chore: ensure metrics are correctly exposed

Author: xeniape

rust/operator-binary/src/druid_controller.rs

@@ -73,10 +73,11 @@ use crate::{
         Container, DB_PASSWORD_ENV, DB_USERNAME_ENV, DRUID_CONFIG_DIRECTORY, DS_BUCKET,
         DeepStorageSpec, DruidClusterStatus, DruidRole, EXTENSIONS_LOADLIST, HDFS_CONFIG_DIRECTORY,
         JVM_CONFIG, JVM_SECURITY_PROPERTIES_FILE, LOG_CONFIG_DIRECTORY, MAX_DRUID_LOG_FILES_SIZE,
-        OPERATOR_NAME, RUNTIME_PROPS, RW_CONFIG_DIRECTORY, S3_ACCESS_KEY, S3_ENDPOINT_URL,
-        S3_PATH_STYLE_ACCESS, S3_SECRET_KEY, STACKABLE_LOG_DIR, ZOOKEEPER_CONNECTION_STRING,
-        authentication::AuthenticationClassesResolved, authorization::DruidAuthorization,
-        build_recommended_labels, build_string_list, security::DruidTlsSecurity, v1alpha1,
+        METRICS_PORT, METRICS_PORT_NAME, OPERATOR_NAME, RUNTIME_PROPS, RW_CONFIG_DIRECTORY,
+        S3_ACCESS_KEY, S3_ENDPOINT_URL, S3_PATH_STYLE_ACCESS, S3_SECRET_KEY, STACKABLE_LOG_DIR,
+        ZOOKEEPER_CONNECTION_STRING, authentication::AuthenticationClassesResolved,
+        authorization::DruidAuthorization, build_recommended_labels, build_string_list,
+        security::DruidTlsSecurity, v1alpha1,
     },
     discovery::{self, build_discovery_configmaps},
     extensions::get_extension_list,
@@ -1078,6 +1079,7 @@ fn build_rolegroup_statefulset(
         .args(vec![main_container_commands.join("\n")])
         .add_env_vars(rest_env)
         .add_container_ports(druid_tls_security.container_ports(role))
+        .add_container_port(METRICS_PORT_NAME, METRICS_PORT.into())
         // 10s * 30 = 300s to come up
         .startup_probe(druid_tls_security.get_tcp_socket_probe(30, 10, 30, 3))
         // 10s * 1 = 10s to get removed from service

rust/operator-binary/src/service.rs

@@ -4,7 +4,7 @@ use snafu::{ResultExt, Snafu};
 use stackable_operator::{
     builder::meta::ObjectMetaBuilder,
     k8s_openapi::api::core::v1::{Service, ServicePort, ServiceSpec},
-    kvp::{Label, ObjectLabels},
+    kvp::{Annotations, Label, ObjectLabels},
     role_utils::RoleGroupRef,
 };
 
@@ -85,6 +85,7 @@ pub fn build_rolegroup_metrics_service(
             .with_recommended_labels(object_labels)
             .context(MetadataBuildSnafu)?
             .with_label(Label::try_from(("prometheus.io/scrape", "true")).context(LabelBuildSnafu)?)
+            .with_annotations(prometheus_annotations())
             .build(),
         spec: Some(ServiceSpec {
             // Internal communication does not need to be exposed
@@ -117,3 +118,18 @@ fn rolegroup_metrics_service_name(role_group_ref_object_name: &str) -> String {
 pub fn rolegroup_headless_service_name(role_group_ref_object_name: &str) -> String {
     format!("{role_group_ref_object_name}-{HEADLESS_SERVICE_SUFFIX}")
 }
+
+/// Common annotations for Prometheus
+///
+/// These annotations can be used in a ServiceMonitor.
+///
+/// see also <https://github.com/prometheus-community/helm-charts/blob/prometheus-27.32.0/charts/prometheus/values.yaml#L983-L1036>
+fn prometheus_annotations() -> Annotations {
+    Annotations::try_from([
+        ("prometheus.io/path".to_owned(), "/metrics".to_owned()),
+        ("prometheus.io/port".to_owned(), METRICS_PORT.to_string()),
+        ("prometheus.io/scheme".to_owned(), "http".to_owned()),
+        ("prometheus.io/scrape".to_owned(), "true".to_owned()),
+    ])
+    .expect("should be valid annotations")
+}