Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Reduce spam in Events from scheduler by counter aggregation of failure #35932

Merged
merged 1 commit into from
Nov 7, 2016
Merged

Reduce spam in Events from scheduler by counter aggregation of failure #35932

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Jump to
Jump to file
Failed to load files.
Diff view
Diff view
20 changes: 15 additions & 5 deletions plugin/pkg/scheduler/generic_scheduler.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -49,14 +49,24 @@ var ErrNoNodesAvailable = fmt.Errorf("no nodes available to schedule pods")
func (f *FitError) Error() string {
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is only called at the end right?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

yup. ScheduleOne is the call that will grab the evnts and push em to etcd.

var buf bytes.Buffer
buf.WriteString(fmt.Sprintf("pod (%s) failed to fit in any node\n", f.Pod.Name))
for node, predicates := range f.FailedPredicates {
reasons := make([]string, 0)
reasons := make(map[string]int)
for _, predicates := range f.FailedPredicates {
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We're still storing internally (N), it's just the outside summary will now be a bin'd histogram.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

event summation will be rollup, store O(unique failure types)

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yeah I get that, just denoting that internally we are storing f.FailedPredicates which will still be (N).

Copy link
Member Author

@jayunit100 jayunit100 Nov 2, 2016 •
edited

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

yup, the parellelization factor i think will determine the total memory consumed by failed nodes

for _, pred := range predicates {
reasons = append(reasons, pred.GetReason())
reasons[pred.GetReason()] += 1
}
reasonMsg := fmt.Sprintf("fit failure on node (%s): %s\n", node, strings.Join(reasons, ", "))
buf.WriteString(reasonMsg)
}

sortReasonsHistogram := func() []string {
reasonStrings := []string{}
for k, v := range reasons {
reasonStrings = append(reasonStrings, fmt.Sprintf("%v (%v)", k, v))
}
sort.Strings(reasonStrings)
return reasonStrings
}

reasonMsg := fmt.Sprintf("fit failure summary on nodes : %v", strings.Join(sortReasonsHistogram(), ", "))
buf.WriteString(reasonMsg)
return buf.String()
}

Expand Down
76 changes: 47 additions & 29 deletions plugin/pkg/scheduler/scheduler_test.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,6 +18,7 @@ package scheduler

import (
"errors"
"fmt"
"reflect"
"testing"
"time"
Expand Down Expand Up @@ -331,49 +332,66 @@ func TestSchedulerFailedSchedulingReasons(t *testing.T) {
defer close(stop)
queuedPodStore := clientcache.NewFIFO(clientcache.MetaNamespaceKeyFunc)
scache := schedulercache.New(10*time.Minute, stop)
node := api.Node{
ObjectMeta: api.ObjectMeta{Name: "machine1"},
Status: api.NodeStatus{
Capacity: api.ResourceList{
api.ResourceCPU: *(resource.NewQuantity(2, resource.DecimalSI)),
api.ResourceMemory: *(resource.NewQuantity(100, resource.DecimalSI)),
api.ResourcePods: *(resource.NewQuantity(10, resource.DecimalSI)),
},
Allocatable: api.ResourceList{
api.ResourceCPU: *(resource.NewQuantity(2, resource.DecimalSI)),
api.ResourceMemory: *(resource.NewQuantity(100, resource.DecimalSI)),
api.ResourcePods: *(resource.NewQuantity(10, resource.DecimalSI)),
}},

// Design the baseline for the pods, and we will make nodes that dont fit it later.
var cpu = int64(4)
var mem = int64(500)
podWithTooBigResourceRequests := podWithResources("bar", "", api.ResourceList{
api.ResourceCPU: *(resource.NewQuantity(cpu, resource.DecimalSI)),
api.ResourceMemory: *(resource.NewQuantity(mem, resource.DecimalSI)),
}, api.ResourceList{
api.ResourceCPU: *(resource.NewQuantity(cpu, resource.DecimalSI)),
api.ResourceMemory: *(resource.NewQuantity(mem, resource.DecimalSI)),
})

