github/kubernetes
1
0
Fork 0
mirror of https://github.com/k3s-io/kubernetes.git synced 2025-07-24 04:06:03 +00:00
Code Issues Projects Releases Wiki Activity

Merge pull request #92537 from wojtek-t/fix_reflector_not_making_progress

Browse Source 
Fix bug in reflector not recovering from "Too large resource version"…
This commit is contained in:
Kubernetes Prow Robot 2020-07-01 00:07:47 -07:00 committed by GitHub
commit ed06981eab
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
4 changed files with 89 additions and 30 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
staging/src/k8s.io/apimachinery/pkg/apis/meta/v1/types.go
View File

@ -903,6 +903,9 @@ const (
// FieldManagerConflict is used to report when another client claims to manage this field,
// It should only be returned for a request using server-side apply.
CauseTypeFieldManagerConflict CauseType = "FieldManagerConflict"
// CauseTypeResourceVersionTooLarge is used to report that the requested resource version
// is newer than the data observed by the API server, so the request cannot be served.
CauseTypeResourceVersionTooLarge CauseType = "ResourceVersionTooLarge"
)
// +k8s:deepcopy-gen:interfaces=k8s.io/apimachinery/pkg/runtime.Object

19
staging/src/k8s.io/apiserver/pkg/storage/errors.go
View File

@ -177,7 +177,12 @@ var tooLargeResourceVersionCauseMsg = "Too large resource version"
// a minimum resource version that is larger than the largest currently available resource version for a requested resource.
func NewTooLargeResourceVersionError(minimumResourceVersion, currentRevision uint64, retrySeconds int) error {
err := errors.NewTimeoutError(fmt.Sprintf("Too large resource version: %d, current: %d", minimumResourceVersion, currentRevision), retrySeconds)
err.ErrStatus.Details.Causes = []metav1.StatusCause{{Message: tooLargeResourceVersionCauseMsg}}
err.ErrStatus.Details.Causes = []metav1.StatusCause{
{
Type: metav1.CauseTypeResourceVersionTooLarge,
Message: tooLargeResourceVersionCauseMsg,
},
}
return err
}
@ -186,15 +191,5 @@ func IsTooLargeResourceVersion(err error) bool {
if !errors.IsTimeout(err) {
return false
}
switch t := err.(type) {
case errors.APIStatus:
if d := t.Status().Details; d != nil {
for _, cause := range d.Causes {
if cause.Message == tooLargeResourceVersionCauseMsg {
return true
}
}
}
}
return false
return errors.HasStatusCause(err, metav1.CauseTypeResourceVersionTooLarge)
}

