
When the pods are transitioning there are several cases where containers might not be in valid state. There were several cases where the stats where failing hard but we should just continue on as they are transient and will be picked up again when kubelet queries for the stats again. Signed-off-by: James Sturtevant <jstur@microsoft.com> Signed-off-by: Mark Rossetti <marosset@microsoft.com>
437 lines
15 KiB
Go
437 lines
15 KiB
Go
/*
|
|
Copyright The containerd Authors.
|
|
|
|
Licensed under the Apache License, Version 2.0 (the "License");
|
|
you may not use this file except in compliance with the License.
|
|
You may obtain a copy of the License at
|
|
|
|
http://www.apache.org/licenses/LICENSE-2.0
|
|
|
|
Unless required by applicable law or agreed to in writing, software
|
|
distributed under the License is distributed on an "AS IS" BASIS,
|
|
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
See the License for the specific language governing permissions and
|
|
limitations under the License.
|
|
*/
|
|
|
|
package server
|
|
|
|
import (
|
|
"context"
|
|
"fmt"
|
|
"time"
|
|
|
|
"github.com/Microsoft/hcsshim"
|
|
wstats "github.com/Microsoft/hcsshim/cmd/containerd-shim-runhcs-v1/stats"
|
|
"github.com/Microsoft/hcsshim/hcn"
|
|
"github.com/containerd/containerd/api/services/tasks/v1"
|
|
"github.com/containerd/containerd/api/types"
|
|
"github.com/containerd/containerd/errdefs"
|
|
"github.com/containerd/containerd/log"
|
|
containerstore "github.com/containerd/containerd/pkg/cri/store/container"
|
|
sandboxstore "github.com/containerd/containerd/pkg/cri/store/sandbox"
|
|
"github.com/containerd/containerd/pkg/cri/store/stats"
|
|
"github.com/containerd/typeurl/v2"
|
|
runtime "k8s.io/cri-api/pkg/apis/runtime/v1"
|
|
)
|
|
|
|
func (c *criService) podSandboxStats(
|
|
ctx context.Context,
|
|
sandbox sandboxstore.Sandbox) (*runtime.PodSandboxStats, error) {
|
|
meta := sandbox.Metadata
|
|
|
|
if sandbox.Status.Get().State != sandboxstore.StateReady {
|
|
return nil, fmt.Errorf("failed to get pod sandbox stats since sandbox container %q is not in ready state: %w", meta.ID, errdefs.ErrUnavailable)
|
|
}
|
|
|
|
timestamp := time.Now()
|
|
podSandboxStats := &runtime.PodSandboxStats{
|
|
Windows: &runtime.WindowsPodSandboxStats{},
|
|
Attributes: &runtime.PodSandboxAttributes{
|
|
Id: meta.ID,
|
|
Metadata: meta.Config.GetMetadata(),
|
|
Labels: meta.Config.GetLabels(),
|
|
Annotations: meta.Config.GetAnnotations(),
|
|
},
|
|
}
|
|
|
|
metrics, containers, err := c.listWindowsMetricsForSandbox(ctx, sandbox)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("failed to obtain container stats during podSandboxStats call: %w", err)
|
|
}
|
|
|
|
statsMap, err := convertMetricsToWindowsStats(metrics, sandbox)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("failed to convert stats: %w", err)
|
|
}
|
|
|
|
podCPU, containerStats, err := c.toPodSandboxStats(sandbox, statsMap, containers, timestamp)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("failed to convert container stats during podSandboxStats call: %w", err)
|
|
}
|
|
podSandboxStats.Windows.Cpu = podCPU.Cpu
|
|
podSandboxStats.Windows.Memory = podCPU.Memory
|
|
podSandboxStats.Windows.Containers = containerStats
|
|
podSandboxStats.Windows.Network = windowsNetworkUsage(ctx, sandbox, timestamp)
|
|
|
|
pidCount, err := c.getSandboxPidCount(ctx, sandbox)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("failed to get pid count: %w", err)
|
|
}
|
|
|
|
podSandboxStats.Windows.Process = &runtime.WindowsProcessUsage{
|
|
Timestamp: timestamp.UnixNano(),
|
|
