[nvidia] Capture more nvidia commands

Capture commands related to nvidia container toolkit. Related: RHEL-58172 Signed-off-by: Jose Castillo <[email protected]>
sosreport · Sep 12, 2024 · 5ca4480 · 5ca4480
1 parent a946343
commit 5ca4480
Showing 1 changed file with 10 additions and 1 deletion.
diff --git a/sos/report/plugins/nvidia.py b/sos/report/plugins/nvidia.py
@@ -16,7 +16,7 @@ class Nvidia(Plugin, IndependentPlugin):
 
     short_desc = 'Nvidia GPU information'
     plugin_name = 'nvidia'
-    commands = ('nvidia-smi',)
+    commands = ('nvidia-smi', 'nvidia-ctk',)
 
     def setup(self):
         subcmds = [
@@ -30,8 +30,17 @@ def setup(self):
             'nvlink -e'
         ]
 
+        ctk_subcmds = [
+            'cdi list',
+            '--version',
+        ]
+
+        self.add_copy_spec("/etc/cdi/nvidia.yaml")
         self.add_service_status("nvidia-persistenced")
+        self.add_service_status("nvidia-fabricmanager")
+        self.add_service_status("nvidia-toolkit-firstboot")
         self.add_cmd_output([f"nvidia-smi {cmd}" for cmd in subcmds])
+        self.add_cmd_output([f"nvidia-ctk {cmd}" for cmd in ctk_subcmds])
 
         query = ('gpu_name,gpu_bus_id,vbios_version,temperature.gpu,'
                  'utilization.gpu,memory.total,memory.free,memory.used,'