containerized-data-importer/pkg/controller/clone/host-clone.go
Arnon Gilboa d365661e30
Fix progress metric registration and parsing (#3292)
* Fix progress metric registration and parsing

Use default metric registration. We shouldn't use the controller-runtime
registration as we have no controller here and it will not register the
metric correctly.

Fix the metric parsing to match its new name. Otherwise the DV progress
will not be updated until its 100%.

Regression introduced in #3254

Signed-off-by: Arnon Gilboa <agilboa@redhat.com>

* Add kubevirt_cdi_import_progress_total metric

Use it in the importer instead of kubevirt_cdi_clone_progress_total and
fix metric parsing accordingly.

Signed-off-by: Arnon Gilboa <agilboa@redhat.com>

* Move ProgressFromClaim to host-clone

Nobody else is using it.

Signed-off-by: Arnon Gilboa <agilboa@redhat.com>

* Add ProgressMetric interface

ProgressReader can now work with either import or clone progress metric.
FIXME: consider removing the direct Add/Get and use only via interface.

Signed-off-by: Arnon Gilboa <agilboa@redhat.com>

* Refactor ProgressMetric interface

Signed-off-by: Arnon Gilboa <agilboa@redhat.com>

* Refactor progress parsing

Signed-off-by: Arnon Gilboa <agilboa@redhat.com>

* Refer metric names from the metrics package

Signed-off-by: Arnon Gilboa <agilboa@redhat.com>

---------

Signed-off-by: Arnon Gilboa <agilboa@redhat.com>
2024-06-03 13:25:37 +02:00

215 lines
5.7 KiB
Go

package clone
import (
"context"
"fmt"
"net/http"
"strconv"
"time"
"github.com/go-logr/logr"
corev1 "k8s.io/api/core/v1"
k8serrors "k8s.io/apimachinery/pkg/api/errors"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"k8s.io/client-go/tools/record"
"sigs.k8s.io/controller-runtime/pkg/client"
"sigs.k8s.io/controller-runtime/pkg/reconcile"
cdiv1 "kubevirt.io/containerized-data-importer-api/pkg/apis/core/v1beta1"
cc "kubevirt.io/containerized-data-importer/pkg/controller/common"
metrics "kubevirt.io/containerized-data-importer/pkg/monitoring/metrics/cdi-cloner"
)
// HostClonePhaseName is the name of the host clone phase
const HostClonePhaseName = "HostClone"
// HostClonePhase creates and monitors a dumb clone operation
type HostClonePhase struct {
Owner client.Object
Namespace string
SourceName string
DesiredClaim *corev1.PersistentVolumeClaim
ImmediateBind bool
OwnershipLabel string
Preallocation bool
PriorityClassName string
Client client.Client
Log logr.Logger
Recorder record.EventRecorder
}
var _ Phase = &HostClonePhase{}
var _ StatusReporter = &HostClonePhase{}
var httpClient *http.Client
func init() {
httpClient = cc.BuildHTTPClient(httpClient)
}
// Name returns the name of the phase
func (p *HostClonePhase) Name() string {
return HostClonePhaseName
}
// Status returns the phase status
func (p *HostClonePhase) Status(ctx context.Context) (*PhaseStatus, error) {
result := &PhaseStatus{}
pvc := &corev1.PersistentVolumeClaim{}
exists, err := getResource(ctx, p.Client, p.Namespace, p.DesiredClaim.Name, pvc)
if err != nil {
return nil, err
}
if !exists {
return result, nil
}
result.Annotations = pvc.Annotations
podName := pvc.Annotations[cc.AnnCloneSourcePod]
if podName == "" {
return result, nil
}
args := &progressFromClaimArgs{
Client: p.Client,
HTTPClient: httpClient,
Claim: pvc,
PodNamespace: p.Namespace,
PodName: podName,
OwnerUID: string(p.Owner.GetUID()),
}
progress, err := progressFromClaim(ctx, args)
if err != nil {
return nil, err
}
result.Progress = progress
return result, nil
}