ProcessCount: &runtime.UInt64Value{Value: pidCount},
|
|
}
|
|
|
|
c.saveSandBoxMetrics(podSandboxStats.Attributes.Id, podSandboxStats)
|
|
|
|
return podSandboxStats, nil
|
|
}
|
|
|
|
func convertMetricsToWindowsStats(metrics []*types.Metric, sandbox sandboxstore.Sandbox) (map[string]*wstats.Statistics, error) {
|
|
isHostProcess := sandbox.Config.GetWindows().GetSecurityContext().GetHostProcess()
|
|
|
|
statsMap := make(map[string]*wstats.Statistics)
|
|
for _, stat := range metrics {
|
|
containerStatsData, err := typeurl.UnmarshalAny(stat.Data)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("failed to extract metrics for container with id %s: %w", stat.ID, err)
|
|
}
|
|
|
|
// extract the metrics if available for this container
|
|
// containerStatsData can be nil for pods that don't have an actual podsandbox container such as HPC
|
|
// In the case of HostProcess sandbox container we will use the nil value for the statsmap which is used later
|
|
// otherwise return an error since we should have gotten stats
|
|
containerStats, ok := containerStatsData.(*wstats.Statistics)
|
|
if !ok && !(isHostProcess && sandbox.ID == stat.ID) {
|
|
return nil, fmt.Errorf("failed to extract metrics for container with id %s: %w", stat.ID, err)
|
|
}
|
|
|
|
statsMap[stat.ID] = containerStats
|
|
}
|
|
return statsMap, nil
|
|
}
|
|
|
|
func (c *criService) toPodSandboxStats(sandbox sandboxstore.Sandbox, statsMap map[string]*wstats.Statistics, containers []containerstore.Container, timestamp time.Time) (*runtime.WindowsContainerStats, []*runtime.WindowsContainerStats, error) {
|
|
podMetric, ok := statsMap[sandbox.ID]
|
|
if !ok {
|
|
return nil, nil, fmt.Errorf("failed to find container metric for pod with id %s", sandbox.ID)
|
|
}
|
|
|
|
podRuntimeStats, err := c.convertToCRIStats(podMetric)
|
|
if err != nil {
|
|
return nil, nil, fmt.Errorf("failed to covert container metrics for sandbox with id %s: %w", sandbox.ID, err)
|
|
}
|
|
|
|
windowsContainerStats := make([]*runtime.WindowsContainerStats, 0, len(statsMap))
|
|
for _, cntr := range containers {
|
|
containerMetric := statsMap[cntr.ID]
|
|
|
|
if cntr.Status.Get().State() != runtime.ContainerState_CONTAINER_RUNNING {
|
|
// containers that are just created, in a failed state or exited (init containers) will not have stats
|
|
log.L.Warnf("failed to get container stats since container %q is not in running state", cntr.ID)
|
|
continue
|
|
}
|
|
|
|
if containerMetric == nil {
|
|
log.L.Warnf("no metrics found for container %q", cntr.ID)
|
|
continue
|
|
}
|
|
|
|
containerStats, err := c.convertToCRIStats(containerMetric)
|
|
if err != nil {
|
|
return nil, nil, fmt.Errorf("failed to convert metrics for container with id %s: %w", cntr.ID, err)
|
|
}
|
|
|
|
// Calculate NanoCores for container
|
|
if containerStats.Cpu != nil && containerStats.Cpu.UsageCoreNanoSeconds != nil {
|
|
nanoCoreUsage := getUsageNanoCores(containerStats.Cpu.UsageCoreNanoSeconds.Value, cntr.Stats, containerStats.Cpu.Timestamp)
|
|
containerStats.Cpu.UsageNanoCores = &runtime.UInt64Value{Value: nanoCoreUsage}
|
|
}
|
|
|
|
// On Windows we need to add up all the podStatsData to get the Total for the Pod as there isn't something
|
|
// like a parent cgroup that queried for all the pod podStatsData
|
|
appendCPUPodStats(podRuntimeStats, containerStats, timestamp)
|
|
appendMemoryPodStats(podRuntimeStats, containerStats, timestamp)
|
|
|
|
// If snapshotstore doesn't have cached snapshot information
|
|
// set WritableLayer usage to zero
|
|
var usedBytes uint64
|
|
sn, err := c.snapshotStore.Get(cntr.ID)
|
|
if err == nil {
