parameter devicePlugin.deviceSplitCount does not work

i use helm to install k8s-vgpu-scheduler, set devicePlugin.deviceSplitCount = 5. after deployed successfully, i run 'kubectl describe node ', i can see the allocatable resources 'nvidia.com/gpu' count 40 (it has 8 A40 card in machine). Then i create 6 pod, every pod assign 1 'nvidia.com/gpu'， but when i create a pod which needs 3 'nvidia.com/gpu'，the k8s said the pod can't not be schedulerd.

the logs of vgpu-scheduler is showed below, it seems said only 2 gpu card can usable？
`I0313 00:58:35.594437 1 score.go:65] "devices status"
I0313 00:58:35.594467 1 score.go:67] "device status" device id="GPU-0707087e-8264-4ba4-bc45-30c70272ec4a" device detail={"Id":"GPU-0707087e-8264-4ba4-bc45-30c70272ec4a","Index":0,"Used":0,"Count":10,"Usedmem":0,"Totalmem":46068,"Totalcore":100,"Usedcores":0,"Numa":0,"Type":"NVIDIA-NVIDIA A40","Health":true}
I0313 00:58:35.594519 1 score.go:67] "device status" device id="GPU-b3e35ad4-81ee-0aee-9865-4787748b93ce" device detail={"Id":"GPU-b3e35ad4-81ee-0aee-9865-4787748b93ce","Index":1,"Used":0,"Count":10,"Usedmem":0,"Totalmem":46068,"Totalcore":100,"Usedcores":0,"Numa":0,"Type":"NVIDIA-NVIDIA A40","Health":true}
I0313 00:58:35.594542 1 score.go:67] "device status" device id="GPU-d38a391c-9f2f-395e-2f91-1785a648f6c4" device detail={"Id":"GPU-d38a391c-9f2f-395e-2f91-1785a648f6c4","Index":2,"Used":1,"Count":10,"Usedmem":46068,"Totalmem":46068,"Totalcore":100,"Usedcores":0,"Numa":0,"Type":"NVIDIA-NVIDIA A40","Health":true}
I0313 00:58:35.594568 1 score.go:67] "device status" device id="GPU-7099a282-5a75-55f8-0cd0-a4b48098ae1e" device detail={"Id":"GPU-7099a282-5a75-55f8-0cd0-a4b48098ae1e","Index":3,"Used":1,"Count":10,"Usedmem":46068,"Totalmem":46068,"Totalcore":100,"Usedcores":0,"Numa":0,"Type":"NVIDIA-NVIDIA A40","Health":true}
I0313 00:58:35.594600 1 score.go:67] "device status" device id="GPU-56967eb2-30b7-c808-367a-225b8bd8a12e" device detail={"Id":"GPU-56967eb2-30b7-c808-367a-225b8bd8a12e","Index":4,"Used":1,"Count":10,"Usedmem":46068,"Totalmem":46068,"Totalcore":100,"Usedcores":0,"Numa":0,"Type":"NVIDIA-NVIDIA A40","Health":true}
I0313 00:58:35.594639 1 score.go:67] "device status" device id="GPU-54191405-e5a9-2f7b-8ac4-f4e86c6669cb" device detail={"Id":"GPU-54191405-e5a9-2f7b-8ac4-f4e86c6669cb","Index":5,"Used":1,"Count":10,"Usedmem":46068,"Totalmem":46068,"Totalcore":100,"Usedcores":0,"Numa":0,"Type":"NVIDIA-NVIDIA A40","Health":true}
I0313 00:58:35.594671 1 score.go:67] "device status" device id="GPU-e731cd15-879f-6d00-485d-d1b468589de9" device detail={"Id":"GPU-e731cd15-879f-6d00-485d-d1b468589de9","Index":6,"Used":1,"Count":10,"Usedmem":46068,"Totalmem":46068,"Totalcore":100,"Usedcores":0,"Numa":0,"Type":"NVIDIA-NVIDIA A40","Health":true}
