fix panic when the gpu is faulty

Signed-off-by: guoqinwill <guoqinwill@163.com>

pkg/scheduler/api/resource_info.go

@@ -260,6 +260,14 @@ func (r *Resource) Sub(rr *Resource) *Resource {
 	return r.sub(rr)
 }
 
+// WithoutAssertSub subtracts two Resource objects without assertion.
+func (r *Resource) WithoutAssertSub(rr *Resource) *Resource {
+	if !rr.LessEqual(r, Zero) {
+		klog.Errorf("resource is not sufficient to do operation: <%v> sub <%v> , the subtraction result is <%v>", r, rr, r.sub(rr))
+	}
+	return r.sub(rr)
+}
+
 // sub subtracts two Resource objects.
 func (r *Resource) sub(rr *Resource) *Resource {
 	r.MilliCPU -= rr.MilliCPU

pkg/scheduler/plugins/overcommit/overcommit.go

@@ -88,7 +88,7 @@ func (op *overcommitPlugin) OnSessionOpen(ssn *framework.Session) {
 	for _, node := range ssn.Nodes {
 		used.Add(node.Used)
 	}
-	op.idleResource = op.totalResource.Clone().Multi(op.overCommitFactor).Sub(used)
+	op.idleResource = op.totalResource.Clone().Multi(op.overCommitFactor).WithoutAssertSub(used)
 
 	for _, job := range ssn.Jobs {
 		// calculate inqueue job resources