|
|
usedBytes = sn.Size
|
|
}
|
|
containerStats.WritableLayer = &runtime.WindowsFilesystemUsage{
|
|
Timestamp: sn.Timestamp,
|
|
FsId: &runtime.FilesystemIdentifier{
|
|
Mountpoint: c.imageFSPath,
|
|
},
|
|
UsedBytes: &runtime.UInt64Value{Value: usedBytes},
|
|
}
|
|
|
|
containerStats.Attributes = &runtime.ContainerAttributes{
|
|
Id: cntr.ID,
|
|
Metadata: cntr.Config.GetMetadata(),
|
|
Labels: cntr.Config.GetLabels(),
|
|
Annotations: cntr.Config.GetAnnotations(),
|
|
}
|
|
|
|
windowsContainerStats = append(windowsContainerStats, containerStats)
|
|
}
|
|
|
|
// Calculate NanoCores for pod after adding containers cpu including the pods cpu
|
|
if podRuntimeStats.Cpu != nil && podRuntimeStats.Cpu.UsageCoreNanoSeconds != nil {
|
|
nanoCoreUsage := getUsageNanoCores(podRuntimeStats.Cpu.UsageCoreNanoSeconds.Value, sandbox.Stats, podRuntimeStats.Cpu.Timestamp)
|
|
podRuntimeStats.Cpu.UsageNanoCores = &runtime.UInt64Value{Value: nanoCoreUsage}
|
|
}
|
|
|
|
return podRuntimeStats, windowsContainerStats, nil
|
|
}
|
|
|
|
func appendCPUPodStats(podRuntimeStats *runtime.WindowsContainerStats, containerRunTimeStats *runtime.WindowsContainerStats, timestamp time.Time) {
|
|
// protect against missing stats in case container hasn't started yet
|
|
if containerRunTimeStats.Cpu == nil || containerRunTimeStats.Cpu.UsageCoreNanoSeconds == nil {
|
|
return
|
|
}
|
|
|
|
// It is possible the pod sandbox might not be populated with values if it doesn't exist
|
|
// HostProcess pods are an example where there is no actual pod sandbox running and therefor no stats
|
|
if podRuntimeStats.Cpu == nil {
|
|
podRuntimeStats.Cpu = &runtime.WindowsCpuUsage{
|
|
Timestamp: timestamp.UnixNano(),
|
|
UsageCoreNanoSeconds: &runtime.UInt64Value{Value: 0},
|
|
}
|
|
}
|
|
|
|
if podRuntimeStats.Cpu.UsageCoreNanoSeconds == nil {
|
|
podRuntimeStats.Cpu.UsageCoreNanoSeconds = &runtime.UInt64Value{Value: 0}
|
|
}
|
|
|
|
podRuntimeStats.Cpu.UsageCoreNanoSeconds.Value += containerRunTimeStats.Cpu.UsageCoreNanoSeconds.Value
|
|
}
|
|
|
|
func appendMemoryPodStats(podRuntimeStats *runtime.WindowsContainerStats, containerRunTimeStats *runtime.WindowsContainerStats, timestamp time.Time) {
|
|
// protect against missing stats in case container hasn't started yet
|
|
if containerRunTimeStats.Memory == nil {
|
|
return
|
|
}
|
|
|
|
// It is possible the pod sandbox might not be populated with values if it doesn't exist
|
|
// HostProcess pods are an example where there is no actual pod sandbox running and therefor no stats
|
|
if podRuntimeStats.Memory == nil {
|
|
podRuntimeStats.Memory = &runtime.WindowsMemoryUsage{
|
|
Timestamp: timestamp.UnixNano(),
|
|
WorkingSetBytes: &runtime.UInt64Value{Value: 0},
|
|
AvailableBytes: &runtime.UInt64Value{Value: 0},
|
|
PageFaults: &runtime.UInt64Value{Value: 0},
|
|
}
|
|
}
|
|
|
|
if containerRunTimeStats.Memory.WorkingSetBytes != nil {
|
|
if podRuntimeStats.Memory.WorkingSetBytes == nil {
|
|
podRuntimeStats.Memory.WorkingSetBytes = &runtime.UInt64Value{Value: 0}
|
|
}
|
|
podRuntimeStats.Memory.WorkingSetBytes.Value += containerRunTimeStats.Memory.WorkingSetBytes.Value
|
|
}
|
|
|
|
if containerRunTimeStats.Memory.AvailableBytes != nil {
|
|
if podRuntimeStats.Memory.AvailableBytes == nil {
|
|
podRuntimeStats.Memory.AvailableBytes = &runtime.UInt64Value{Value: 0}
|
|
}
|
|
podRuntimeStats.Memory.AvailableBytes.Value += containerRunTimeStats.Memory.AvailableBytes.Value
|
|
}
|
|
|
|
if containerRunTimeStats.Memory.PageFaults != nil {
|
|
if podRuntimeStats.Memory.PageFaults == nil {
|
|
podRuntimeStats.Memory.PageFaults = &runtime.UInt64Value{Value: 0}