// create several nodes which cannot schedule the above pod
nodes := []*api.Node{}
for i := 0; i < 100; i++ {
node := api.Node{
ObjectMeta: api.ObjectMeta{Name: fmt.Sprintf("machine%v", i)},
Status: api.NodeStatus{
Capacity: api.ResourceList{
api.ResourceCPU: *(resource.NewQuantity(cpu/2, resource.DecimalSI)),
api.ResourceMemory: *(resource.NewQuantity(mem/5, resource.DecimalSI)),
api.ResourcePods: *(resource.NewQuantity(10, resource.DecimalSI)),
},
Allocatable: api.ResourceList{
api.ResourceCPU: *(resource.NewQuantity(cpu/2, resource.DecimalSI)),
api.ResourceMemory: *(resource.NewQuantity(mem/5, resource.DecimalSI)),
api.ResourcePods: *(resource.NewQuantity(10, resource.DecimalSI)),
}},
}
scache.AddNode(&node)
nodes = append(nodes, &node)
}
scache.AddNode(&node)
nodeLister := algorithm.FakeNodeLister([]*api.Node{&node})
nodeLister := algorithm.FakeNodeLister(nodes)
predicateMap := map[string]algorithm.FitPredicate{
"PodFitsResources": predicates.PodFitsResources,
}

// Create expected failure reasons for all the nodes. Hopefully they will get rolled up into a non-spammy summary.
failedPredicatesMap := FailedPredicateMap{}
for _, node := range nodes {
failedPredicatesMap[node.Name] = []algorithm.PredicateFailureReason{
predicates.NewInsufficientResourceError(api.ResourceCPU, 4000, 0, 2000),
predicates.NewInsufficientResourceError(api.ResourceMemory, 500, 0, 100),
}
}
scheduler, _, errChan := setupTestScheduler(queuedPodStore, scache, nodeLister, predicateMap)

podWithTooBigResourceRequests := podWithResources("bar", "", api.ResourceList{
api.ResourceCPU: *(resource.NewQuantity(4, resource.DecimalSI)),
api.ResourceMemory: *(resource.NewQuantity(500, resource.DecimalSI)),
}, api.ResourceList{
api.ResourceCPU: *(resource.NewQuantity(4, resource.DecimalSI)),
api.ResourceMemory: *(resource.NewQuantity(500, resource.DecimalSI)),
})
queuedPodStore.Add(podWithTooBigResourceRequests)
scheduler.scheduleOne()

select {
case err := <-errChan:
expectErr := &FitError{
Pod: podWithTooBigResourceRequests,
FailedPredicates: FailedPredicateMap{node.Name: []algorithm.PredicateFailureReason{
predicates.NewInsufficientResourceError(api.ResourceCPU, 4000, 0, 2000),
predicates.NewInsufficientResourceError(api.ResourceMemory, 500, 0, 100),
}},
Pod: podWithTooBigResourceRequests,
FailedPredicates: failedPredicatesMap,
}
if len(fmt.Sprint(expectErr)) > 150 {
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Shouldn't we could lines vs. len()?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

hmmm, yup considered that to, but this is more precise, missing \n-> miss large non delimited values. ?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

+0 either way is ok. i think \n is more coupled ....

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I only ask b/c 150 is arbitrary vs. looking for > 1 line.. which is what the rollup should produce.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

thats true, 150 seemed a little magic to me, 1 line more precise. Maybe we should say 32K max string size or something?

t.Errorf("message is too spammy ! %v ", len(fmt.Sprint(expectErr)))
}
if !reflect.DeepEqual(expectErr, err) {
t.Errorf("err want=%+v, get=%+v", expectErr, err)
t.Errorf("\n err \nWANT=%+v,\nGOT=%+v", expectErr, err)
}
case <-time.After(wait.ForeverTestTimeout):
t.Fatalf("timeout after %v", wait.ForeverTestTimeout)
Expand Down