// CreateOrUpdate attempts to update the current etcd state with the provided
// allocation.
func (e *Etcd) CreateOrUpdate(snapshot *api.RangeAllocation) error {
e.lock.Lock()
defer e.lock.Unlock()
last := ""
err := e.storage.GuaranteedUpdate(context.TODO(), e.baseKey, &api.RangeAllocation{}, true,
storage.SimpleUpdate(func(input runtime.Object) (output runtime.Object, err error) {
existing := input.(*api.RangeAllocation)
switch {
case len(snapshot.ResourceVersion) != 0 && len(existing.ResourceVersion) != 0:
if snapshot.ResourceVersion != existing.ResourceVersion {
return nil, k8serr.NewConflict(e.kind, "", fmt.Errorf("the provided resource version does not match"))
}
case len(existing.ResourceVersion) != 0:
return nil, k8serr.NewConflict(e.kind, "", fmt.Errorf("another caller has already initialized the resource"))
}
last = snapshot.ResourceVersion
return snapshot, nil
}),
)
if err != nil {
return etcderr.InterpretUpdateError(err, e.kind, "")
}
err = e.alloc.Restore(snapshot.Range, snapshot.Data)
if err == nil {
e.last = last
}
return err
}
// UpdateRoute replaces an existing Route.
func (registry *Etcd) UpdateRoute(ctx kapi.Context, route *api.Route) error {
key, err := makeRouteKey(ctx, route.Name)
if err != nil {
return err
}
err = registry.Set(key, route, nil, 0)
return etcderr.InterpretUpdateError(err, "route", route.Name)
}
// UpdateService replaces an existing Service.
func (r *Registry) UpdateService(ctx api.Context, svc *api.Service) (*api.Service, error) {
key, err := makeServiceKey(ctx, svc.Name)
if err != nil {
return nil, err
}
out := &api.Service{}
err = r.Set(key, svc, out, 0)
return out, etcderr.InterpretUpdateError(err, "service", svc.Name)
}
// assignPod assigns the given pod to the given machine.
func (r *BindingREST) assignPod(ctx api.Context, podID string, machine string, annotations map[string]string) (err error) {
if _, err = r.setPodHostAndAnnotations(ctx, podID, "", machine, annotations); err != nil {
err = etcderr.InterpretGetError(err, "pod", podID)
err = etcderr.InterpretUpdateError(err, "pod", podID)
if _, ok := err.(*errors.StatusError); !ok {
err = errors.NewConflict("binding", podID, err)
}
}
return
}
func (r *RollbackREST) rollbackDeployment(ctx api.Context, deploymentID string, config *extensions.RollbackConfig, annotations map[string]string) (err error) {
if _, err = r.setDeploymentRollback(ctx, deploymentID, config, annotations); err != nil {
err = etcderr.InterpretGetError(err, extensions.Resource("deployments"), deploymentID)
err = etcderr.InterpretUpdateError(err, extensions.Resource("deployments"), deploymentID)
if _, ok := err.(*errors.StatusError); !ok {
err = errors.NewConflict(extensions.Resource("deployments/rollback"), deploymentID, err)
}
}
return
}
// assignPod assigns the given pod to the given machine.
func (r *BindingREST) assignPod(ctx api.Context, podID string, machine string, annotations map[string]string, cpuSet string, network api.Network) (err error) {
if _, err = r.setPodHostAndAnnotations(ctx, podID, "", machine, annotations, cpuSet, network); err != nil {
err = etcderr.InterpretGetError(err, api.Resource("pods"), podID)
err = etcderr.InterpretUpdateError(err, api.Resource("pods"), podID)
if _, ok := err.(*errors.StatusError); !ok {
err = errors.NewConflict(api.Resource("pods/binding"), podID, err)
}
}
return
}
// UpdateWithName updates the item with the provided name
// DEPRECATED: use Update instead
func (e *Etcd) UpdateWithName(ctx api.Context, name string, obj runtime.Object) error {
key, err := e.KeyFunc(ctx, name)
if err != nil {
return err
}
ttl, err := e.calculateTTL(obj, 0, true)
if err != nil {
return err
}
err = e.Storage.Set(key, obj, nil, ttl)
err = etcderr.InterpretUpdateError(err, e.EndpointName, name)
if err == nil && e.Decorator != nil {
err = e.Decorator(obj)
}
return err
}
// Delete removes the item from etcd.
