feat(monitoring): enable JMX and OTEL for frontend pods (#5834)

docker/datahub-frontend/Dockerfile

@@ -18,7 +18,23 @@ RUN apk --no-cache --update-cache --available upgrade \
 ARG USE_SYSTEM_NODE="true"
 ENV CI=true
 ENV GRADLE_OPTS="-Xms256m -Xmx512m"
-COPY . datahub-src
+
+# copy only what's realy needed for the build
+# to get the benefits of caching as much as possible
+COPY ./datahub-frontend ./datahub-src/datahub-frontend
+COPY ./entity-registry ./datahub-src/entity-registry
+COPY ./buildSrc ./datahub-src/buildSrc
+COPY ./datahub-web-react ./datahub-src/datahub-web-react
+COPY ./li-utils ./datahub-src/li-utils
+COPY ./metadata-models ./datahub-src/metadata-models
+COPY ./metadata-models-validator ./datahub-src/metadata-models-validator
+COPY ./metadata-utils ./datahub-src/metadata-utils
+COPY ./metadata-service ./datahub-src/metadata-service
+COPY ./metadata-io ./datahub-src/metadata-io
+COPY ./datahub-graphql-core ./datahub-src/datahub-graphql-core
+COPY ./gradle ./datahub-src/gradle
+COPY repositories.gradle gradle.properties gradlew settings.gradle build.gradle ./datahub-src/
+
 RUN cd datahub-src \
     && ./gradlew :datahub-web-react:build -x test -x yarnTest -x yarnLint \
     && ./gradlew :datahub-frontend:dist -PuseSystemNode=${USE_SYSTEM_NODE} -x test -x yarnTest -x yarnLint \
@@ -27,6 +43,9 @@ RUN cd datahub-src \
 
 FROM base as prod-install
 COPY --from=prod-build /datahub-frontend /datahub-frontend/
+COPY ./docker/monitoring/client-prometheus-config.yaml /datahub-frontend/
+RUN wget https://github.com/open-telemetry/opentelemetry-java-instrumentation/releases/download/v1.4.1/opentelemetry-javaagent-all.jar -O opentelemetry-javaagent-all.jar \
+    && wget https://repo1.maven.org/maven2/io/prometheus/jmx/jmx_prometheus_javaagent/0.16.1/jmx_prometheus_javaagent-0.16.1.jar -O jmx_prometheus_javaagent.jar
 RUN chown -R datahub:datahub /datahub-frontend && chmod 755 /datahub-frontend
 
 FROM base as dev-install
@@ -43,14 +62,5 @@ RUN echo $SERVER_PORT
 EXPOSE $SERVER_PORT
 
 HEALTHCHECK --start-period=2m --retries=4 CMD curl --fail http://localhost:$SERVER_PORT/admin || exit 1
-
-ENV JAVA_OPTS=" \
-   -Xms512m \
-   -Xmx1024m \
-   -Dhttp.port=$SERVER_PORT \
-   -Dconfig.file=datahub-frontend/conf/application.conf \
-   -Djava.security.auth.login.config=datahub-frontend/conf/jaas.conf \
-   -Dlogback.configurationFile=datahub-frontend/conf/logback.xml \
-   -Dlogback.debug=false \
-   -Dpidfile.path=/dev/null"
-CMD ["datahub-frontend/bin/datahub-frontend"]
+COPY ./docker/datahub-frontend/start.sh /
+CMD ./start.sh

