From f3cfdb0ef74883fcc18356db9a19ada665efd0b8 Mon Sep 17 00:00:00 2001 From: Piotr Szczesniak Date: Fri, 2 Oct 2015 12:08:22 +0200 Subject: [PATCH] Extended prediction algorithm in Initial Resource to use also namespace --- hack/verify-flags/known-flags.txt | 1 + .../admission/initialresources/admission.go | 70 ++++++++--- .../initialresources/admission_test.go | 119 ++++++++++++------ .../admission/initialresources/data_source.go | 2 +- plugin/pkg/admission/initialresources/gcm.go | 20 +-- .../admission/initialresources/hawkular.go | 2 +- .../admission/initialresources/influxdb.go | 12 +- 7 files changed, 153 insertions(+), 73 deletions(-) diff --git a/hack/verify-flags/known-flags.txt b/hack/verify-flags/known-flags.txt index 5458afe036..d6121fdfc1 100644 --- a/hack/verify-flags/known-flags.txt +++ b/hack/verify-flags/known-flags.txt @@ -122,6 +122,7 @@ iptables-sync-period ir-data-source ir-dbname ir-influxdb-host +ir-namespace-only ir-password ir-user jenkins-host diff --git a/plugin/pkg/admission/initialresources/admission.go b/plugin/pkg/admission/initialresources/admission.go index b151ed8bbd..98aeb8f337 100644 --- a/plugin/pkg/admission/initialresources/admission.go +++ b/plugin/pkg/admission/initialresources/admission.go @@ -32,8 +32,9 @@ import ( ) var ( - source = flag.String("ir-data-source", "influxdb", "Data source used by InitialResources. Supported options: influxdb.") + source = flag.String("ir-data-source", "influxdb", "Data source used by InitialResources. Supported options: influxdb, gcm.") percentile = flag.Int64("ir-percentile", 90, "Which percentile of samples should InitialResources use when estimating resources. For experiment purposes.") + nsOnly = flag.Bool("ir-namespace-only", false, "Whether the estimation should be made only based on data from the same namespace.") ) const ( @@ -50,19 +51,23 @@ func init() { if err != nil { return nil, err } - return newInitialResources(s), nil + return newInitialResources(s, *percentile, *nsOnly), nil }) } type initialResources struct { *admission.Handler - source dataSource + source dataSource + percentile int64 + nsOnly bool } -func newInitialResources(source dataSource) admission.Interface { +func newInitialResources(source dataSource, percentile int64, nsOnly bool) admission.Interface { return &initialResources{ - Handler: admission.NewHandler(admission.Create), - source: source, + Handler: admission.NewHandler(admission.Create), + source: source, + percentile: percentile, + nsOnly: nsOnly, } } @@ -92,7 +97,7 @@ func (ir initialResources) estimateAndFillResourcesIfNotSet(pod *api.Pod) { var err error if _, ok := req[api.ResourceCPU]; !ok { if _, ok2 := lim[api.ResourceCPU]; !ok2 { - cpu, err = ir.getEstimation(api.ResourceCPU, c) + cpu, err = ir.getEstimation(api.ResourceCPU, c, pod.ObjectMeta.Namespace) if err != nil { glog.Errorf("Error while trying to estimate resources: %v", err) } @@ -100,7 +105,7 @@ func (ir initialResources) estimateAndFillResourcesIfNotSet(pod *api.Pod) { } if _, ok := req[api.ResourceMemory]; !ok { if _, ok2 := lim[api.ResourceMemory]; !ok2 { - mem, err = ir.getEstimation(api.ResourceMemory, c) + mem, err = ir.getEstimation(api.ResourceMemory, c, pod.ObjectMeta.Namespace) if err != nil { glog.Errorf("Error while trying to estimate resources: %v", err) } @@ -138,29 +143,56 @@ func (ir initialResources) estimateAndFillResourcesIfNotSet(pod *api.Pod) { } } -func (ir initialResources) getEstimation(kind api.ResourceName, c *api.Container) (*resource.Quantity, error) { +func (ir initialResources) getEstimation(kind api.ResourceName, c *api.Container, ns string) (*resource.Quantity, error) { end := time.Now() start := end.Add(-week) var usage, samples int64 var err error - // Historical data from last 7 days for the same image:tag. - if usage, samples, err = ir.source.GetUsagePercentile(kind, *percentile, c.Image, true, start, end); err != nil { + // Historical data from last 7 days for the same image:tag