chore: replace jmx exporter with built in Prometheus support

CHANGELOG.md

@@ -33,6 +33,7 @@ All notable changes to this project will be documented in this file.
   - The `runAsUser` and `runAsGroup` fields will not be set anymore by the operator
   - The defaults from the docker images itself will now apply, which will be different from 1000/0 going forward
   - This is marked as breaking because tools and policies might exist, which require these fields to be set
+- BREAKING: the JMX exporter has been an replaced with the built-in Prometheus servlet. The history server pods do not expose metrics anymore ([#584])
 
 ### Fixed
 
@@ -59,6 +60,7 @@ All notable changes to this project will be documented in this file.
 [#574]: https://github.com/stackabletech/spark-k8s-operator/pull/574
 [#580]: https://github.com/stackabletech/spark-k8s-operator/pull/580
 [#575]: https://github.com/stackabletech/spark-k8s-operator/pull/575
+[#584]: https://github.com/stackabletech/spark-k8s-operator/pull/584
 
 ## [25.3.0] - 2025-03-21
 

docs/modules/spark-k8s/pages/usage-guide/history-server.adoc

@@ -156,3 +156,11 @@ spark-history-node-cleaner   NodePort    10.96.203.43    <none>        18080:325
 By setting up port forwarding on 18080 the UI can be opened by pointing your browser to `http://localhost:18080`:
 
 image::history-server-ui.png[History Server Console]
+
+== Metrics
+
+[NOTE]
+====
+Up to version 25.3 of the Stackable Data Platform, the history server used the JMX exporter to expose metrics on a separate port.
+Starting with version 25.7 the JMX exporter has been removed and the history server doesn't expose metrics as of Spark version 3.5.6.
+====

docs/modules/spark-k8s/pages/usage-guide/operations/applications.adoc

@@ -8,3 +8,18 @@ As the operator creates the necessary resources, the status of the application t
 
 NOTE: The operator never reconciles an application once it has been created.
 To resubmit an application, a new SparkApplication resource must be created.
+
+== Metrics
+
+[NOTE]
+====
+Up to version 25.3 of the Stackable Data Platform, Spark applications used the JMX exporter to expose metrics on a separate port.
+Starting with version 25.7, the built-in Prometheus servlet is used instead.
+====
+
+Application driver pods expose Prometheus metrics at the following endpoints:
+
+* `/metrics/prometheus` for driver instances
+* `/metrics/executors/prometheus` for executor instances.
+
+These endpoints are available on the same port as the Spark UI, which is 4040 by default.

rust/operator-binary/src/connect/common.rs

@@ -34,6 +34,11 @@ pub enum Error {
     JvmSecurityProperties {
         source: product_config::writer::PropertiesWriterError,
     },
+
+    #[snafu(display("failed to serialize metrics properties",))]
+    MetricsProperties {
+        source: product_config::writer::PropertiesWriterError,
+    },
 }
 
 pub(crate) fn labels<'a, T>(
@@ -149,5 +154,5 @@ pub(crate) fn metrics_properties(
         );
     }
 
-    to_java_properties_string(result.iter()).context(JvmSecurityPropertiesSnafu)
+    to_java_properties_string(result.iter()).context(MetricsPropertiesSnafu)
 }

rust/operator-binary/src/crd/constants.rs

@@ -86,7 +86,6 @@ pub const SPARK_DEFAULTS_FILE_NAME: &str = "spark-defaults.conf";
 pub const SPARK_ENV_SH_FILE_NAME: &str = "spark-env.sh";
 
 pub const SPARK_CLUSTER_ROLE: &str = "spark-k8s-clusterrole";
-pub const METRICS_PORT: u16 = 18081;
 pub const HISTORY_UI_PORT: u16 = 18080;
 
 pub const LISTENER_VOLUME_NAME: &str = "listener";

rust/operator-binary/src/crd/mod.rs

@@ -649,6 +649,14 @@ impl v1alpha1::SparkApplication {
             )]);
         }
 
