prometheus: kube-prometheus
+  interval: "30s"
+  coordinator:
+    enabled: true
+    labels:
+      prometheus: kube-prometheus
+  worker:
+    enabled: true
+    labels:
+      prometheus: kube-prometheus

distributed-databases/trino/trino-keda.yaml

@@ -0,0 +1,42 @@
+apiVersion: keda.sh/v1alpha1
+kind: ScaledObject
+metadata:
+  name: keda-scaler-trino-worker
+  namespace: ${trino_namespace}
+spec:
+  scaleTargetRef:
+    name: trino-worker
+  minReplicaCount: 3
+  maxReplicaCount: 15
+  pollingInterval: 30  # Seconds
+  cooldownPeriod: 600 # Seconds
+  fallback:
+    failureThreshold: 3
+    replicas: 6
+  advanced:
+    horizontalPodAutoscalerConfig:
+      behavior:
+        scaleDown:
+          stabilizationWindowSeconds: 600
+          policies:
+          - type: Pods
+            value: 1
+            periodSeconds: 300
+        scaleUp:
+          stabilizationWindowSeconds: 0
+          policies:
+          - type: Pods
+            value: 1
+            periodSeconds: 120
+  triggers:
+  - type: cpu
+    metricType: Utilization
+    metadata:
+      value: '80'  # Target CPU utilization percentage
+  - type: prometheus
+    metricType: Value
+    metadata:
+      serverAddress: http://kube-prometheus-stack-prometheus.kube-prometheus-stack.svc.cluster.local:9090
+      threshold: '1'
+      metricName: queued_queries
+      query: sum by (job) (avg_over_time(trino_execution_QueryManager_QueuedQueries{job="trino"}[1m]))

distributed-databases/trino/trino.tf

@@ -149,11 +149,15 @@ resource "aws_iam_policy" "trino_exchange_bucket_policy" {
 # Trino Helm Add-on
 #---------------------------------------
 module "trino_addon" {
+  depends_on = [
+    module.eks_blueprints_addons,
+  ]
+
   source  = "aws-ia/eks-blueprints-addon/aws"
   version = "~> 1.1.1" #ensure to update this to the latest/desired version
 
   chart            = "trino"
-  chart_version    = "0.33.0"
+  chart_version    = "0.34.0"
   repository       = "https://trinodb.github.io/charts"
   description      = "Trino Helm Chart deployment"
   namespace        = local.trino_namespace
@@ -190,3 +194,19 @@ module "trino_addon" {
     }
   }
 }
+
+
+#---------------------------------------------------------------
+# KEDA ScaleObject - Trino Prometheus
+#---------------------------------------------------------------
+resource "kubectl_manifest" "trino_keda" {
+
+  yaml_body = templatefile("${path.module}/trino-keda.yaml", {
+    trino_namespace = local.trino_namespace
+  })
+
+  depends_on = [
+    module.eks_blueprints_addons,
+    module.trino_addon
+  ]
+}