within the same namespace. + if usage, samples, err = ir.source.GetUsagePercentile(kind, ir.percentile, c.Image, ns, true, start, end); err != nil { return nil, err } if samples < samplesThreshold { - // Historical data from last 30 days for the same image:tag. + // Historical data from last 30 days for the same image:tag within the same namespace. start := end.Add(-month) - if usage, samples, err = ir.source.GetUsagePercentile(kind, *percentile, c.Image, true, start, end); err != nil { + if usage, samples, err = ir.source.GetUsagePercentile(kind, ir.percentile, c.Image, ns, true, start, end); err != nil { return nil, err } } - if samples < samplesThreshold { - // Historical data from last 30 days for the same image. - start := end.Add(-month) - image := strings.Split(c.Image, ":")[0] - if usage, samples, err = ir.source.GetUsagePercentile(kind, *percentile, image, false, start, end); err != nil { - return nil, err + + // If we are allowed to estimate only based on data from the same namespace. + if ir.nsOnly { + if samples < samplesThreshold { + // Historical data from last 30 days for the same image within the same namespace. + start := end.Add(-month) + image := strings.Split(c.Image, ":")[0] + if usage, samples, err = ir.source.GetUsagePercentile(kind, ir.percentile, image, ns, false, start, end); err != nil { + return nil, err + } + } + } else { + if samples < samplesThreshold { + // Historical data from last 7 days for the same image:tag within all namespaces. + start := end.Add(-week) + if usage, samples, err = ir.source.GetUsagePercentile(kind, ir.percentile, c.Image, "", true, start, end); err != nil { + return nil, err + } + } + if samples < samplesThreshold { + // Historical data from last 30 days for the same image:tag within all namespaces. + start := end.Add(-month) + if usage, samples, err = ir.source.GetUsagePercentile(kind, ir.percentile, c.Image, "", true, start, end); err != nil { + return nil, err + } + } + if samples < samplesThreshold { + // Historical data from last 30 days for the same image within all namespaces. + start := end.Add(-month) + image := strings.Split(c.Image, ":")[0] + if usage, samples, err = ir.source.GetUsagePercentile(kind, ir.percentile, image, "", false, start, end); err != nil { + return nil, err + } } } diff --git a/plugin/pkg/admission/initialresources/admission_test.go b/plugin/pkg/admission/initialresources/admission_test.go index 40ec120cf4..8fe0ea01f6 100644 --- a/plugin/pkg/admission/initialresources/admission_test.go +++ b/plugin/pkg/admission/initialresources/admission_test.go @@ -26,11 +26,11 @@ import ( ) type fakeSource struct { - f func(kind api.ResourceName, perc int64, image string, exactMatch bool, start, end time.Time) (int64, int64, error) + f func(kind api.ResourceName, perc int64, image, namespace string, exactMatch bool, start, end time.Time) (int64, int64, error) } -func (s *fakeSource) GetUsagePercentile(kind api.ResourceName, perc int64, image string, exactMatch bool, start, end time.Time) (usage int64, samples int64, err error) { - return s.f(kind, perc, image, exactMatch, start, end) +func (s *fakeSource) GetUsagePercentile(kind api.ResourceName, perc int64, image, namespace string, exactMatch bool, start, end time.Time) (usage int64, samples int64, err error) { + return s.f(kind, perc, image, namespace, exactMatch, start, end) } func parseReq(cpu, mem string) api.ResourceList { @@ -57,7 +57,7 @@ func addContainer(pod *api.Pod, name, image string, request api.ResourceList) { func createPod(name string, image string, request api.ResourceList) *api.Pod { pod := &api.Pod{ - ObjectMeta: api.ObjectMeta{Name: name, Namespace: "test"}, + ObjectMeta: api.ObjectMeta{Name: name, Namespace: "test-ns"}, Spec: api.PodSpec{}, } pod.Spec.Containers = []api.Container{} @@ -113,15 +113,7 @@ func admit(t *testing.T, ir admission.Interface, pods []*api.Pod) { } } -func TestEstimationBasedOnTheSameImage7d(t *testing.T) { - f := func(_ api.ResourceName, _ int64, _ string, exactMatch bool, start, end time.Time) (int64, int64, error) { - if exactMatch && end.Sub(start) == week { - return 100, 120, nil - } - return 200, 120, nil - } - ir := newInitialResources(&fakeSource{f: f}) - +func performTest(t *testing.T, ir admission.Interface) { pods := getPods() admit(t, ir, pods) @@ -136,50 +128,74 @@ func TestEstimationBasedOnTheSameImage7d(t *testing.T) { expectNoAnnotation(t, pods[3]) } -func TestEstimationBasedOnTheSameImage30d(t *testing.T) { - f := func(_ api.ResourceName, _ int64, _ string, exactMatch bool, start, end time.Time) (int64, int64, error) { - if exactMatch && end.Sub(start) == week { - return 200, 20, nil - } - if exactMatch && end.Sub(start) == month { +func TestEstimationBasedOnTheSameImageSameNamespace7d(t *testing.T) { + f := func(_ api.ResourceName, _ int64, _, ns string, exactMatch bool, start, end time.Time) (int64, int64, error) { + if exactMatch && end.Sub(start) == week && ns == "test-ns" { return 100, 120, nil } return 200, 120, nil } - ir := newInitialResources(&fakeSource{f: f}) + performTest(t, newInitialResources(&fakeSource{f: f}, 90, false)) - pods := getPods() - admit(t, ir, pods) +} - verifyPod(t, pods[0], 100, 100) - verifyPod(t, pods[1], 100, 300) - verifyPod(t, pods[2], 300, 100) - verifyPod(t, pods[3], 300, 300) +func TestEstimationBasedOnTheSameImageSameNamespace30d(t *testing.T) { + f := func(_ api.ResourceName, _ int64, _, ns string, exactMatch bool, start, end time.Time) (int64, int64, error) { + if exactMatch && end.Sub(start) == week && ns == "test-ns" { + return 200, 20, nil + } + if exactMatch && end.Sub(start) == month && ns == "test-ns" { + return 100, 120, nil + } + return 200, 120, nil + } + performTest(t, newInitialResources(&fakeSource{f: f}, 90, false)) +} + +func TestEstimationBasedOnTheSameImageAllNamespaces7d(t *testing.T) { + f := func(_ api.ResourceName, _ int64, _, ns string, exactMatch bool, start, end time.Time) (int64, int64, error) { + if exactMatch && ns == "test-ns" { + return 200, 20, nil + } + if exactMatch && end.Sub(start) == week && ns == "" { + return 100, 120, nil + } + return 200, 120, nil + } + performTest(t, newInitialResources(&fakeSource{f: f}, 90, false)) +} + +func TestEstimationBasedOnTheSameImageAllNamespaces30d(t *testing.T) { + f := func(_ api.ResourceName, _ int64, _, ns string, exactMatch bool, start, end time.Time) (int64, int64, error) { + if exactMatch && ns == "test-ns" { + return 200, 20, nil + } + if exactMatch && end.Sub(start) == week && ns == "" { + return 200, 20, nil + } + if exactMatch && end.Sub(start) == month && ns == "" { + return 100, 120, nil + } + return 200, 120, nil + } + performTest(t, newInitialResources(&fakeSource{f: f}, 90, false)) } func TestEstimationBasedOnOtherImages(t *testing.T) { - f := func(_ api.ResourceName, _ int64, image string, exactMatch bool, _, _ time.Time) (int64, int64, error) { - if image == "image" && !exactMatch { + f := func(_ api.ResourceName, _ int64, image, ns string, exactMatch bool, _, _ time.Time) (int64, int64, error) { + if image == "image" && !exactMatch && ns == "" { return 100, 5, nil } return 200, 20, nil } - ir := newInitialResources(&fakeSource{f: f}) - - pods := getPods() - admit(t, ir, pods) - - verifyPod(t, pods[0], 100, 100) - verifyPod(t, pods[1], 100, 300) - verifyPod(t, pods[2], 300, 100) - verifyPod(t, pods[3], 300, 300) + performTest(t, newInitialResources(&fakeSource{f: f}, 90, false)) } func TestNoData(t *testing.T) { - f := func(_ api.ResourceName, _ int64, _ string, _ bool, _, _ time.Time) (int64, int64, error) { + f := func(_ api.ResourceName, _ int64, _, ns string, _ bool, _, _ time.Time) (int64, int64, error) { return 200, 0, nil } - ir := newInitialResources(&fakeSource{f: f}) + ir := newInitialResources(&fakeSource{f: f}, 90, false) pods := []*api.Pod{ createPod("p0", "image:v0", parseReq("", "")), @@ -194,13 +210,13 @@ func TestNoData(t *testing.T) { } func TestManyContainers(t *testing.T) { - f := func(_ api.ResourceName, _ int64, _ string, exactMatch bool, _, _ time.Time) (int64, int64, error) { + f := func(_ api.ResourceName, _ int64, _, ns string, exactMatch bool, _, _ time.Time) (int64, int64, error) { if exactMatch { return 100, 120, nil } return 200, 30, nil } - ir := newInitialResources(&fakeSource{f: f}) + ir := newInitialResources(&fakeSource{f: f}, 90, false) pod := createPod("p", "image:v0", parseReq("", "")) addContainer(pod, "c1", "image:v1", parseReq("", "300")) @@ -215,3 +231,24 @@ func TestManyContainers(t *testing.T) { verifyAnnotation(t, pod, "Initial Resources plugin set: cpu, memory request for container c0; cpu request for container c1; memory request for container c2") } + +func TestNamespaceAware(t *testing.T) { + f := func(_ api.ResourceName, _ int64, _, ns string, exactMatch bool, start, end time.Time) (int64, int64, error) { + if ns == "test-ns" { + return 200, 0, nil + } + return 200, 120, nil + } + ir := newInitialResources(&fakeSource{f: f}, 90, true) + + pods := []*api.Pod{ + createPod("p0", "image:v0", parseReq("", "")), + } + admit(t, ir, pods) + + if pods[0].Spec.Containers[0].Resources.Requests != nil { + t.Errorf("Unexpected resource estimation") + } + + expectNoAnnotation(t, pods[0]) +} diff --git a/plugin/pkg/admission/initialresources/data_source.go b/plugin/pkg/admission/initialresources/data_source.go index d6cb43bf33..c391c007f7 100644 --- a/plugin/pkg/admission/initialresources/data_source.go +++ b/plugin/pkg/admission/initialresources/data_source.go @@ -39,7 +39,7 @@ type dataSource interface { // withing time range (start, end), number of samples considered and error if occured. // If then take only samples that concern the same image (both name and take are the same), // otherwise consider also samples with the same image a possibly different tag. - GetUsagePercentile(kind api.ResourceName, perc int64, image string, exactMatch bool, start, end time.Time) (usage int64, samples int64, err error) + GetUsagePercentile(kind api.ResourceName, perc int64, image, namespace string, exactMatch bool, start, end time.Time) (usage int64, samples int64, err error) } func newDataSource(kind string) (dataSource, error) { diff --git a/plugin/pkg/admission/initialresources/gcm.go b/plugin/pkg/admission/initialresources/gcm.go index be6fae9347..e78556df93 100644 --- a/plugin/pkg/admission/initialresources/gcm.go +++ b/plugin/pkg/admission/initialresources/gcm.go @@ -34,6 +34,7 @@ const ( cpuMetricName = kubePrefix + "cpu/usage_rate" memMetricName = kubePrefix + "memory/usage" labelImage = kubePrefix + "label/container_base_image" + labelNs = kubePrefix + "label/pod_namespace" ) type gcmSource struct { @@ -61,12 +62,14 @@ func newGcmSource() (dataSource, error) { }, nil } -func (s *gcmSource) query(metric, oldest, youngest, label, pageToken string) (*gcm.ListTimeseriesResponse, error) { +func (s *gcmSource) query(metric, oldest, youngest string, labels []string, pageToken string) (*gcm.ListTimeseriesResponse, error) { req := s.gcmService.Timeseries.List(s.project, metric, youngest, nil). Oldest(oldest). - Labels(label). Aggregator("mean"). Window("1m") + for _, l := range labels { + req = req.Labels(l) + } if pageToken != "" { req = req.PageToken(pageToken) } @@ -81,7 +84,7 @@ func retrieveRawSamples(res *gcm.ListTimeseriesResponse, output *[]int) { } } -func (s *gcmSource) GetUsagePercentile(kind api.ResourceName, perc int64, image string, exactMatch bool, start, end time.Time) (int64, int64, error) { +func (s *gcmSource) GetUsagePercentile(kind api.ResourceName, perc int64, image, namespace string, exactMatch bool, start, end time.Time) (int64, int64, error) { var metric string if kind == api.ResourceCPU { metric = cpuMetricName @@ -89,11 +92,14 @@ func (s *gcmSource) GetUsagePercentile(kind api.ResourceName, perc int64, image metric = memMetricName } - var label string + var labels []string if exactMatch { - label = labelImage + "==" + image + labels = append(labels, labelImage+"=="+image) } else { - label = labelImage + "=~" + image + ".