+        // Enable Prometheus metrics export
+        submit_cmd.extend(vec![
+            "--conf spark.metrics.conf.\\*.sink.prometheusServlet.class=org.apache.spark.metrics.sink.PrometheusServlet".to_string(),
+            "--conf spark.metrics.conf.\\*.sink.prometheusServlet.path=/metrics/prometheus".to_string(),
+            "--conf spark.ui.prometheus.enabled=true".to_string(),
+            "--conf spark.sql.streaming.metricsEnabled=true".to_string(),
+        ]);
+
         // some command elements need to be initially stored in a map (to allow overwrites) and
         // then added to the vector once complete.
         let mut submit_conf: BTreeMap<String, String> = BTreeMap::new();

rust/operator-binary/src/history/config/jvm.rs

@@ -5,9 +5,8 @@ use stackable_operator::role_utils::{
 
 use crate::crd::{
     constants::{
-        JVM_SECURITY_PROPERTIES_FILE, LOG4J2_CONFIG_FILE, METRICS_PORT,
-        STACKABLE_TLS_STORE_PASSWORD, STACKABLE_TRUST_STORE, VOLUME_MOUNT_PATH_CONFIG,
-        VOLUME_MOUNT_PATH_LOG_CONFIG,
+        JVM_SECURITY_PROPERTIES_FILE, LOG4J2_CONFIG_FILE, STACKABLE_TLS_STORE_PASSWORD,
+        STACKABLE_TRUST_STORE, VOLUME_MOUNT_PATH_CONFIG, VOLUME_MOUNT_PATH_LOG_CONFIG,
     },
     history::HistoryConfigFragment,
     logdir::ResolvedLogDir,
@@ -33,9 +32,6 @@ pub fn construct_history_jvm_args(
         format!(
             "-Djava.security.properties={VOLUME_MOUNT_PATH_CONFIG}/{JVM_SECURITY_PROPERTIES_FILE}"
         ),
-        format!(
-            "-javaagent:/stackable/jmx/jmx_prometheus_javaagent.jar={METRICS_PORT}:/stackable/jmx/config.yaml"
-        ),
     ];
 
     if logdir.tls_enabled() {
@@ -86,8 +82,7 @@ mod tests {
         assert_eq!(
             jvm_config,
             "-Dlog4j.configurationFile=/stackable/log_config/log4j2.properties \
-            -Djava.security.properties=/stackable/spark/conf/security.properties \
-            -javaagent:/stackable/jmx/jmx_prometheus_javaagent.jar=18081:/stackable/jmx/config.yaml"
+            -Djava.security.properties=/stackable/spark/conf/security.properties"
         );
     }
 
@@ -130,7 +125,6 @@ mod tests {
             jvm_config,
             "-Dlog4j.configurationFile=/stackable/log_config/log4j2.properties \
             -Djava.security.properties=/stackable/spark/conf/security.properties \
-            -javaagent:/stackable/jmx/jmx_prometheus_javaagent.jar=18081:/stackable/jmx/config.yaml \
             -Dhttps.proxyHost=proxy.my.corp \
             -Djava.net.preferIPv4Stack=true \
             -Dhttps.proxyPort=1234"

rust/operator-binary/src/history/history_controller.rs

@@ -56,11 +56,11 @@ use crate::{
         constants::{
             ACCESS_KEY_ID, HISTORY_APP_NAME, HISTORY_CONTROLLER_NAME, HISTORY_ROLE_NAME,
             HISTORY_UI_PORT, JVM_SECURITY_PROPERTIES_FILE, LISTENER_VOLUME_DIR,
-            LISTENER_VOLUME_NAME, MAX_SPARK_LOG_FILES_SIZE, METRICS_PORT, OPERATOR_NAME,
-            SECRET_ACCESS_KEY, SPARK_DEFAULTS_FILE_NAME, SPARK_ENV_SH_FILE_NAME,
-            SPARK_IMAGE_BASE_NAME, STACKABLE_TRUST_STORE, VOLUME_MOUNT_NAME_CONFIG,
-            VOLUME_MOUNT_NAME_LOG, VOLUME_MOUNT_NAME_LOG_CONFIG, VOLUME_MOUNT_PATH_CONFIG,
-            VOLUME_MOUNT_PATH_LOG, VOLUME_MOUNT_PATH_LOG_CONFIG,
+            LISTENER_VOLUME_NAME, MAX_SPARK_LOG_FILES_SIZE, OPERATOR_NAME, SECRET_ACCESS_KEY,
+            SPARK_DEFAULTS_FILE_NAME, SPARK_ENV_SH_FILE_NAME, SPARK_IMAGE_BASE_NAME,
+            STACKABLE_TRUST_STORE, VOLUME_MOUNT_NAME_CONFIG, VOLUME_MOUNT_NAME_LOG,
+            VOLUME_MOUNT_NAME_LOG_CONFIG, VOLUME_MOUNT_PATH_CONFIG, VOLUME_MOUNT_PATH_LOG,
+            VOLUME_MOUNT_PATH_LOG_CONFIG,
         },
         history::{self, HistoryConfig, SparkHistoryServerContainer, v1alpha1},
         listener_ext,
@@ -574,7 +574,6 @@ fn build_stateful_set(
         ])
         .args(command_args(log_dir))
         .add_container_port("http", HISTORY_UI_PORT.into())
-        .add_container_port("metrics", METRICS_PORT.into())
         .add_env_vars(merged_env)
         .add_volume_mounts(log_dir.volume_mounts())
         .context(AddVolumeMountSnafu)?

