avoid preempt of failed co-schd pods (#317)

Author: asm582

pkg/controller/queuejobresources/utils.go

@@ -61,6 +61,10 @@ func PendingPodsFailedSchd(pods []*v1.Pod) map[string][]v1.PodCondition {
 if strings.Contains(cond.Message, "pgName") && strings.Contains(cond.Message, "last") && strings.Contains(cond.Message, "failed") && strings.Contains(cond.Message, "deny") {
 //ignore co-scheduled pending pods for coscheduler version:0.22.6
 continue
+} else if strings.Contains(cond.Message, "optimistic") && strings.Contains(cond.Message, "rejection") && strings.Contains(cond.Message, "PostFilter") ||
+strings.Contains(cond.Message, "cannot") && strings.Contains(cond.Message, "find") && strings.Contains(cond.Message, "enough") && strings.Contains(cond.Message, "sibling") {
+//ignore co-scheduled pending pods for coscheduler version:0.23.10
+continue
 } else {
 podName := string(pods[i].Name)
 podCondition[podName] = append(podCondition[podName], *cond.DeepCopy())
@@ -74,41 +78,41 @@ func PendingPodsFailedSchd(pods []*v1.Pod) map[string][]v1.PodCondition {
 
 // filterPods returns pods based on their phase.
 func GetPodResourcesByPhase(phase v1.PodPhase, pods []*v1.Pod) *clusterstateapi.Resource {
- req := clusterstateapi.EmptyResource()
- for i := range pods {
- if  pods[i].Status.Phase == phase {
- for _, c := range pods[i].Spec.Containers {
- req.Add(clusterstateapi.NewResource(c.Resources.Requests))
- }
- }
- }
- return req
+req := clusterstateapi.EmptyResource()
+for i := range pods {
+if pods[i].Status.Phase == phase {
+for _, c := range pods[i].Spec.Containers {
+req.Add(clusterstateapi.NewResource(c.Resources.Requests))
+}
+}
+}
+return req
 }
 
 func GetPodResources(template *v1.PodTemplateSpec) *clusterstateapi.Resource {
- total := clusterstateapi.EmptyResource()
- req := clusterstateapi.EmptyResource()
- limit := clusterstateapi.EmptyResource()
- spec := template.Spec
+total := clusterstateapi.EmptyResource()
+req := clusterstateapi.EmptyResource()
+limit := clusterstateapi.EmptyResource()
+spec := template.Spec
 
- if &spec == nil {
- klog.Errorf("Pod Spec not found in Pod Template: %+v. Aggregated resources set to 0.", template)
- return total
- }
+if &spec == nil {
+klog.Errorf("Pod Spec not found in Pod Template: %+v. Aggregated resources set to 0.", template)
+return total
+}
 
- for _, c := range template.Spec.Containers {
- req.Add(clusterstateapi.NewResource(c.Resources.Requests))
- limit.Add(clusterstateapi.NewResource(c.Resources.Limits))
- }
- if req.MilliCPU < limit.MilliCPU {
- req.MilliCPU = limit.MilliCPU
- }
- if req.Memory < limit.Memory {
- req.Memory = limit.Memory
- }
- if req.GPU < limit.GPU {
- req.GPU = limit.GPU
- }
- total = total.Add(req)
- return total
+for _, c := range template.Spec.Containers {
+req.Add(clusterstateapi.NewResource(c.Resources.Requests))
+limit.Add(clusterstateapi.NewResource(c.Resources.Limits))
+}
+if req.MilliCPU < limit.MilliCPU {
+req.MilliCPU = limit.MilliCPU
+}
+if req.Memory < limit.Memory {
+req.Memory = limit.Memory
+}
+if req.GPU < limit.GPU {
+req.GPU = limit.GPU
+}
+total = total.Add(req)
+return total
 }