41
staging/src/k8s.io/client-go/tools/cache/reflector.go vendored
View File

@ -82,9 +82,9 @@ type Reflector struct {
// observed when doing a sync with the underlying store
// it is thread safe, but not synchronized with the underlying store
lastSyncResourceVersion string
// isLastSyncResourceVersionGone is true if the previous list or watch request with lastSyncResourceVersion
// failed with an HTTP 410 (Gone) status code.
isLastSyncResourceVersionGone bool
// isLastSyncResourceVersionUnavailable is true if the previous list or watch request with
// lastSyncResourceVersion failed with an "expired" or "too large resource version" error.
isLastSyncResourceVersionUnavailable bool
// lastSyncResourceVersionMutex guards read/write access to lastSyncResourceVersion
lastSyncResourceVersionMutex sync.RWMutex
// WatchListPageSize is the requested chunk size of initial and resync watch lists.
@ -115,7 +115,7 @@ type WatchErrorHandler func(r *Reflector, err error)
func DefaultWatchErrorHandler(r *Reflector, err error) {
switch {
case isExpiredError(err):
// Don't set LastSyncResourceVersionExpired - LIST call with ResourceVersion=RV already
// Don't set LastSyncResourceVersionUnavailable - LIST call with ResourceVersion=RV already
// has a semantic that it returns data at least as fresh as provided RV.
// So first try to LIST with setting RV to resource version of last observed object.
klog.V(4).Infof("%s: watch of %v closed with: %v", r.name, r.expectedTypeName, err)
@ -288,13 +288,14 @@ func (r *Reflector) ListAndWatch(stopCh <-chan struct{}) error {
}
list, paginatedResult, err = pager.List(context.Background(), options)
if isExpiredError(err) {
r.setIsLastSyncResourceVersionExpired(true)
// Retry immediately if the resource version used to list is expired.
if isExpiredError(err) || isTooLargeResourceVersionError(err) {
r.setIsLastSyncResourceVersionUnavailable(true)
// Retry immediately if the resource version used to list is unavailable.
// The pager already falls back to full list if paginated list calls fail due to an "Expired" error on
// continuation pages, but the pager might not be enabled, or the full list might fail because the
// resource version it is listing at is expired, so we need to fallback to resourceVersion="" in all
// to recover and ensure the reflector makes forward progress.
// continuation pages, but the pager might not be enabled, the full list might fail because the
// resource version it is listing at is expired or the cache may not yet be synced to the provided
// resource version. So we need to fallback to resourceVersion="" in all to recover and ensure
// the reflector makes forward progress.
list, paginatedResult, err = pager.List(context.Background(), metav1.ListOptions{ResourceVersion: r.relistResourceVersion()})
}
close(listCh)
@ -324,7 +325,7 @@ func (r *Reflector) ListAndWatch(stopCh <-chan struct{}) error {
r.paginatedResult = true
}
r.setIsLastSyncResourceVersionExpired(false) // list was successful
r.setIsLastSyncResourceVersionUnavailable(false) // list was successful
initTrace.Step("Objects listed")
listMetaInterface, err := meta.ListAccessor(list)
if err != nil {
@ -415,7 +416,7 @@ func (r *Reflector) ListAndWatch(stopCh <-chan struct{}) error {
if err != errorStopRequested {
switch {
case isExpiredError(err):
// Don't set LastSyncResourceVersionExpired - LIST call with ResourceVersion=RV already
// Don't set LastSyncResourceVersionUnavailable - LIST call with ResourceVersion=RV already
// has a semantic that it returns data at least as fresh as provided RV.
// So first try to LIST with setting RV to resource version of last observed object.
klog.V(4).Infof("%s: watch of %v closed with: %v", r.name, r.expectedTypeName, err)
@ -538,9 +539,9 @@ func (r *Reflector) relistResourceVersion() string {
r.lastSyncResourceVersionMutex.RLock()
defer r.lastSyncResourceVersionMutex.RUnlock()
if r.isLastSyncResourceVersionGone {
if r.isLastSyncResourceVersionUnavailable {
// Since this reflector makes paginated list requests, and all paginated list requests skip the watch cache
// if the lastSyncResourceVersion is expired, we set ResourceVersion="" and list again to re-establish reflector
// if the lastSyncResourceVersion is unavailable, we set ResourceVersion="" and list again to re-establish reflector
// to the latest available ResourceVersion, using a consistent read from etcd.
return ""
}
@ -552,12 +553,12 @@ func (r *Reflector) relistResourceVersion() string {
return r.lastSyncResourceVersion
}
// setIsLastSyncResourceVersionExpired sets if the last list or watch request with lastSyncResourceVersion returned a
// expired error: HTTP 410 (Gone) Status Code.
func (r *Reflector) setIsLastSyncResourceVersionExpired(isExpired bool) {
// setIsLastSyncResourceVersionUnavailable sets if the last list or watch request with lastSyncResourceVersion returned
// "expired" or "too large resource version" error.
func (r *Reflector) setIsLastSyncResourceVersionUnavailable(isUnavailable bool) {
r.lastSyncResourceVersionMutex.Lock()
defer r.lastSyncResourceVersionMutex.Unlock()
r.isLastSyncResourceVersionGone = isExpired
r.isLastSyncResourceVersionUnavailable = isUnavailable
}
func isExpiredError(err error) bool {
@ -567,3 +568,7 @@ func isExpiredError(err error) bool {
// check when we fully drop support for Kubernetes 1.17 servers from reflectors.
return apierrors.IsResourceExpired(err) || apierrors.IsGone(err)
}
func isTooLargeResourceVersionError(err error) bool {
return apierrors.HasStatusCause(err, metav1.CauseTypeResourceVersionTooLarge)
}

56
staging/src/k8s.io/client-go/tools/cache/reflector_test.go vendored
View File

@ -714,6 +714,62 @@ func TestReflectorFullListIfExpired(t *testing.T) {
}
}
func TestReflectorFullListIfTooLarge(t *testing.T) {
stopCh := make(chan struct{})
s := NewStore(MetaNamespaceKeyFunc)
listCallRVs := []string{}
lw := &testLW{
WatchFunc: func(options metav1.ListOptions) (watch.Interface, error) {
// Stop once the reflector begins watching since we're only interested in the list.
close(stopCh)
fw := watch.NewFake()
return fw, nil
},
ListFunc: func(options metav1.ListOptions) (runtime.Object, error) {
listCallRVs = append(listCallRVs, options.ResourceVersion)
switch options.ResourceVersion {
// initial list
case "0":
return &v1.PodList{ListMeta: metav1.ListMeta{ResourceVersion: "20"}}, nil
// relist after the initial list
case "20":
err := apierrors.NewTimeoutError("too large resource version", 1)
err.ErrStatus.Details.Causes = []metav1.StatusCause{{Type: metav1.CauseTypeResourceVersionTooLarge}}
return nil, err
// relist from etcd after "too large" error
case "":
return &v1.PodList{ListMeta: metav1.ListMeta{ResourceVersion: "10"}}, nil
default:
return nil, fmt.Errorf("unexpected List call: %s", options.ResourceVersion)
}
},
}
r := NewReflector(lw, &v1.Pod{}, s, 0)
// Initial list should use RV=0
if err := r.ListAndWatch(stopCh); err != nil {
t.Fatal(err)
}
// Relist from the future version.
// This may happen, as watchcache is initialized from "current global etcd resource version"
// when kube-apiserver is starting and if no objects are changing after that each kube-apiserver
// may be synced to a different version and they will never converge.
// TODO: We should use etcd progress-notify feature to avoid this behavior but until this is
// done we simply try to relist from now to avoid continuous errors on relists.
stopCh = make(chan struct{})
if err := r.ListAndWatch(stopCh); err != nil {
t.Fatal(err)
}
expectedRVs := []string{"0", "20", ""}
if !reflect.DeepEqual(listCallRVs, expectedRVs) {
t.Errorf("Expected series of list calls with resource version of %#v but got: %#v", expectedRVs, listCallRVs)
}
}
func TestReflectorSetExpectedType(t *testing.T) {
obj := &unstructured.Unstructured{}
gvk := schema.GroupVersionKind{