|
|
}
|
|
podRuntimeStats.Memory.PageFaults.Value += containerRunTimeStats.Memory.PageFaults.Value
|
|
}
|
|
}
|
|
|
|
func (c *criService) listWindowsMetricsForSandbox(ctx context.Context, sandbox sandboxstore.Sandbox) ([]*types.Metric, []containerstore.Container, error) {
|
|
req := &tasks.MetricsRequest{}
|
|
var containers []containerstore.Container
|
|
for _, cntr := range c.containerStore.List() {
|
|
if cntr.SandboxID != sandbox.ID {
|
|
continue
|
|
}
|
|
containers = append(containers, cntr)
|
|
req.Filters = append(req.Filters, "id=="+cntr.ID)
|
|
}
|
|
|
|
//add sandbox container as well
|
|
req.Filters = append(req.Filters, "id=="+sandbox.ID)
|
|
|
|
resp, err := c.client.TaskService().Metrics(ctx, req)
|
|
if err != nil {
|
|
return nil, nil, fmt.Errorf("failed to fetch metrics for tasks: %w", err)
|
|
}
|
|
return resp.Metrics, containers, nil
|
|
}
|
|
|
|
func (c *criService) convertToCRIStats(stats *wstats.Statistics) (*runtime.WindowsContainerStats, error) {
|
|
var cs runtime.WindowsContainerStats
|
|
// the metric should exist but stats or stats.container will be nil for HostProcess sandbox containers
|
|
// this can also be the case when the container has not started yet
|
|
if stats != nil && stats.Container != nil {
|
|
wstats := stats.GetWindows()
|
|
if wstats == nil {
|
|
return nil, fmt.Errorf("windows stats is empty")
|
|
}
|
|
if wstats.Processor != nil {
|
|
cs.Cpu = &runtime.WindowsCpuUsage{
|
|
Timestamp: wstats.Timestamp.UnixNano(),
|
|
UsageCoreNanoSeconds: &runtime.UInt64Value{Value: wstats.Processor.TotalRuntimeNS},
|
|
}
|
|
}
|
|
|
|
if wstats.Memory != nil {
|
|
cs.Memory = &runtime.WindowsMemoryUsage{
|
|
Timestamp: wstats.Timestamp.UnixNano(),
|
|
WorkingSetBytes: &runtime.UInt64Value{
|
|
Value: wstats.Memory.MemoryUsagePrivateWorkingSetBytes,
|
|
},
|
|
}
|
|
}
|
|
|
|
}
|
|
return &cs, nil
|
|
}
|
|
|
|
func getUsageNanoCores(usageCoreNanoSeconds uint64, oldStats *stats.ContainerStats, newtimestamp int64) uint64 {
|
|
if oldStats == nil {
|
|
return 0
|
|
}
|
|
|
|
nanoSeconds := newtimestamp - oldStats.Timestamp.UnixNano()
|
|
|
|
// zero or negative interval
|
|
if nanoSeconds <= 0 {
|
|
return 0
|
|
}
|
|
|
|
return uint64(float64(usageCoreNanoSeconds-oldStats.UsageCoreNanoSeconds) /
|
|
float64(nanoSeconds) * float64(time.Second/time.Nanosecond))
|
|
}
|
|
|
|
func windowsNetworkUsage(ctx context.Context, sandbox sandboxstore.Sandbox, timestamp time.Time) *runtime.WindowsNetworkUsage {
|
|
eps, err := hcn.GetNamespaceEndpointIds(sandbox.NetNSPath)
|
|
if err != nil {
|
|
log.G(ctx).WithField("podsandboxid", sandbox.ID).WithError(err).Errorf("unable to retrieve windows endpoint metrics for netNsPath: %v", sandbox.NetNSPath)
|
|
return nil
|
|
}
|
|
networkUsage := &runtime.WindowsNetworkUsage{
|
|
Timestamp: timestamp.UnixNano(),
|
|
}
|
|
for _, ep := range eps {
|
|
endpointStats, err := hcsshim.GetHNSEndpointStats(ep)
|
|
if err != nil {
|
|
log.G(ctx).WithError(err).Errorf("unable to gather stats for endpoint: %s", ep)
|
|
continue
|
|
}
|
|
rtStats := runtime.WindowsNetworkInterfaceUsage{
|
|
Name: endpointStats.EndpointID,
|
|
RxBytes: &runtime.UInt64Value{Value: endpointStats.BytesReceived},
|
|
RxPacketsDropped: &runtime.UInt64Value{Value: endpointStats.DroppedPacketsIncoming},
|
|
TxBytes: &runtime.UInt64Value{Value: endpointStats.BytesSent},
|
|
TxPacketsDropped: &runtime.UInt64Value{Value: endpointStats.DroppedPacketsOutgoing},
|
|
}
|
|
networkUsage.Interfaces = append(networkUsage.Interfaces, &rtStats)
|
|
|
|
// if the default interface isn't set add it.
