From 7a35d3bbf9405e38fc53a8d0bd269aa61e78f8f0 Mon Sep 17 00:00:00 2001
From: Maneesha Herath <145188213+Maneeshaherath@users.noreply.github.com>
Date: Sat, 9 May 2026 17:57:22 +0530
Subject: [PATCH] fix(worker): remove duplicate metrics bind; add Prometheus
 scrape for worker

---
 infra/k8s/base/kustomization.yaml          |  2 ++
 infra/k8s/base/worker-deployment.yaml      |  3 +++
 infra/k8s/base/worker-metrics-service.yaml | 15 +++++++++++++++
 infra/k8s/base/worker-servicemonitor.yaml  | 16 ++++++++++++++++
 src/d2/config.py                           |  5 +----
 5 files changed, 37 insertions(+), 4 deletions(-)
 create mode 100644 infra/k8s/base/worker-metrics-service.yaml
 create mode 100644 infra/k8s/base/worker-servicemonitor.yaml

diff --git a/infra/k8s/base/kustomization.yaml b/infra/k8s/base/kustomization.yaml
index 6fea0ec..cffcc1d 100644
--- a/infra/k8s/base/kustomization.yaml
+++ b/infra/k8s/base/kustomization.yaml
@@ -13,5 +13,7 @@ resources:
   - api-ingress-tls.yaml
   - api-service.yaml
   - api-servicemonitor.yaml
+  - worker-metrics-service.yaml
+  - worker-servicemonitor.yaml
   - migrate-job.yaml
   - worker-deployment.yaml
diff --git a/infra/k8s/base/worker-deployment.yaml b/infra/k8s/base/worker-deployment.yaml
index 3177687..4214091 100644
--- a/infra/k8s/base/worker-deployment.yaml
+++ b/infra/k8s/base/worker-deployment.yaml
@@ -30,6 +30,9 @@ spec:
         - name: worker
           image: ghcr.io/healthcare-monitoring-system/d2-data-intelligence:sha-a280ecc715a943d47c8b3ed635dc61e7f9f69704
           imagePullPolicy: Always
+          ports:
+            - name: metrics
+              containerPort: 8001
           command:
             - python
             - -m
diff --git a/infra/k8s/base/worker-metrics-service.yaml b/infra/k8s/base/worker-metrics-service.yaml
new file mode 100644
index 0000000..78e51ba
--- /dev/null
+++ b/infra/k8s/base/worker-metrics-service.yaml
@@ -0,0 +1,15 @@
+# Headless Prometheus scrape target for worker /metrics (ML + Kafka pipeline counters).
+apiVersion: v1
+kind: Service
+metadata:
+  name: d2-worker-metrics
+  labels:
+    app: d2-worker
+    d2-metrics: worker
+spec:
+  selector:
+    app: d2-worker
+  ports:
+    - name: metrics
+      port: 8001
+      targetPort: metrics
diff --git a/infra/k8s/base/worker-servicemonitor.yaml b/infra/k8s/base/worker-servicemonitor.yaml
new file mode 100644
index 0000000..48371dd
--- /dev/null
+++ b/infra/k8s/base/worker-servicemonitor.yaml
@@ -0,0 +1,16 @@
+# Prometheus Operator: scrape d2-worker /metrics when release label matches the kube-stack.
+apiVersion: monitoring.coreos.com/v1
+kind: ServiceMonitor
+metadata:
+  name: d2-worker
+  labels:
+    release: monitoring
+spec:
+  selector:
+    matchLabels:
+      app: d2-worker
+      d2-metrics: worker
+  endpoints:
+    - port: metrics
+      path: /metrics
+      interval: 30s
diff --git a/src/d2/config.py b/src/d2/config.py
index 2b84e84..4fcd05d 100644
--- a/src/d2/config.py
+++ b/src/d2/config.py
@@ -85,7 +85,7 @@ def influx_configured(self) -> bool:
     # Observability
     worker_metrics_enabled: bool = True
     worker_metrics_host: str = "0.0.0.0"
-    worker_metrics_port: int = 9102
+    worker_metrics_port: int = 8001
 
     # Optional Keycloak checks (D4 will supply values)
     keycloak_well_known_url: str | None = None
@@ -104,9 +104,6 @@ def ensure_asyncpg_driver_in_database_url(self) -> Self:
             self.database_url = "postgresql+asyncpg://" + raw.removeprefix("postgres://")
         return self
 
-    # Observability — port the worker exposes Prometheus metrics on. Set to 0 to disable.
-    worker_metrics_port: int = 8001
-
 
 @lru_cache
 def get_settings() -> Settings: