Initial commit (with migration template)

2026-07-17 15:37:40 +00:00
commit 6cfbe6dcb5
63 changed files with 4992 additions and 0 deletions
--- a/monitoring/monitoring-stack/prometheus/prometheus.yml
+++ b/monitoring/monitoring-stack/prometheus/prometheus.yml
@@ -0,0 +1,21 @@
+global:
+  scrape_interval: 15s
+  evaluation_interval: 15s
+
+scrape_configs:
+  - job_name: 'litellm-proxy'
+    metrics_path: '/metrics'
+    static_configs:
+      - targets: ['litellm:4000']
+
+  - job_name: 'dgx-node-metrics'
+    static_configs:
+      - targets:
+          - '192.168.0.99:9100'
+          - '192.168.0.100:9100'
+
+  - job_name: 'dgx-gpu-metrics'
+    static_configs:
+      - targets:
+          - '192.168.0.99:9400'
+          - '192.168.0.100:9400'