func (e *Etcd) Delete(ctx api.Context, name string, options *api.DeleteOptions) (runtime.Object, error) {
key, err := e.KeyFunc(ctx, name)
if err != nil {
return nil, err
}
obj := e.NewFunc()
trace := util.NewTrace("Delete " + reflect.TypeOf(obj).String())
defer trace.LogIfLong(time.Second)
trace.Step("About to read object")
if err := e.Storage.Get(key, obj, false); err != nil {
return nil, etcderr.InterpretDeleteError(err, e.EndpointName, name)
}
// support older consumers of delete by treating "nil" as delete immediately
if options == nil {
options = api.NewDeleteOptions(0)
}
graceful, pendingGraceful, err := rest.BeforeDelete(e.DeleteStrategy, ctx, obj, options)
if err != nil {
return nil, err
}
if pendingGraceful {
return e.finalizeDelete(obj, false)
}
if graceful && *options.GracePeriodSeconds != 0 {
trace.Step("Graceful deletion")
out := e.NewFunc()
if err := e.Storage.Set(key, obj, out, uint64(*options.GracePeriodSeconds)); err != nil {
return nil, etcderr.InterpretUpdateError(err, e.EndpointName, name)
}
return e.finalizeDelete(out, true)
}
// delete immediately, or no graceful deletion supported
out := e.NewFunc()
trace.Step("About to delete object")
if err := e.Storage.Delete(key, out); err != nil {
return nil, etcderr.InterpretDeleteError(err, e.EndpointName, name)
}
return e.finalizeDelete(out, true)
}
// tryUpdate performs a read-update to persist the latest snapshot state of allocation.
func (e *Etcd) tryUpdate(fn func() error) error {
err := e.storage.GuaranteedUpdate(context.TODO(), e.baseKey, &api.RangeAllocation{}, true,
storage.SimpleUpdate(func(input runtime.Object) (output runtime.Object, err error) {
existing := input.(*api.RangeAllocation)
if len(existing.ResourceVersion) == 0 {
return nil, fmt.Errorf("cannot allocate resources of type %s at this time", e.kind)
}
if existing.ResourceVersion != e.last {
if err := e.alloc.Restore(existing.Range, existing.Data); err != nil {
return nil, err
}
if err := fn(); err != nil {
return nil, err
}
}
e.last = existing.ResourceVersion
rangeSpec, data := e.alloc.Snapshot()
existing.Range = rangeSpec
existing.Data = data
return existing, nil
}),
)
return etcderr.InterpretUpdateError(err, e.kind, "")
}
// Delete removes the item from etcd.
func (e *Etcd) Delete(ctx api.Context, name string, options *api.DeleteOptions) (runtime.Object, error) {
key, err := e.KeyFunc(ctx, name)
if err != nil {
return nil, err
}
obj := e.NewFunc()
trace := util.NewTrace("Delete " + reflect.TypeOf(obj).String())
defer trace.LogIfLong(time.Second)
trace.Step("About to read object")
if err := e.Storage.Get(ctx, key, obj, false); err != nil {
return nil, etcderr.InterpretDeleteError(err, e.EndpointName, name)
}
// support older consumers of delete by treating "nil" as delete immediately
if options == nil {
options = api.NewDeleteOptions(0)
}
graceful, pendingGraceful, err := rest.BeforeDelete(e.DeleteStrategy, ctx, obj, options)
if err != nil {
return nil, err
}
if pendingGraceful {
return e.finalizeDelete(obj, false)
}
if graceful {
trace.Step("Graceful deletion")
out := e.NewFunc()
lastGraceful := int64(0)
err := e.Storage.GuaranteedUpdate(
ctx, key, out, false,
storage.SimpleUpdate(func(existing runtime.Object) (runtime.Object, error) {
graceful, pendingGraceful, err := rest.BeforeDelete(e.DeleteStrategy, ctx, existing, options)
if err != nil {
return nil, err
}
if pendingGraceful {
return nil, errAlreadyDeleting
}
if !graceful {
return nil, errDeleteNow
}
lastGraceful = *options.GracePeriodSeconds
return existing, nil
}),
)
switch err {
case nil:
if lastGraceful > 0 {
return out, nil
}
// fall through and delete immediately
case errDeleteNow:
// we've updated the object to have a zero grace period, or it's already at 0, so
// we should fall through and truly delete the object.
case errAlreadyDeleting:
return e.finalizeDelete(obj, true)
default:
return nil, etcderr.InterpretUpdateError(err, e.EndpointName, name)
}
}
// delete immediately, or no graceful deletion supported
out := e.NewFunc()
trace.Step("About to delete object")
if err := e.Storage.Delete(ctx, key, out); err != nil {
return nil, etcderr.InterpretDeleteError(err, e.EndpointName, name)
}
return e.finalizeDelete(out, true)
}
// Update performs an atomic update and set of the object. Returns the result of the update
// or an error. If the registry allows create-on-update, the create flow will be executed.