|
|
// We don't have a way to determine the default interface in windows
|
|
if networkUsage.DefaultInterface == nil {
|
|
networkUsage.DefaultInterface = &rtStats
|
|
}
|
|
}
|
|
|
|
return networkUsage
|
|
}
|
|
|
|
func (c *criService) saveSandBoxMetrics(sandboxID string, sandboxStats *runtime.PodSandboxStats) error {
|
|
// we may not have stats since container hasn't started yet so skip saving to cache
|
|
if sandboxStats == nil || sandboxStats.Windows == nil || sandboxStats.Windows.Cpu == nil ||
|
|
sandboxStats.Windows.Cpu.UsageCoreNanoSeconds == nil {
|
|
return nil
|
|
}
|
|
|
|
newStats := &stats.ContainerStats{
|
|
UsageCoreNanoSeconds: sandboxStats.Windows.Cpu.UsageCoreNanoSeconds.Value,
|
|
Timestamp: time.Unix(0, sandboxStats.Windows.Cpu.Timestamp),
|
|
}
|
|
err := c.sandboxStore.UpdateContainerStats(sandboxID, newStats)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
// We queried the stats when getting sandbox stats. We need to save the query to cache
|
|
for _, cntr := range sandboxStats.Windows.Containers {
|
|
// we may not have stats since container hasn't started yet so skip saving to cache
|
|
if cntr == nil || cntr.Cpu == nil || cntr.Cpu.UsageCoreNanoSeconds == nil {
|
|
return nil
|
|
}
|
|
|
|
newStats := &stats.ContainerStats{
|
|
UsageCoreNanoSeconds: cntr.Cpu.UsageCoreNanoSeconds.Value,
|
|
Timestamp: time.Unix(0, cntr.Cpu.Timestamp),
|
|
}
|
|
err = c.containerStore.UpdateContainerStats(cntr.Attributes.Id, newStats)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
func (c *criService) getSandboxPidCount(ctx context.Context, sandbox sandboxstore.Sandbox) (uint64, error) {
|
|
var pidCount uint64
|
|
|
|
// get process count inside PodSandbox for Windows
|
|
task, err := sandbox.Container.Task(ctx, nil)
|
|
if err != nil {
|
|
if errdefs.IsNotFound(err) {
|
|
return 0, nil
|
|
}
|
|
return 0, err
|
|
}
|
|
processes, err := task.Pids(ctx)
|
|
if err != nil {
|
|
if errdefs.IsNotFound(err) {
|
|
return 0, nil
|
|
}
|
|
return 0, err
|
|
}
|
|
pidCount += uint64(len(processes))
|
|
|
|
for _, cntr := range c.containerStore.List() {
|
|
if cntr.SandboxID != sandbox.ID {
|
|
continue
|
|
}
|
|
|
|
state := cntr.Status.Get().State()
|
|
if state != runtime.ContainerState_CONTAINER_RUNNING {
|
|
continue
|
|
}
|
|
|
|
task, err := cntr.Container.Task(ctx, nil)
|
|
if err != nil {
|
|
return 0, err
|
|
}
|
|
|
|
processes, err := task.Pids(ctx)
|
|
if err != nil {
|
|
if errdefs.IsNotFound(err) {
|
|
continue
|
|
}
|
|
return 0, err
|
|
}
|
|
pidCount += uint64(len(processes))
|
|
|
|
}
|
|
|
|
return pidCount, nil
|
|
}
|