I0313 00:58:35.594693 1 score.go:67] "device status" device id="GPU-865edbf8-5d63-8e57-5e14-36682179eaf6" device detail={"Id":"GPU-865edbf8-5d63-8e57-5e14-36682179eaf6","Index":7,"Used":1,"Count":10,"Usedmem":46068,"Totalmem":46068,"Totalcore":100,"Usedcores":0,"Numa":0,"Type":"NVIDIA-NVIDIA A40","Health":true}
I0313 00:58:35.594725 1 score.go:90] "Allocating device for container request" pod="default/gpu-pod-2" card request={"Nums":5,"Type":"NVIDIA","Memreq":0,"MemPercentagereq":100,"Coresreq":0}
I0313 00:58:35.594757 1 score.go:93] "scoring pod" pod="default/gpu-pod-2" Memreq=0 MemPercentagereq=100 Coresreq=0 Nums=5 device index=7 device="GPU-b3e35ad4-81ee-0aee-9865-4787748b93ce"
I0313 00:58:35.594800 1 score.go:140] "first fitted" pod="default/gpu-pod-2" device="GPU-b3e35ad4-81ee-0aee-9865-4787748b93ce"
I0313 00:58:35.594829 1 score.go:93] "scoring pod" pod="default/gpu-pod-2" Memreq=0 MemPercentagereq=100 Coresreq=0 Nums=4 device index=6 device="GPU-0707087e-8264-4ba4-bc45-30c70272ec4a"
I0313 00:58:35.594850 1 score.go:140] "first fitted" pod="default/gpu-pod-2" device="GPU-0707087e-8264-4ba4-bc45-30c70272ec4a"
I0313 00:58:35.594869 1 score.go:93] "scoring pod" pod="default/gpu-pod-2" Memreq=0 MemPercentagereq=100 Coresreq=0 Nums=3 device index=5 device="GPU-865edbf8-5d63-8e57-5e14-36682179eaf6"
I0313 00:58:35.594889 1 score.go:93] "scoring pod" pod="default/gpu-pod-2" Memreq=0 MemPercentagereq=100 Coresreq=0 Nums=3 device index=4 device="GPU-e731cd15-879f-6d00-485d-d1b468589de9"
I0313 00:58:35.594911 1 score.go:93] "scoring pod" pod="default/gpu-pod-2" Memreq=0 MemPercentagereq=100 Coresreq=0 Nums=3 device index=3 device="GPU-54191405-e5a9-2f7b-8ac4-f4e86c6669cb"
I0313 00:58:35.594929 1 score.go:93] "scoring pod" pod="default/gpu-pod-2" Memreq=0 MemPercentagereq=100 Coresreq=0 Nums=3 device index=2 device="GPU-56967eb2-30b7-c808-367a-225b8bd8a12e"
I0313 00:58:35.594948 1 score.go:93] "scoring pod" pod="default/gpu-pod-2" Memreq=0 MemPercentagereq=100 Coresreq=0 Nums=3 device index=1 device="GPU-7099a282-5a75-55f8-0cd0-a4b48098ae1e"
I0313 00:58:35.594966 1 score.go:93] "scoring pod" pod="default/gpu-pod-2" Memreq=0 MemPercentagereq=100 Coresreq=0 Nums=3 device index=0 device="GPU-d38a391c-9f2f-395e-2f91-1785a648f6c4"
I0313 00:58:35.594989 1 score.go:211] "calcScore:node not fit pod" pod="default/gpu-pod-2" node="gpu-230"`

the kubectl describe node gpu-230 said:

the nvidia-smi said:

so somebody can solve this issue? thanks

The issue is about the devicePlugin.deviceSplitCount Helm parameter not working as expected. Start by examining the Helm chart values and the device plugin's configuration logic. Look at the scheduler's scoring logic in score.go (referenced in logs) to understand how device counts are calculated. Check how the allocatable resources are reported versus how the scheduler interprets them. The fix likely involves aligning the device split configuration with the scheduler's device counting and allocation logic.

Vom Indexierungsmodell aus dem Issue-Text verfasst.