Update ingress godeps
This commit is contained in:
parent
d43021b3f1
commit
28db8fb16d
1068 changed files with 461467 additions and 117300 deletions
161
vendor/k8s.io/kubernetes/pkg/storage/etcd3/compact.go
generated
vendored
Normal file
161
vendor/k8s.io/kubernetes/pkg/storage/etcd3/compact.go
generated
vendored
Normal file
|
|
@ -0,0 +1,161 @@
|
|||
/*
|
||||
Copyright 2016 The Kubernetes 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 etcd3
|
||||
|
||||
import (
|
||||
"strconv"
|
||||
"sync"
|
||||
"time"
|
||||
|
||||
"github.com/coreos/etcd/clientv3"
|
||||
"github.com/golang/glog"
|
||||
"golang.org/x/net/context"
|
||||
)
|
||||
|
||||
const (
|
||||
compactInterval = 10 * time.Minute
|
||||
compactRevKey = "compact_rev_key"
|
||||
)
|
||||
|
||||
var (
|
||||
endpointsMapMu sync.Mutex
|
||||
endpointsMap map[string]struct{}
|
||||
)
|
||||
|
||||
func init() {
|
||||
endpointsMap = make(map[string]struct{})
|
||||
}
|
||||
|
||||
// StartCompactor starts a compactor in the background to compact old version of keys that's not needed.
|
||||
// By default, we save the most recent 10 minutes data and compact versions > 10minutes ago.
|
||||
// It should be enough for slow watchers and to tolerate burst.
|
||||
// TODO: We might keep a longer history (12h) in the future once storage API can take advantage of past version of keys.
|
||||
func StartCompactor(ctx context.Context, client *clientv3.Client) {
|
||||
endpointsMapMu.Lock()
|
||||
defer endpointsMapMu.Unlock()
|
||||
|
||||
// In one process, we can have only one compactor for one cluster.
|
||||
// Currently we rely on endpoints to differentiate clusters.
|
||||
for _, ep := range client.Endpoints() {
|
||||
if _, ok := endpointsMap[ep]; ok {
|
||||
glog.V(4).Infof("compactor already exists for endpoints %v", client.Endpoints())
|
||||
return
|
||||
}
|
||||
}
|
||||
for _, ep := range client.Endpoints() {
|
||||
endpointsMap[ep] = struct{}{}
|
||||
}
|
||||
|
||||
go compactor(ctx, client, compactInterval)
|
||||
}
|
||||
|
||||
// compactor periodically compacts historical versions of keys in etcd.
|
||||
// It will compact keys with versions older than given interval.
|
||||
// In other words, after compaction, it will only contain keys set during last interval.
|
||||
// Any API call for the older versions of keys will return error.
|
||||
// Interval is the time interval between each compaction. The first compaction happens after "interval".
|
||||
func compactor(ctx context.Context, client *clientv3.Client, interval time.Duration) {
|
||||
// Technical definitions:
|
||||
// We have a special key in etcd defined as *compactRevKey*.
|
||||
// compactRevKey's value will be set to the string of last compacted revision.
|
||||
// compactRevKey's version will be used as logical time for comparison. THe version is referred as compact time.
|
||||
// Initially, because the key doesn't exist, the compact time (version) is 0.
|
||||
//
|
||||
// Algorithm:
|
||||
// - Compare to see if (local compact_time) = (remote compact_time).
|
||||
// - If yes, increment both local and remote compact_time, and do a compaction.
|
||||
// - If not, set local to remote compact_time.
|
||||
//
|
||||
// Technical details/insights:
|
||||
//
|
||||
// The protocol here is lease based. If one compactor CAS successfully, the others would know it when they fail in
|
||||
// CAS later and would try again in 10 minutes. If an APIServer crashed, another one would "take over" the lease.
|
||||
//
|
||||
// For example, in the following diagram, we have a compactor C1 doing compaction in t1, t2. Another compactor C2
|
||||
// at t1' (t1 < t1' < t2) would CAS fail, set its known oldRev to rev at t1', and try again in t2' (t2' > t2).
