fix panic when the gpu is faulty

Signed-off-by: guoqinwill <guoqinwill@163.com>
volcano-sh · Mar 20, 2024 · 42de699 · 42de699
1 parent 0f7626f
commit 42de699
Show file tree

Hide file tree

Showing 2 changed files with 4 additions and 3 deletions.
diff --git a/pkg/scheduler/api/resource_info.go b/pkg/scheduler/api/resource_info.go
@@ -260,8 +260,9 @@ 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 {
+// SubWithoutAssert subtracts two Resource objects without assertion,
+// this function is added because some resource subtraction allows negative results, while others do not.
+func (r *Resource) SubWithoutAssert(rr *Resource) *Resource {
 	ok, resources := rr.LessEqualWithResourcesName(r, Zero)
 	if !ok {
 		klog.Errorf("resources <%v> are not sufficient to do operation: <%v> sub <%v>", resources, r, rr)

diff --git a/pkg/scheduler/plugins/overcommit/overcommit.go b/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).WithoutAssertSub(used)
+	op.idleResource = op.totalResource.Clone().Multi(op.overCommitFactor).SubWithoutAssert(used)
 
 	for _, job := range ssn.Jobs {
 		// calculate inqueue job resources