From cbe08eb53dca65469abd2a23f276206c35d2df65 Mon Sep 17 00:00:00 2001
From: Jason Kincl <jkincl@redhat.com>
Date: Mon, 30 Oct 2023 11:58:57 -0400
Subject: [PATCH] Adding service monitor scraping for slurm

---
 slurm/manifests/kustomization.yaml  |  1 +
 slurm/manifests/namespace.yaml      |  4 ++-
 slurm/manifests/service.yaml        |  7 ++---
 slurm/manifests/servicemonitor.yaml | 42 +++++++++++++++++++++++++++++
 4 files changed, 50 insertions(+), 4 deletions(-)
 create mode 100644 slurm/manifests/servicemonitor.yaml

diff --git a/slurm/manifests/kustomization.yaml b/slurm/manifests/kustomization.yaml
index d83f087..c633e30 100644
--- a/slurm/manifests/kustomization.yaml
+++ b/slurm/manifests/kustomization.yaml
@@ -11,6 +11,7 @@ resources:
   - serviceaccount.yaml
   - clusterrolebinding.yaml
   - claim.yaml
+  - servicemonitor.yaml
 
 configMapGenerator:
 - name: slurm-conf
diff --git a/slurm/manifests/namespace.yaml b/slurm/manifests/namespace.yaml
index e8a5ec6..d4bbd52 100644
--- a/slurm/manifests/namespace.yaml
+++ b/slurm/manifests/namespace.yaml
@@ -1,4 +1,6 @@
 apiVersion: v1
 kind: Namespace
 metadata:
-  name: slurm-system
\ No newline at end of file
+  name: slurm-system
+  labels:
+    openshift.io/cluster-monitoring: 'true'
\ No newline at end of file
diff --git a/slurm/manifests/service.yaml b/slurm/manifests/service.yaml
index d38733f..9a75d09 100644
--- a/slurm/manifests/service.yaml
+++ b/slurm/manifests/service.yaml
@@ -8,8 +8,9 @@ spec:
   ports:
   - port: 6817
     name: slurmctld
-  - port: 6818
-    name: slurmd
+  - port: 8080
+    name: scrape
   clusterIP: None
   selector:
-    app: slurm
\ No newline at end of file
+    app: slurm
+    component: head
\ No newline at end of file
diff --git a/slurm/manifests/servicemonitor.yaml b/slurm/manifests/servicemonitor.yaml
new file mode 100644
index 0000000..ed6ddb6
--- /dev/null
+++ b/slurm/manifests/servicemonitor.yaml
@@ -0,0 +1,42 @@
+apiVersion: monitoring.coreos.com/v1
+kind: ServiceMonitor
+metadata:
+  name: slurm
+spec:
+  endpoints:
+  - interval: 30s
+    port: scrape
+    scheme: http
+  selector:
+    matchLabels:
+      app: slurm
+      component: head
+---
+kind: RoleBinding
+apiVersion: rbac.authorization.k8s.io/v1
+metadata:
+  name: prometheus-k8s
+subjects:
+  - kind: ServiceAccount
+    name: prometheus-k8s
+    namespace: openshift-monitoring
+roleRef:
+  apiGroup: rbac.authorization.k8s.io
+  kind: Role
+  name: prometheus-k8s
+---
+kind: Role
+apiVersion: rbac.authorization.k8s.io/v1
+metadata:
+  name: prometheus-k8s
+rules:
+  - verbs:
+      - get
+      - list
+      - watch
+    apiGroups:
+      - ''
+    resources:
+      - services
+      - endpoints
+      - pods