// A bool is returned along with the object and any errors, to indicate object creation.
func (e *Etcd) Update(ctx api.Context, obj runtime.Object) (runtime.Object, bool, error) {
trace := util.NewTrace("Update " + reflect.TypeOf(obj).String())
defer trace.LogIfLong(time.Second)
name, err := e.ObjectNameFunc(obj)
if err != nil {
return nil, false, err
}
key, err := e.KeyFunc(ctx, name)
if err != nil {
return nil, false, err
}
// If AllowUnconditionalUpdate() is true and the object specified by the user does not have a resource version,
// then we populate it with the latest version.
// Else, we check that the version specified by the user matches the version of latest etcd object.
resourceVersion, err := e.Storage.Versioner().ObjectResourceVersion(obj)
if err != nil {
return nil, false, err
}
doUnconditionalUpdate := resourceVersion == 0 && e.UpdateStrategy.AllowUnconditionalUpdate()
// TODO: expose TTL
creating := false
out := e.NewFunc()
err = e.Storage.GuaranteedUpdate(ctx, key, out, true, func(existing runtime.Object, res storage.ResponseMeta) (runtime.Object, *uint64, error) {
version, err := e.Storage.Versioner().ObjectResourceVersion(existing)
if err != nil {
return nil, nil, err
}
if version == 0 {
if !e.UpdateStrategy.AllowCreateOnUpdate() {
return nil, nil, kubeerr.NewNotFound(e.EndpointName, name)
}
creating = true
if err := rest.BeforeCreate(e.CreateStrategy, ctx, obj); err != nil {
return nil, nil, err
}
ttl, err := e.calculateTTL(obj, 0, false)
if err != nil {
return nil, nil, err
}
return obj, &ttl, nil
}
creating = false
if doUnconditionalUpdate {
// Update the object's resource version to match the latest etcd object's resource version.
err = e.Storage.Versioner().UpdateObject(obj, res.Expiration, res.ResourceVersion)
if err != nil {
return nil, nil, err
}
} else {
// Check if the object's resource version matches the latest resource version.
newVersion, err := e.Storage.Versioner().ObjectResourceVersion(obj)
if err != nil {
return nil, nil, err
}
if newVersion != version {
return nil, nil, kubeerr.NewConflict(e.EndpointName, name, fmt.Errorf("the object has been modified; please apply your changes to the latest version and try again"))
}
}
if err := rest.BeforeUpdate(e.UpdateStrategy, ctx, obj, existing); err != nil {
return nil, nil, err
}
ttl, err := e.calculateTTL(obj, res.TTL, true)
if err != nil {
return nil, nil, err
}
if int64(ttl) != res.TTL {
return obj, &ttl, nil
}
return obj, nil, nil
})
if err != nil {
if creating {
err = etcderr.InterpretCreateError(err, e.EndpointName, name)
err = rest.CheckGeneratedNameError(e.CreateStrategy, err, obj)
} else {
err = etcderr.InterpretUpdateError(err, e.EndpointName, name)
}
return nil, false, err
}
if creating {
if e.AfterCreate != nil {
if err := e.AfterCreate(out); err != nil {
return nil, false, err
}
}
} else {
if e.AfterUpdate != nil {
if err := e.AfterUpdate(out); err != nil {
return nil, false, err
}
}
}
if e.Decorator != nil {
if err := e.Decorator(obj); err != nil {
return nil, false, err
}
}
return out, creating, nil
}
// Update performs an atomic update and set of the object. Returns the result of the update
// or an error. If the registry allows create-on-update, the create flow will be executed.
// A bool is returned along with the object and any errors, to indicate object creation.
func (e *Etcd) Update(ctx api.Context, obj runtime.Object) (runtime.Object, bool, error) {
name, err := e.ObjectNameFunc(obj)
if err != nil {
return nil, false, err
}
key, err := e.KeyFunc(ctx, name)
if err != nil {
return nil, false, err
}
// If AllowUnconditionalUpdate() is true and the object specified by the user does not have a resource version,
// then we populate it with the latest version.
