prometheus.yaml

prometheus.yamlinfrastructure/kubernetes/observability/prometheus.yamlraw ← observability3,357 B← Back to the app
1# =============================================================================
2# Prometheus (TASK-0135, phase-15) — Deployment + PVC + Service, ns `monitoring`.
3#
4# Config is generated into `prometheus-config` by kustomization.yaml FROM
5# infrastructure/observability/prometheus/ (single source of truth):
6#   /etc/prometheus/prometheus.yml                 (scrape + rule_files + alerting)
7#   /etc/prometheus/rules/recording-rules.yaml
8#   /etc/prometheus/rules/alerts.yaml
9# Retention 7 d (the 14-day cert alert needs >14 d of history for its
10# avg_over_time-free expr, but 7 d is the staging default — bump if the
11# dashboards' 30 d views matter).
12# Image pinned: prom/prometheus:v2.51.0 (the 2.51 LTS line — classic text
13# format + the rule files above are 2.x syntax).
14# =============================================================================
15apiVersion: v1
16kind: PersistentVolumeClaim
17metadata:
18  name: prometheus-data
19  namespace: monitoring
20  labels:
21    app.kubernetes.io/name: prometheus
22    app.kubernetes.io/part-of: ai-software-factory
23spec:
24  accessModes: ["ReadWriteOnce"]
25  resources:
26    requests:
27      storage: 10Gi
28---
29apiVersion: apps/v1
30kind: Deployment
31metadata:
32  name: prometheus
33  namespace: monitoring
34  labels:
35    app.kubernetes.io/name: prometheus
36    app.kubernetes.io/part-of: ai-software-factory
37spec:
38  replicas: 1
39  selector:
40    matchLabels:
41      app.kubernetes.io/name: prometheus
42  strategy:
43    type: Recreate
44  template:
45    metadata:
46      labels:
47        app.kubernetes.io/name: prometheus
48    spec:
49      securityContext:
50        runAsNonRoot: true
51        runAsUser: 65534  # the `prom` user in the official image
52        runAsGroup: 65534
53        fsGroup: 65534
54      containers:
55        - name: prometheus
56          image: prom/prometheus:v2.51.0
57          imagePullPolicy: IfNotPresent
58          args:
59            - --config.file=/etc/prometheus/prometheus.yml
60            - --storage.tsdb.path=/prometheus
61            - --storage.tsdb.retention.time=7d
62            - --web.enable-lifecycle
63          ports:
64            - containerPort: 9090
65              name: http
66          volumeMounts:
67            - name: config
68              mountPath: /etc/prometheus
69              readOnly: true
70            - name: data
71              mountPath: /prometheus
72          readinessProbe:
73            httpGet:
74              path: /-/ready
75              port: 9090
76            initialDelaySeconds: 10
77            periodSeconds: 10
78          livenessProbe:
79            httpGet:
80              path: /-/healthy
81              port: 9090
82            initialDelaySeconds: 30
83            periodSeconds: 15
84          resources:
85            requests:
86              cpu: 250m
87              memory: 512Mi
88            limits:
89              cpu: "1"
90              memory: 2Gi
91      volumes:
92        - name: config
93          configMap:
94            name: prometheus-config  # name rewritten to the hashed name by kustomize
95        - name: data
96          persistentVolumeClaim:
97            claimName: prometheus-data
98---
99apiVersion: v1
100kind: Service
101metadata:
102  name: prometheus
103  namespace: monitoring
104  labels:
105    app.kubernetes.io/name: prometheus
106    app.kubernetes.io/part-of: ai-software-factory
107spec:
108  selector:
109    app.kubernetes.io/name: prometheus
110  ports:
111    - name: http
112      port: 9090
113      targetPort: 9090
114