rust/operator-binary/src/spark_k8s_controller.rs

@@ -37,6 +37,7 @@ use stackable_operator::{
         core::{DeserializeGuard, error_boundary},
         runtime::{controller::Action, reflector::ObjectRef},
     },
+    kvp::Label,
     logging::controller::ReconcilerError,
     product_config_utils::ValidatedRoleConfigByPropertyKind,
     product_logging::{
@@ -610,27 +611,32 @@ fn pod_template(
         );
     }
 
+    let mut omb = ObjectMetaBuilder::new();
+    omb.name(&container_name)
+        // this reference is not pointing to a controller but only provides a UID that can used to clean up resources
+        // cleanly (specifically driver pods and related config maps) when the spark application is deleted.
+        .ownerreference_from_resource(spark_application, None, None)
+        .context(ObjectMissingMetadataForOwnerRefSnafu)?
+        .with_recommended_labels(
+            spark_application
+                .build_recommended_labels(&spark_image.app_version_label, &container_name),
+        )
+        .context(MetadataBuildSnafu)?;
+
+    // Only the driver pod should be scraped by Prometheus
+    // because the executor metrics are also available via /metrics/executors/prometheus/
+    if role == SparkApplicationRole::Driver {
+        omb.with_label(Label::try_from(("prometheus.io/scrape", "true")).context(LabelBuildSnafu)?);
+    }
+
     let mut pb = PodBuilder::new();
-    pb.metadata(
-        ObjectMetaBuilder::new()
-            .name(&container_name)
-            // this reference is not pointing to a controller but only provides a UID that can used to clean up resources
-            // cleanly (specifically driver pods and related config maps) when the spark application is deleted.
-            .ownerreference_from_resource(spark_application, None, None)
-            .context(ObjectMissingMetadataForOwnerRefSnafu)?
-            .with_recommended_labels(
-                spark_application
-                    .build_recommended_labels(&spark_image.app_version_label, &container_name),
-            )
-            .context(MetadataBuildSnafu)?
-            .build(),
-    )
-    .add_container(cb.build())
-    .add_volumes(volumes.to_vec())
-    .context(AddVolumeSnafu)?
-    .security_context(security_context())
-    .image_pull_secrets_from_product_image(spark_image)
-    .affinity(&config.affinity);
+    pb.metadata(omb.build())
+        .add_container(cb.build())
+        .add_volumes(volumes.to_vec())
+        .context(AddVolumeSnafu)?
+        .security_context(security_context())
+        .image_pull_secrets_from_product_image(spark_image)
+        .affinity(&config.affinity);
 
     let init_containers = init_containers(
         spark_application,

tests/templates/kuttl/spark-connect/30-assert.yaml

@@ -0,0 +1,14 @@
+---
+apiVersion: kuttl.dev/v1beta1
+kind: TestAssert
+timeout: 300
+commands:
+  - script: |
+      # This endpoint (/metrics/prometheus) is also used as liveliness probe
+      echo test Prometheus endpoint for driver metrics
+      DRIVER_METRIC_COUNT=$(kubectl exec spark-connect-server-0 -c spark -n $NAMESPACE -- curl localhost:4040/metrics/prometheus | grep _driver_ | wc -l)
+      test 0 -lt "$DRIVER_METRIC_COUNT"
+
+      echo test Prometheus endpoint for executor metrics
+      EXECUTOR_METRIC_COUNT=$(kubectl exec spark-connect-server-0 -c spark -n $NAMESPACE -- curl localhost:4040/metrics/executors/prometheus | grep _executor_ | wc -l)
+      test 0 -lt "$EXECUTOR_METRIC_COUNT"