// Else, we check that the version specified by the user matches the version of latest etcd object.
resourceVersion, err := e.Storage.Versioner().ObjectResourceVersion(obj)
if err != nil {
return nil, false, err
}
doUnconditionalUpdate := resourceVersion == 0 && e.UpdateStrategy.AllowUnconditionalUpdate()
// TODO: expose TTL
creating := false
out := e.NewFunc()
err = e.Storage.GuaranteedUpdate(ctx, key, out, true, func(existing runtime.Object, res storage.ResponseMeta) (runtime.Object, *uint64, error) {
// Since we return 'obj' from this function and it can be modified outside this
// function, we are resetting resourceVersion to the initial value here.
//
// TODO: In fact, we should probably return a DeepCopy of obj in all places.
err := e.Storage.Versioner().UpdateObject(obj, nil, resourceVersion)
if err != nil {
return nil, nil, err
}
version, err := e.Storage.Versioner().ObjectResourceVersion(existing)
if err != nil {
return nil, nil, err
}
if version == 0 {
if !e.UpdateStrategy.AllowCreateOnUpdate() {
return nil, nil, kubeerr.NewNotFound(e.QualifiedResource, name)
}
creating = true
if err := rest.BeforeCreate(e.CreateStrategy, ctx, obj); err != nil {
return nil, nil, err
}
ttl, err := e.calculateTTL(obj, 0, false)
if err != nil {
return nil, nil, err
}
return obj, &ttl, nil
}
creating = false
if doUnconditionalUpdate {
// Update the object's resource version to match the latest etcd object's resource version.
err = e.Storage.Versioner().UpdateObject(obj, res.Expiration, res.ResourceVersion)
if err != nil {
return nil, nil, err
}
} else {
// Check if the object's resource version matches the latest resource version.
newVersion, err := e.Storage.Versioner().ObjectResourceVersion(obj)
if err != nil {
return nil, nil, err
}
if newVersion == 0 {
// TODO: The Invalid error should has a field for Resource.
// After that field is added, we should fill the Resource and
// leave the Kind field empty. See the discussion in #18526.
qualifiedKind := unversioned.GroupKind{e.QualifiedResource.Group, e.QualifiedResource.Resource}
fieldErrList := field.ErrorList{field.Invalid(field.NewPath("metadata").Child("resourceVersion"), newVersion, "must be specified for an update")}
return nil, nil, kubeerr.NewInvalid(qualifiedKind, name, fieldErrList)
}
if newVersion != version {
return nil, nil, kubeerr.NewConflict(e.QualifiedResource, name, fmt.Errorf("the object has been modified; please apply your changes to the latest version and try again"))
}
}
if err := rest.BeforeUpdate(e.UpdateStrategy, ctx, obj, existing); err != nil {
return nil, nil, err
}
ttl, err := e.calculateTTL(obj, res.TTL, true)
if err != nil {
return nil, nil, err
}
if int64(ttl) != res.TTL {
return obj, &ttl, nil
}
return obj, nil, nil
})
if err != nil {
if creating {
err = etcderr.InterpretCreateError(err, e.QualifiedResource, name)
err = rest.CheckGeneratedNameError(e.CreateStrategy, err, obj)
} else {
err = etcderr.InterpretUpdateError(err, e.QualifiedResource, name)
}
return nil, false, err
}
if creating {
if e.AfterCreate != nil {
if err := e.AfterCreate(out); err != nil {
return nil, false, err
}
}
} else {
if e.AfterUpdate != nil {
if err := e.AfterUpdate(out); err != nil {
return nil, false, err
}
}
}
if e.Decorator != nil {
if err := e.Decorator(obj); err != nil {
return nil, false, err
}
}
return out, creating, nil
}
// Delete enforces life-cycle rules for namespace termination
func (r *REST) Delete(ctx api.Context, name string, options *api.DeleteOptions) (runtime.Object, error) {
nsObj, err := r.Get(ctx, name)
if err != nil {
return nil, err
}
namespace := nsObj.(*api.Namespace)
// upon first request to delete, we switch the phase to start namespace termination
// TODO: enhance graceful deletion's calls to DeleteStrategy to allow phase change and finalizer patterns
if namespace.DeletionTimestamp.IsZero() {
key, err := r.Etcd.KeyFunc(ctx, name)
if err != nil {
return nil, err
}
out := r.Etcd.NewFunc()
err = r.Etcd.Storage.GuaranteedUpdate(
ctx, key, out, false,
storage.SimpleUpdate(func(existing runtime.Object) (runtime.Object, error) {
