From ff7cbd02951904abd40ae9587e011edea86d7f08 Mon Sep 17 00:00:00 2001
From: Frank Blechschmidt <contact@frank-blechschmidt.com>
Date: Sun, 18 Jul 2021 19:13:07 -0700
Subject: [PATCH] Add support for Prometheus metrics in the Spark Driver (#5)

---
 charts/bullet/Chart.yaml                      |  2 +-
 .../templates/spark-backend-deployment.yaml   |  2 +
 .../templates/spark-backend-podmonitor.yaml   | 48 +++++++++++++++++++
 charts/bullet/values.yaml                     |  9 ++++
 4 files changed, 60 insertions(+), 1 deletion(-)
 create mode 100644 charts/bullet/templates/spark-backend-podmonitor.yaml

diff --git a/charts/bullet/Chart.yaml b/charts/bullet/Chart.yaml
index fd96e50..acbde49 100644
--- a/charts/bullet/Chart.yaml
+++ b/charts/bullet/Chart.yaml
@@ -2,7 +2,7 @@ apiVersion: v2
 name: bullet
 description: A real-time query engine for very large data streams
 type: application
-version: 0.2.0
+version: 0.3.0
 appVersion: "1.0.4"
 home: https://github.com/SketchBench/bullet-k8s
 # icon:
diff --git a/charts/bullet/templates/spark-backend-deployment.yaml b/charts/bullet/templates/spark-backend-deployment.yaml
index 7ec7ea3..4350f25 100644
--- a/charts/bullet/templates/spark-backend-deployment.yaml
+++ b/charts/bullet/templates/spark-backend-deployment.yaml
@@ -55,6 +55,8 @@ spec:
             - {{ printf "spark.executor.cores=%g" .Values.sparkBackend.executorCores }}
             - --conf
             - {{ printf "spark.executor.memory=%s" .Values.sparkBackend.executorMemory }}
+            - --conf
+            - {{ printf "spark.ui.prometheus.enabled=%t" .Values.sparkBackend.metrics.enabled }}
             - --class
             - com.yahoo.bullet.spark.BulletSparkStreamingMain
             - --jars
diff --git a/charts/bullet/templates/spark-backend-podmonitor.yaml b/charts/bullet/templates/spark-backend-podmonitor.yaml
new file mode 100644
index 0000000..b082cfb
--- /dev/null
+++ b/charts/bullet/templates/spark-backend-podmonitor.yaml
@@ -0,0 +1,48 @@
+{{- if and .Values.sparkBackend.metrics.enabled .Values.sparkBackend.metrics.podMonitor.enabled }}
+apiVersion: monitoring.coreos.com/v1
+kind: PodMonitor
+metadata:
+  name: {{ include "common.names.fullname" . }}-spark-backend
+  {{- if .Values.sparkBackend.metrics.podMonitor.namespace }}
+  namespace: {{ .Values.sparkBackend.metrics.podMonitor.namespace }}
+  {{- else }}
+  namespace: {{ .Release.Namespace }}
+  {{- end }}
+  labels:
+    {{- include "common.labels.standard" . | nindent 4 }}
+    app.kubernetes.io/component: metrics
+    {{- if .Values.sparkBackend.metrics.podMonitor.additionalLabels }}
+    {{- include "common.tplvalues.render" (dict "value" .Values.sparkBackend.metrics.podMonitor.additionalLabels "context" $) | nindent 4 }}
+    {{- end }}
+    {{- if .Values.commonLabels }}
+    {{- include "common.tplvalues.render" ( dict "value" .Values.commonLabels "context" $ ) | nindent 4 }}
+    {{- end }}
+  {{- if .Values.commonAnnotations }}
+  annotations: {{- include "common.tplvalues.render" ( dict "value" .Values.commonAnnotations "context" $ ) | nindent 4 }}
+  {{- end }}
+spec:
+  podMetricsEndpoints:
+    - port: "4040"
+      path: /metrics/executors/prometheus/
+      {{- if .Values.sparkBackend.metrics.podMonitor.interval }}
+      interval: {{ .Values.sparkBackend.metrics.podMonitor.interval }}
+      {{- end }}
+      {{- if .Values.sparkBackend.metrics.podMonitor.scrapeTimeout }}
+      scrapeTimeout: {{ .Values.sparkBackend.metrics.podMonitor.scrapeTimeout }}
+      {{- end }}
+  {{- range .Values.sparkBackend.metrics.podMonitor.extraMetricsEndpoints }}
+    - port: {{ .port }}
+      path: {{ .path }}
+    {{- if $.Values.sparkBackend.metrics.podMonitor.interval }}
+      interval: {{ $.Values.sparkBackend.metrics.podMonitor.interval }}
+    {{- end }}
+    {{- if $.Values.sparkBackend.metrics.podMonitor.scrapeTimeout }}
+      scrapeTimeout: {{ $.Values.sparkBackend.metrics.podMonitor.scrapeTimeout }}
+    {{- end }}
+  {{- end }}
+  namespaceSelector:
+    matchNames:
+      - {{ .Release.Namespace }}
+  selector:
+    matchLabels: {{- include "common.labels.matchLabels" . | nindent 6 }}
+{{- end }}
diff --git a/charts/bullet/values.yaml b/charts/bullet/values.yaml
index 6baed5d..9329056 100644
--- a/charts/bullet/values.yaml
+++ b/charts/bullet/values.yaml
@@ -69,6 +69,15 @@ sparkBackend:
   schemaDir: /bullet/schemas
   executorCores: 2
   executorMemory: 3G
+  metrics:
+    enabled: false
+    podMonitor:
+      enabled: false
+      extraMetricsEndpoints: []
+      namespace: ""
+      interval: 30s
+      scrapeTimeout:
+      additionalLabels: {}
   config:
     # See https://github.com/bullet-db/bullet-spark/blob/master/src/main/resources/bullet_spark_defaults.yaml
     ### Bullet ###