// progressFromClaimArgs are the args for ProgressFromClaim
type progressFromClaimArgs struct {
Client client.Client
HTTPClient *http.Client
Claim *corev1.PersistentVolumeClaim
OwnerUID string
PodNamespace string
PodName string
}
// progressFromClaim returns the progres
func progressFromClaim(ctx context.Context, args *progressFromClaimArgs) (string, error) {
// Just set 100.0% if pod is succeeded
if args.Claim.Annotations[cc.AnnPodPhase] == string(corev1.PodSucceeded) {
return cc.ProgressDone, nil
}
pod := &corev1.Pod{
ObjectMeta: metav1.ObjectMeta{
Namespace: args.PodNamespace,
Name: args.PodName,
},
}
if err := args.Client.Get(ctx, client.ObjectKeyFromObject(pod), pod); err != nil {
if k8serrors.IsNotFound(err) {
return "", nil
}
return "", err
}
// This will only work when the clone source pod is running
if pod.Status.Phase != corev1.PodRunning {
return "", nil
}
url, err := cc.GetMetricsURL(pod)
if err != nil {
return "", err
}
if url == "" {
return "", nil
}
// We fetch the clone progress from the clone source pod metrics
progressReport, err := cc.GetProgressReportFromURL(url, args.HTTPClient, metrics.CloneProgressMetricName, args.OwnerUID)
if err != nil {
return "", err
}
if progressReport != "" {
if f, err := strconv.ParseFloat(progressReport, 64); err == nil {
return fmt.Sprintf("%.2f%%", f), nil
}
}
return "", nil
}
// Reconcile creates the desired pvc and waits for the operation to complete
func (p *HostClonePhase) Reconcile(ctx context.Context) (*reconcile.Result, error) {
actualClaim := &corev1.PersistentVolumeClaim{}
exists, err := getResource(ctx, p.Client, p.Namespace, p.DesiredClaim.Name, actualClaim)
if err != nil {
return nil, err
}
if !exists {
actualClaim, err = p.createClaim(ctx)
if err != nil {
return nil, err
}
}
if !p.hostCloneComplete(actualClaim) {
// requeue to update status
return &reconcile.Result{RequeueAfter: 3 * time.Second}, nil
}
return nil, nil
}
func (p *HostClonePhase) createClaim(ctx context.Context) (*corev1.PersistentVolumeClaim, error) {
claim := p.DesiredClaim.DeepCopy()
claim.Namespace = p.Namespace
cc.AddAnnotation(claim, cc.AnnPreallocationRequested, fmt.Sprintf("%t", p.Preallocation))
cc.AddAnnotation(claim, cc.AnnOwnerUID, string(p.Owner.GetUID()))
cc.AddAnnotation(claim, cc.AnnPodRestarts, "0")
cc.AddAnnotation(claim, cc.AnnCloneRequest, fmt.Sprintf("%s/%s", p.Namespace, p.SourceName))
cc.AddAnnotation(claim, cc.AnnPopulatorKind, cdiv1.VolumeCloneSourceRef)
cc.AddAnnotation(claim, cc.AnnEventSourceKind, p.Owner.GetObjectKind().GroupVersionKind().Kind)
cc.AddAnnotation(claim, cc.AnnEventSource, fmt.Sprintf("%s/%s", p.Owner.GetNamespace(), p.Owner.GetName()))
if p.OwnershipLabel != "" {
AddOwnershipLabel(p.OwnershipLabel, claim, p.Owner)
}
if p.ImmediateBind {
cc.AddAnnotation(claim, cc.AnnImmediateBinding, "")
}
if p.PriorityClassName != "" {
cc.AddAnnotation(claim, cc.AnnPriorityClassName, p.PriorityClassName)
}
if err := p.Client.Create(ctx, claim); err != nil {
checkQuotaExceeded(p.Recorder, p.Owner, err)
return nil, err
}
return claim, nil
}
func (p *HostClonePhase) hostCloneComplete(pvc *corev1.PersistentVolumeClaim) bool {
// this is awfully lame
// both the upload controller and clone controller update the PVC status to succeeded
// but only the clone controller will set the preallocation annotation
// so we have to wait for that
if p.Preallocation && pvc.Annotations[cc.AnnPreallocationApplied] != "true" {
return false
}
return pvc.Annotations[cc.AnnPodPhase] == string(cdiv1.Succeeded)
}