existingNamespace, ok := existing.(*api.Namespace)
if !ok {
// wrong type
return nil, fmt.Errorf("expected *api.Namespace, got %v", existing)
}
if existingNamespace.UID != namespace.UID {
return nil, apierrors.NewConflict(
api.Resource("namespaces"),
name,
fmt.Errorf("UID in precondition: %v, UID in object meta: %v", namespace.UID, existingNamespace.UID),
)
}
// Set the deletion timestamp if needed
if existingNamespace.DeletionTimestamp.IsZero() {
now := unversioned.Now()
existingNamespace.DeletionTimestamp = &now
}
// Set the namespace phase to terminating, if needed
if existingNamespace.Status.Phase != api.NamespaceTerminating {
existingNamespace.Status.Phase = api.NamespaceTerminating
}
return existingNamespace, nil
}),
)
if err != nil {
err = storageerr.InterpretGetError(err, api.Resource("namespaces"), name)
err = storageerr.InterpretUpdateError(err, api.Resource("namespaces"), name)
if _, ok := err.(*apierrors.StatusError); !ok {
err = apierrors.NewInternalError(err)
}
return nil, err
}
return out, nil
}
// prior to final deletion, we must ensure that finalizers is empty
if len(namespace.Spec.Finalizers) != 0 {
err = apierrors.NewConflict(api.Resource("namespaces"), namespace.Name, fmt.Errorf("The system is ensuring all content is removed from this namespace. Upon completion, this namespace will automatically be purged by the system."))
return nil, err
}
return r.Etcd.Delete(ctx, name, nil)
}
// Delete removes the item from etcd.
func (e *Etcd) Delete(ctx api.Context, name string, options *api.DeleteOptions) (runtime.Object, error) {
key, err := e.KeyFunc(ctx, name)
if err != nil {
return nil, err
}
obj := e.NewFunc()
if err := e.Storage.Get(ctx, key, obj, false); err != nil {
return nil, etcderr.InterpretDeleteError(err, e.QualifiedResource, name)
}
// support older consumers of delete by treating "nil" as delete immediately
if options == nil {
options = api.NewDeleteOptions(0)
}
graceful, pendingGraceful, err := rest.BeforeDelete(e.DeleteStrategy, ctx, obj, options)
if err != nil {
return nil, err
}
if pendingGraceful {
return e.finalizeDelete(obj, false)
}
var ignoreNotFound bool = false
var lastExisting runtime.Object = nil
if graceful {
out := e.NewFunc()
lastGraceful := int64(0)
err := e.Storage.GuaranteedUpdate(
ctx, key, out, false,
storage.SimpleUpdate(func(existing runtime.Object) (runtime.Object, error) {
graceful, pendingGraceful, err := rest.BeforeDelete(e.DeleteStrategy, ctx, existing, options)
if err != nil {
return nil, err
}
if pendingGraceful {
return nil, errAlreadyDeleting
}
if !graceful {
return nil, errDeleteNow
}
lastGraceful = *options.GracePeriodSeconds
lastExisting = existing
return existing, nil
}),
)
switch err {
case nil:
if lastGraceful > 0 {
return out, nil
}
// If we are here, the registry supports grace period mechanism and
// we are intentionally delete gracelessly. In this case, we may
// enter a race with other k8s components. If other component wins
// the race, the object will not be found, and we should tolerate
// the NotFound error. See
// https://github.com/kubernetes/kubernetes/issues/19403 for
// details.
ignoreNotFound = true
// exit the switch and delete immediately
case errDeleteNow:
// we've updated the object to have a zero grace period, or it's already at 0, so
// we should fall through and truly delete the object.
case errAlreadyDeleting:
return e.finalizeDelete(obj, true)
default:
return nil, etcderr.InterpretUpdateError(err, e.QualifiedResource, name)
}
}
// delete immediately, or no graceful deletion supported
out := e.NewFunc()
if err := e.Storage.Delete(ctx, key, out); err != nil {
// Please refer to the place where we set ignoreNotFound for the reason
// why we ignore the NotFound error .
if storage.IsNotFound(err) && ignoreNotFound && lastExisting != nil {
// The lastExisting object may not be the last state of the object
// before its deletion, but it's the best approximation.
return e.finalizeDelete(lastExisting, true)
}
return nil, etcderr.InterpretDeleteError(err, e.QualifiedResource, name)
}
return e.finalizeDelete(out, true)
}