Treat zero cluster UUID or clique ID as non-fabric-attached

elezar · elezar · commit fdb3f6038af5 · 2025-02-12T14:13:39.000+01:00
Signed-off-by: Evan Lezar &lt;elezar@nvidia.com&gt;
diff --git a/pkg/nvlib/device/device.go b/pkg/nvlib/device/device.go
@@ -219,6 +219,9 @@ func (d *device) IsFabricAttached() (bool, error) {
 		if ret != nvml.SUCCESS {
 			return false, fmt.Errorf("error getting GPU Fabric Info: %v", ret)
 		}
+		if info.ClusterUuid == [16]uint8{} || info.CliqueId == 0 {
+			return false, nil
+		}
 		if info.State != nvml.GPU_FABRIC_STATE_COMPLETED {
 			return false, nil
 		}
@@ -237,6 +240,9 @@ func (d *device) IsFabricAttached() (bool, error) {
 		if ret != nvml.SUCCESS {
 			return false, fmt.Errorf("error getting GPU Fabric Info: %v", ret)
 		}
+		if info.ClusterUuid == [16]uint8{} || info.CliqueId == 0 {
+			return false, nil
+		}
 		if info.State != nvml.GPU_FABRIC_STATE_COMPLETED {
 			return false, nil
 		}

Original file line number	Diff line number	Diff line change
`@@ -219,6 +219,9 @@ func (d *device) IsFabricAttached() (bool, error) {`
`219`	`219`	`if ret != nvml.SUCCESS {`
`220`	`220`	`return false, fmt.Errorf("error getting GPU Fabric Info: %v", ret)`
`221`	`221`	`}`
	`222`	`+ if info.ClusterUuid == [16]uint8{} \|\| info.CliqueId == 0 {`
	`223`	`+ return false, nil`
	`224`	`+ }`
`222`	`225`	`if info.State != nvml.GPU_FABRIC_STATE_COMPLETED {`
`223`	`226`	`return false, nil`
`224`	`227`	`}`
`@@ -237,6 +240,9 @@ func (d *device) IsFabricAttached() (bool, error) {`
`237`	`240`	`if ret != nvml.SUCCESS {`
`238`	`241`	`return false, fmt.Errorf("error getting GPU Fabric Info: %v", ret)`
`239`	`242`	`}`
	`243`	`+ if info.ClusterUuid == [16]uint8{} \|\| info.CliqueId == 0 {`
	`244`	`+ return false, nil`
	`245`	`+ }`
`240`	`246`	`if info.State != nvml.GPU_FABRIC_STATE_COMPLETED {`
`241`	`247`	`return false, nil`
`242`	`248`	`}`