|
||||
// If C1 crashed and wouldn't compact at t2, C2 would CAS successfully at t2'.
|
||||
//
|
||||
// oldRev(t2) curRev(t2)
|
||||
// +
|
||||
// oldRev curRev |
|
||||
// + + |
|
||||
// | | |
|
||||
// | | t1' | t2'
|
||||
// +---v-------------v----^---------v------^---->
|
||||
// t0 t1 t2
|
||||
//
|
||||
// We have the guarantees:
|
||||
// - in normal cases, the interval is 10 minutes.
|
||||
// - in failover, the interval is >10m and <20m
|
||||
//
|
||||
// FAQ:
|
||||
// - What if time is not accurate? We don't care as long as someone did the compaction. Atomicity is ensured using
|
||||
// etcd API.
|
||||
// - What happened under heavy load scenarios? Initially, each apiserver will do only one compaction
|
||||
// every 10 minutes. This is very unlikely affecting or affected w.r.t. server load.
|
||||
|
||||
var compactTime int64
|
||||
var rev int64
|
||||
var err error
|
||||
for {
|
||||
select {
|
||||
case <-time.After(interval):
|
||||
case <-ctx.Done():
|
||||
return
|
||||
}
|
||||
|
||||
compactTime, rev, err = compact(ctx, client, compactTime, rev)
|
||||
if err != nil {
|
||||
glog.Errorf("etcd: endpoint (%v) compact failed: %v", client.Endpoints(), err)
|
||||
continue
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// compact compacts etcd store and returns current rev.
|
||||
// It will return the current compact time and global revision if no error occurred.
|
||||
// Note that CAS fail will not incur any error.
|
||||
func compact(ctx context.Context, client *clientv3.Client, t, rev int64) (int64, int64, error) {
|
||||
resp, err := client.KV.Txn(ctx).If(
|
||||
clientv3.Compare(clientv3.Version(compactRevKey), "=", t),
|
||||
).Then(
|
||||
clientv3.OpPut(compactRevKey, strconv.FormatInt(rev, 10)), // Expect side effect: increment Version
|
||||
).Else(
|
||||
clientv3.OpGet(compactRevKey),
|
||||
).Commit()
|
||||
if err != nil {
|
||||
return t, rev, err
|
||||
}
|
||||
|
||||
curRev := resp.Header.Revision
|
||||
|
||||
if !resp.Succeeded {
|
||||
curTime := resp.Responses[0].GetResponseRange().Kvs[0].Version
|
||||
return curTime, curRev, nil
|
||||
}
|
||||
curTime := t + 1
|
||||
|
||||
if rev == 0 {
|
||||
// We don't compact on bootstrap.
|
||||
return curTime, curRev, nil
|
||||
}
|
||||
if _, err = client.Compact(ctx, rev); err != nil {
|
||||
return curTime, curRev, err
|
||||
}
|
||||
glog.Infof("etcd: compacted rev (%d), endpoints (%v)", rev, client.Endpoints())
|
||||
return curTime, curRev, nil
|
||||
}
|
||||
50
vendor/k8s.io/kubernetes/pkg/storage/etcd3/event.go
generated
vendored
Normal file
50
vendor/k8s.io/kubernetes/pkg/storage/etcd3/event.go
generated
vendored
Normal file
|
|
@ -0,0 +1,50 @@
|
|||
/*
|
||||
Copyright 2016 The Kubernetes 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 etcd3
|
||||
|
||||
import (
|
||||
"github.com/coreos/etcd/clientv3"
|
||||
"github.com/coreos/etcd/mvcc/mvccpb"
|
||||
)
|
||||
|
||||
type event struct {
|
||||
key string
|
||||
value []byte
|
||||
rev int64
|
||||
isDeleted bool
|
||||
isCreated bool
|
||||
}
|
||||
|
||||
