ai-prio-class

rwlove · Jan 12, 2025 · 4710193 · 4710193
1 parent d531feb
commit 4710193
Show file tree

Hide file tree

Showing 6 changed files with 16 additions and 0 deletions.
diff --git a/kubernetes/main/apps/ai/kustomization.yaml b/kubernetes/main/apps/ai/kustomization.yaml
@@ -3,6 +3,7 @@
 apiVersion: kustomize.config.k8s.io/v1beta1
 kind: Kustomization
 resources:
+  - ./priority-class.yaml
   - ./k8sgpt/ks.yaml
   - ./ollama/ks.yaml
   - ./stable-diffusion/ks.yaml

diff --git a/kubernetes/main/apps/ai/ollama/app/helmrelease.yaml b/kubernetes/main/apps/ai/ollama/app/helmrelease.yaml
@@ -69,6 +69,8 @@ spec:
 
     runtimeClassName: nvidia
 
+    priorityClassName: ai-gpu-critical
+
     securityContext:
       privileged: true
 

diff --git a/kubernetes/main/apps/ai/priority-class.yaml b/kubernetes/main/apps/ai/priority-class.yaml
@@ -0,0 +1,8 @@
+---
+apiVersion: scheduling.k8s.io/v1
+description: Used for pods that must run on a node with a GPU.
+kind: PriorityClass
+metadata:
+  name: ai-gpu-critical
+preemptionPolicy: PreemptLowerPriority
+value: 100040
diff --git a/kubernetes/main/apps/ai/stable-diffusion/comfyui/helmrelease.yaml b/kubernetes/main/apps/ai/stable-diffusion/comfyui/helmrelease.yaml
@@ -36,6 +36,8 @@ spec:
         pod:
           runtimeClassName: nvidia
 
+          priorityClassName: ai-gpu-critical
+
           affinity:
             nodeAffinity:
               requiredDuringSchedulingIgnoredDuringExecution:

diff --git a/kubernetes/main/apps/media/immich/app/helmrelease.yaml b/kubernetes/main/apps/media/immich/app/helmrelease.yaml
@@ -111,6 +111,8 @@ spec:
     machine-learning:
       runtimeClassName: nvidia
 
+      priorityClassName: ai-gpu-critical
+
       securityContext:
         privileged: true
 

diff --git a/kubernetes/main/apps/monitoring/exporters/dcgm-exporter/app/helm-release.yaml b/kubernetes/main/apps/monitoring/exporters/dcgm-exporter/app/helm-release.yaml
@@ -37,6 +37,7 @@ spec:
     nodeSelector:
       nvidia.com/gpu.present: "true"
     runtimeClassName: nvidia
+    priorityClassName: ai-gpu-critical
     resources:
       limits:
         nvidia.com/gpu: 1