docker/datahub-frontend/start.sh

@@ -0,0 +1,27 @@
+#!/bin/sh
+set -u
+
+PROMETHEUS_AGENT=""
+if [[ ${ENABLE_PROMETHEUS:-false} == true ]]; then
+  PROMETHEUS_AGENT="-javaagent:jmx_prometheus_javaagent.jar=4318:/datahub-frontend/client-prometheus-config.yaml"
+fi
+
+OTEL_AGENT=""
+if [[ ${ENABLE_OTEL:-false} == true ]]; then
+  OTEL_AGENT="-javaagent:/opentelemetry-javaagent-all.jar"
+fi
+
+# make sure there is no whitespace at the beginning and the end of 
+# this string
+export JAVA_OPTS="-Xms512m \
+   -Xmx1024m \
+   -Dhttp.port=$SERVER_PORT \
+   -Dconfig.file=datahub-frontend/conf/application.conf \
+   -Djava.security.auth.login.config=datahub-frontend/conf/jaas.conf \
+   -Dlogback.configurationFile=datahub-frontend/conf/logback.xml \
+   -Dlogback.debug=false \
+   ${PROMETHEUS_AGENT:-} ${OTEL_AGENT:-}
+   -Dpidfile.path=/dev/null"
+
+exec ./datahub-frontend/bin/datahub-frontend
+

docker/monitoring/docker-compose.monitoring.yml

@@ -12,6 +12,17 @@ services:
     ports:
       - "4318"
 
+  datahub-frontend-react:
+    environment:
+      - ENABLE_PROMETHEUS=true
+      - ENABLE_OTEL=true
+      - OTEL_TRACES_EXPORTER=jaeger
+      - OTEL_EXPORTER_JAEGER_ENDPOINT=http://jaeger-all-in-one:14250
+      - OTEL_METRICS_EXPORTER=none
+      - OTEL_SERVICE_NAME=datahub-gms
+    ports:
+      - "4318"
+
   # Jaeger
   jaeger-all-in-one:
     image: jaegertracing/all-in-one:latest

docker/monitoring/prometheus.yaml

@@ -6,4 +6,5 @@ scrape_configs:
         - 'host.docker.internal:4318'
         - 'datahub-gms:4318'
         - 'datahub-mae-consumer:4318'
-        - 'datahub-mce-consumer:4318'
+        - 'datahub-mce-consumer:4318'
+        - 'datahub-frontend-react:4318'

docker/quickstart.sh

@@ -24,6 +24,11 @@ TAG_VERSION=$(cd $DIR && git name-rev --name-only --tags HEAD)
 DEFAULT_VERSION=$(echo $TAG_VERSION | sed 's/undefined/head/')
 export DATAHUB_VERSION=${DATAHUB_VERSION:-${DEFAULT_VERSION}}
 
+M1_COMPOSE=""
+if [[ $(uname -m) == 'arm64' && $(uname) == 'Darwin' ]]; then
+  M1_COMPOSE="-f docker-compose.m1.yml"
+fi
+
 echo "Quickstarting DataHub: version ${DATAHUB_VERSION}"
 if docker volume ls | grep -c -q datahub_neo4jdata
 then
@@ -32,10 +37,7 @@ then
 else
   echo "No Datahub Neo4j volume found, starting with elasticsearch as graph service"
   cd $DIR && \
-  docker-compose \
-    -f quickstart/docker-compose-without-neo4j.quickstart.yml \
-    $MONITORING_COMPOSE $CONSUMERS_COMPOSE pull && \
   docker-compose -p datahub \
     -f quickstart/docker-compose-without-neo4j.quickstart.yml \
-    $MONITORING_COMPOSE $CONSUMERS_COMPOSE up $@
+    $MONITORING_COMPOSE $CONSUMERS_COMPOSE $M1_COMPOSE up $@
 fi

docker/quickstart/docker-compose.monitoring.quickstart.yml

@@ -1,4 +1,14 @@
 services:
+  datahub-frontend-react:
+    environment:
+    - ENABLE_PROMETHEUS=true
+    - ENABLE_OTEL=true
+    - OTEL_TRACES_EXPORTER=jaeger
+    - OTEL_EXPORTER_JAEGER_ENDPOINT=http://jaeger-all-in-one:14250
+    - OTEL_METRICS_EXPORTER=none
+    - OTEL_SERVICE_NAME=datahub-gms
+    ports:
+    - '4318'
   datahub-gms:
     environment:
     - ENABLE_PROMETHEUS=true