func parseKV(kv *mvccpb.KeyValue) *event {
|
||||
return &event{
|
||||
key: string(kv.Key),
|
||||
value: kv.Value,
|
||||
rev: kv.ModRevision,
|
||||
isDeleted: false,
|
||||
isCreated: kv.ModRevision == kv.CreateRevision,
|
||||
}
|
||||
}
|
||||
|
||||
func parseEvent(e *clientv3.Event) *event {
|
||||
return &event{
|
||||
key: string(e.Kv.Key),
|
||||
value: e.Kv.Value,
|
||||
rev: e.Kv.ModRevision,
|
||||
isDeleted: e.Type == clientv3.EventTypeDelete,
|
||||
isCreated: e.IsCreate(),
|
||||
}
|
||||
}
|
||||
455
vendor/k8s.io/kubernetes/pkg/storage/etcd3/store.go
generated
vendored
Normal file
455
vendor/k8s.io/kubernetes/pkg/storage/etcd3/store.go
generated
vendored
Normal file
|
|
@ -0,0 +1,455 @@
|
|||
/*
|
||||
Copyright 2016 The Kubernetes 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 etcd3
|
||||
|
||||
import (
|
||||
"bytes"
|
||||
"errors"
|
||||
"fmt"
|
||||
"path"
|
||||
"reflect"
|
||||
"strings"
|
||||
|
||||
"k8s.io/kubernetes/pkg/api"
|
||||
"k8s.io/kubernetes/pkg/api/meta"
|
||||
"k8s.io/kubernetes/pkg/conversion"
|
||||
"k8s.io/kubernetes/pkg/runtime"
|
||||
"k8s.io/kubernetes/pkg/storage"
|
||||
"k8s.io/kubernetes/pkg/storage/etcd"
|
||||
"k8s.io/kubernetes/pkg/watch"
|
||||
|
||||
"github.com/coreos/etcd/clientv3"
|
||||
"github.com/golang/glog"
|
||||
"golang.org/x/net/context"
|
||||
)
|
||||
|
||||
type store struct {
|
||||
client *clientv3.Client
|
||||
codec runtime.Codec
|
||||
versioner storage.Versioner
|
||||
pathPrefix string
|
||||
watcher *watcher
|
||||
}
|
||||
|
||||
type elemForDecode struct {
|
||||
data []byte
|
||||
rev uint64
|
||||
}
|
||||
|
||||
type objState struct {
|
||||
obj runtime.Object
|
||||
meta *storage.ResponseMeta
|
||||
rev int64
|
||||
data []byte
|
||||
}
|
||||
|
||||
// New returns an etcd3 implementation of storage.Interface.
|
||||
func New(c *clientv3.Client, codec runtime.Codec, prefix string) storage.Interface {
|
||||
return newStore(c, codec, prefix)
|
||||
}
|
||||
|
||||
func newStore(c *clientv3.Client, codec runtime.Codec, prefix string) *store {
|
||||
versioner := etcd.APIObjectVersioner{}
|
||||
return &store{
|
||||
client: c,
|
||||
versioner: versioner,
|
||||
codec: codec,
|
||||
pathPrefix: prefix,
|
||||
watcher: newWatcher(c, codec, versioner),
|
||||
}
|
||||
}
|
||||
|
||||
// Versioner implements storage.Interface.Versioner.
|
||||
func (s *store) Versioner() storage.Versioner {
|
||||
return s.versioner
|
||||
}
|
||||
|
||||
// Get implements storage.Interface.Get.
|
||||
func (s *store) Get(ctx context.Context, key string, out runtime.Object, ignoreNotFound bool) error {
|
||||
key = keyWithPrefix(s.pathPrefix, key)
|
||||
getResp, err := s.client.KV.Get(ctx, key)
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
|
||||
if len(getResp.Kvs) == 0 {
|
||||
if ignoreNotFound {
|
||||
return runtime.SetZeroValue(out)
|
||||
}
|
||||
return storage.NewKeyNotFoundError(key, 0)
|
||||
}
|
||||
kv := getResp.Kvs[0]
|
||||
return decode(s.codec, s.versioner, kv.Value, out, kv.ModRevision)
|
||||
}
|
||||
|
||||
// Create implements storage.Interface.Create.
