diff kube.py @ 268:34ab4aec7d4b

notes and changes for getting nvidia gpu k3d support going, which was very hard
author drewp@bigasterisk.com
date Wed, 14 Feb 2024 18:48:08 -0800
parents 564b62e59484
children 665a199f7c8a
line wrap: on
line diff
--- a/kube.py	Wed Feb 14 18:45:31 2024 -0800
+++ b/kube.py	Wed Feb 14 18:48:08 2024 -0800
@@ -56,6 +56,7 @@
     if is_pi:
         pi_cgroup_setup()
 
+
 # don't try to get aufs-dkms on rpi-- https://github.com/docker/for-linux/issues/709
 def podman_insecure_registry(reg):
     files.template(src='templates/kube/podman_registries.conf.j2', dest='/etc/containers/registries.conf.d/reg.conf', reg=reg)
@@ -91,8 +92,17 @@
         dest=f'/etc/systemd/system/{service_name}',
         role=role,
     )
-    if host.name in ['dash', 'bang', 'ditto']:
-        setupNvidiaToolkit()
+    if host.name in ['bang', 'garage']:
+        # no supported gpu
+        '''
+            kubectl label --overwrite node bang nvidia.com/gpu.deploy.gpu-feature-discovery=false
+            kubectl label --overwrite node bang nvidia.com/gpu.deploy.container-toolkit=false
+            kubectl label --overwrite node bang nvidia.com/gpu.deploy.dcgm-exporter=false
+            kubectl label --overwrite node bang nvidia.com/gpu.deploy.device-plugin=false
+            kubectl label --overwrite node bang nvidia.com/gpu.deploy.driver=false
+            kubectl label --overwrite node bang nvidia.com/gpu.deploy.mig-manager=false
+            kubectl label --overwrite node bang nvidia.com/gpu.deploy.operator-validator=false
+        '''
     systemd.service(service=service_name, daemon_reload=True, enabled=True, restarted=True)