*" + labels = append(labels, labelImage+"=~"+image+".*") + } + if namespace != "" { + labels = append(labels, labelNs+"=="+namespace) } oldest := start.Format(time.RFC3339) @@ -102,7 +108,7 @@ func (s *gcmSource) GetUsagePercentile(kind api.ResourceName, perc int64, image rawSamples := make([]int, 0) pageToken := "" for { - res, err := s.query(metric, oldest, youngest, label, pageToken) + res, err := s.query(metric, oldest, youngest, labels, pageToken) if err != nil { return 0, 0, err } diff --git a/plugin/pkg/admission/initialresources/hawkular.go b/plugin/pkg/admission/initialresources/hawkular.go index ea30100457..1482b0d75c 100644 --- a/plugin/pkg/admission/initialresources/hawkular.go +++ b/plugin/pkg/admission/initialresources/hawkular.go @@ -29,6 +29,6 @@ func newHawkularSource() (dataSource, error) { return nil, fmt.Errorf("hawkular source not implemented") } -func (s *hawkularSource) GetUsagePercentile(kind api.ResourceName, perc int64, image string, exactMatch bool, start, end time.Time) (int64, int64, error) { +func (s *hawkularSource) GetUsagePercentile(kind api.ResourceName, perc int64, image, namespace string, exactMatch bool, start, end time.Time) (int64, int64, error) { return 0, 0, fmt.Errorf("gcm source not implemented") } diff --git a/plugin/pkg/admission/initialresources/influxdb.go b/plugin/pkg/admission/initialresources/influxdb.go index 876df7d641..1664afbb7d 100644 --- a/plugin/pkg/admission/initialresources/influxdb.go +++ b/plugin/pkg/admission/initialresources/influxdb.go @@ -29,8 +29,8 @@ import ( const ( cpuSeriesName = "autoscaling.cpu.usage.1m" memSeriesName = "autoscaling.memory.usage.1m" - cpuContinuousQuery = "select derivative(value) as value from \"cpu/usage_ns_cumulative\" where pod_id <> '' group by pod_id, container_name, container_base_image, time(1m) into " + cpuSeriesName - memContinuousQuery = "select mean(value) as value from \"memory/usage_bytes_gauge\" where pod_id <> '' group by pod_id, container_name, container_base_image, time(1m) into " + memSeriesName + cpuContinuousQuery = "select derivative(value) as value from \"cpu/usage_ns_cumulative\" where pod_id <> '' group by pod_id, pod_namespace, container_name, container_base_image, time(1m) into " + cpuSeriesName + memContinuousQuery = "select mean(value) as value from \"memory/usage_bytes_gauge\" where pod_id <> '' group by pod_id, pod_namespace, container_name, container_base_image, time(1m) into " + memSeriesName timeFormat = "2006-01-02 15:04:05" ) @@ -112,7 +112,7 @@ func (s *influxdbSource) query(query string, precision ...influxdb.TimePrecision return client.Query(query, precision...) } -func (s *influxdbSource) GetUsagePercentile(kind api.ResourceName, perc int64, image string, exactMatch bool, start, end time.Time) (int64, int64, error) { +func (s *influxdbSource) GetUsagePercentile(kind api.ResourceName, perc int64, image, namespace string, exactMatch bool, start, end time.Time) (int64, int64, error) { var series string if kind == api.ResourceCPU { series = cpuSeriesName @@ -126,8 +126,12 @@ func (s *influxdbSource) GetUsagePercentile(kind api.ResourceName, perc int64, i } else { imgPattern = "=~/^" + image + "/" } + var namespaceCond string + if namespace != "" { + namespaceCond = " and pod_namespace='" + namespace + "'" + } - query := fmt.Sprintf("select percentile(value, %v), count(pod_id) from %v where container_base_image%v and time > '%v' and time < '%v'", perc, series, imgPattern, start.UTC().Format(timeFormat), end.UTC().Format(timeFormat)) + query := fmt.Sprintf("select percentile(value, %v), count(pod_id) from %v where container_base_image%v%v and time > '%v' and time < '%v'", perc, series, imgPattern, namespaceCond, start.UTC().Format(timeFormat), end.UTC().Format(timeFormat)) var res []*influxdb.Series var err error if res, err = s.query(query, influxdb.Second); err != nil {