|
||||
func (s *store) Create(ctx context.Context, key string, obj, out runtime.Object, ttl uint64) error {
|
||||
if version, err := s.versioner.ObjectResourceVersion(obj); err == nil && version != 0 {
|
||||
return errors.New("resourceVersion should not be set on objects to be created")
|
||||
}
|
||||
data, err := runtime.Encode(s.codec, obj)
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
key = keyWithPrefix(s.pathPrefix, key)
|
||||
|
||||
opts, err := s.ttlOpts(ctx, int64(ttl))
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
|
||||
txnResp, err := s.client.KV.Txn(ctx).If(
|
||||
notFound(key),
|
||||
).Then(
|
||||
clientv3.OpPut(key, string(data), opts...),
|
||||
).Commit()
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
if !txnResp.Succeeded {
|
||||
return storage.NewKeyExistsError(key, 0)
|
||||
}
|
||||
|
||||
if out != nil {
|
||||
putResp := txnResp.Responses[0].GetResponsePut()
|
||||
return decode(s.codec, s.versioner, data, out, putResp.Header.Revision)
|
||||
}
|
||||
return nil
|
||||
}
|
||||
|
||||
// Delete implements storage.Interface.Delete.
|
||||
func (s *store) Delete(ctx context.Context, key string, out runtime.Object, precondtions *storage.Preconditions) error {
|
||||
v, err := conversion.EnforcePtr(out)
|
||||
if err != nil {
|
||||
panic("unable to convert output object to pointer")
|
||||
}
|
||||
key = keyWithPrefix(s.pathPrefix, key)
|
||||
if precondtions == nil {
|
||||
return s.unconditionalDelete(ctx, key, out)
|
||||
}
|
||||
return s.conditionalDelete(ctx, key, out, v, precondtions)
|
||||
}
|
||||
|
||||
func (s *store) unconditionalDelete(ctx context.Context, key string, out runtime.Object) error {
|
||||
// We need to do get and delete in single transaction in order to
|
||||
// know the value and revision before deleting it.
|
||||
txnResp, err := s.client.KV.Txn(ctx).If().Then(
|
||||
clientv3.OpGet(key),
|
||||
clientv3.OpDelete(key),
|
||||
).Commit()
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
getResp := txnResp.Responses[0].GetResponseRange()
|
||||
if len(getResp.Kvs) == 0 {
|
||||
return storage.NewKeyNotFoundError(key, 0)
|
||||
}
|
||||
|
||||
kv := getResp.Kvs[0]
|
||||
return decode(s.codec, s.versioner, kv.Value, out, kv.ModRevision)
|
||||
}
|
||||
|
||||
func (s *store) conditionalDelete(ctx context.Context, key string, out runtime.Object, v reflect.Value, precondtions *storage.Preconditions) error {
|
||||
getResp, err := s.client.KV.Get(ctx, key)
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
for {
|
||||
origState, err := s.getState(getResp, key, v, false)
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
if err := checkPreconditions(key, precondtions, origState.obj); err != nil {
|
||||
return err
|
||||
}
|
||||
txnResp, err := s.client.KV.Txn(ctx).If(
|
||||
clientv3.Compare(clientv3.ModRevision(key), "=", origState.rev),
|
||||
).Then(
|
||||
clientv3.OpDelete(key),
|
||||
).Else(
|
||||
clientv3.OpGet(key),
|
||||
).Commit()
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
if !txnResp.Succeeded {
|
||||
getResp = (*clientv3.GetResponse)(txnResp.Responses[0].GetResponseRange())
|
||||
glog.V(4).Infof("deletion of %s failed because of a conflict, going to retry", key)
|
||||
continue
|
||||
}
|
||||
return decode(s.codec, s.versioner, origState.data, out, origState.rev)
|
||||
}
|
||||
}
|
||||
|
||||
// GuaranteedUpdate implements storage.Interface.GuaranteedUpdate.
|
||||
func (s *store) GuaranteedUpdate(ctx context.Context, key string, out runtime.Object, ignoreNotFound bool, precondtions *storage.Preconditions, tryUpdate storage.UpdateFunc) error {
|
||||
v, err := conversion.EnforcePtr(out)
|
||||
if err != nil {
|
||||
panic("unable to convert output object to pointer")
|
||||
}
|
||||
key = keyWithPrefix(s.pathPrefix, key)
|
||||
getResp, err := s.client.KV.Get(ctx, key)
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
for {
|
||||
origState, err := s.getState(getResp, key, v, ignoreNotFound)
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
|
||||
if err := checkPreconditions(key, precondtions, origState.obj); err != nil {
|
||||
return err
|
||||
}
|
||||
|
||||
ret, ttl, err := s.updateState(origState, tryUpdate)
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
|
||||
data, err := runtime.Encode(s.codec, ret)
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
if bytes.Equal(data, origState.data) {
|
||||
return decode(s.codec, s.versioner, origState.data, out, origState.rev)
|
||||
}
|
||||
|
||||
opts, err := s.ttlOpts(ctx, int64(ttl))
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
|
||||
txnResp, err := s.client.KV.Txn(ctx).If(
|
||||
clientv3.Compare(clientv3.ModRevision(key), "=", origState.rev),
|
||||
).Then(
|
||||
clientv3.OpPut(key, string(data), opts...),
|
||||
).Else(
|
||||
clientv3.OpGet(key),
|
||||
).Commit()
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
if !txnResp.Succeeded {
|
||||
getResp = (*clientv3.GetResponse)(txnResp.Responses[0].GetResponseRange())
|
||||
glog.V(4).Infof("GuaranteedUpdate of %s failed because of a conflict, going to retry", key)
|
||||
continue
|
||||
}
|
||||
putResp := txnResp.Responses[0].GetResponsePut()
|
||||
return decode(s.codec, s.versioner, data, out, putResp.Header.Revision)
|
||||
}
|
||||
}
|
||||
|
||||
// GetToList implements storage.Interface.GetToList.
|
||||
func (s *store) GetToList(ctx context.Context, key string, filter storage.Filter, listObj runtime.Object) error {
|
||||
listPtr, err := meta.GetItemsPtr(listObj)
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
key = keyWithPrefix(s.pathPrefix, key)
|
||||
|
||||
getResp, err := s.client.KV.Get(ctx, key)
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
if len(getResp.Kvs) == 0 {
|
||||
return nil
|
||||
}
|
||||
elems := []*elemForDecode{{
|
||||
data: getResp.Kvs[0].Value,
|
||||
rev: uint64(getResp.Kvs[0].ModRevision),
|
||||
}}
|
||||
if err := decodeList(elems, filter, listPtr, s.codec, s.versioner); err != nil {
|
||||
return err
|
||||
}
|
||||
// update version with cluster level revision
|
||||
return s.versioner.UpdateList(listObj, uint64(getResp.Header.Revision))
|
||||
}
|
||||
|
||||
// List implements storage.Interface.List.
|
||||
func (s *store) List(ctx context.Context, key, resourceVersion string, filter storage.Filter, listObj runtime.Object) error {
|
||||
listPtr, err := meta.GetItemsPtr(listObj)
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
key = keyWithPrefix(s.pathPrefix, key)
|
||||
// We need to make sure the key ended with "/" so that we only get children "directories".
|
||||
// e.g. if we have key "/a", "/a/b", "/ab", getting keys with prefix "/a" will return all three,
|
||||
// while with prefix "/a/" will return only "/a/b" which is the correct answer.
|
||||
if !strings.HasSuffix(key, "/") {
|
||||
key += "/"
|
||||
}
|
||||
getResp, err := s.client.KV.Get(ctx, key, clientv3.WithPrefix())
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
|
||||
elems := make([]*elemForDecode, len(getResp.Kvs))
|
||||
for i, kv := range getResp.Kvs {
|
||||
elems[i] = &elemForDecode{
|
||||
data: kv.Value,
|
||||
rev: uint64(kv.ModRevision),
|
||||
}
|
||||
}
|
||||
if err := decodeList(elems, filter, listPtr, s.codec, s.versioner); err != nil {
|
||||
return err
|
||||
}
|
||||
// update version with cluster level revision
|
||||
return s.versioner.UpdateList(listObj, uint64(getResp.Header.Revision))
|
||||
}
|
||||
|
||||
// Watch implements storage.Interface.Watch.
|
||||
func (s *store) Watch(ctx context.Context, key string, resourceVersion string, filter storage.Filter) (watch.Interface, error) {
|
||||
return s.watch(ctx, key, resourceVersion, filter, false)
|
||||
}
|
||||
|
||||
// WatchList implements storage.Interface.WatchList.
|
||||
func (s *store) WatchList(ctx context.Context, key string, resourceVersion string, filter storage.Filter) (watch.Interface, error) {
|
||||
return s.watch(ctx, key, resourceVersion, filter, true)
|
||||
}
|
||||
|
||||
func (s *store) watch(ctx context.Context, key string, rv string, filter storage.Filter, recursive bool) (watch.Interface, error) {
|
||||
rev, err := storage.ParseWatchResourceVersion(rv)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
key = keyWithPrefix(s.pathPrefix, key)
|
||||
return s.watcher.Watch(ctx, key, int64(rev), recursive, filter)
|
||||
}
|
||||
|
||||
func (s *store) getState(getResp *clientv3.GetResponse, key string, v reflect.Value, ignoreNotFound bool) (*objState, error) {
|
||||
state := &objState{
|
||||
obj: reflect.New(v.Type()).Interface().(runtime.Object),
|
||||
meta: &storage.ResponseMeta{},
|
||||
}
|
||||
if len(getResp.Kvs) == 0 {
|
||||
if !ignoreNotFound {
|
||||
return nil, storage.NewKeyNotFoundError(key, 0)
|
||||
}
|
||||
if err := runtime.SetZeroValue(state.obj); err != nil {
|
||||
return nil, err
|
||||
}
|
||||
} else {
|
||||
state.rev = getResp.Kvs[0].ModRevision
|
||||
state.meta.ResourceVersion = uint64(state.rev)
|
||||
state.data = getResp.Kvs[0].Value
|
||||
if err := decode(s.codec, s.versioner, state.data, state.obj, state.rev); err != nil {
|
||||
return nil, err
|
||||
}
|
||||
}
|
||||
return state, nil
|
||||
}
|
||||
|
||||
func (s *store) updateState(st *objState, userUpdate storage.UpdateFunc) (runtime.Object, uint64, error) {
|
||||
ret, ttlPtr, err := userUpdate(st.obj, *st.meta)
|
||||
if err != nil {
|
||||
return nil, 0, err
|
||||
}
|
||||
|
||||
version, err := s.versioner.ObjectResourceVersion(ret)
|
||||
if err != nil {
|
||||
return nil, 0, err
|
||||
}
|
||||
if version != 0 {
|
||||
// We cannot store object with resourceVersion in etcd. We need to reset it.
|
||||
if err := s.versioner.UpdateObject(ret, 0); err != nil {
|
||||
return nil, 0, fmt.Errorf("UpdateObject failed: %v", err)
|
||||
}
|
||||
}
|
||||
var ttl uint64
|
||||
if ttlPtr != nil {
|
||||
ttl = *ttlPtr
|
||||
}
|
||||
return ret, ttl, nil
|
||||
}
|
||||
|
||||
// ttlOpts returns client options based on given ttl.
|
||||
// ttl: if ttl is non-zero, it will attach the key to a lease with ttl of roughly the same length
|
||||
func (s *store) ttlOpts(ctx context.Context, ttl int64) ([]clientv3.OpOption, error) {
|
||||
if ttl == 0 {
|
||||
return nil, nil
|
||||
}
|
||||
// TODO: one lease per ttl key is expensive. Based on current use case, we can have a long window to
|
||||
// put keys within into same lease. We shall benchmark this and optimize the performance.
|
||||
lcr, err := s.client.Lease.Grant(ctx, ttl)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
return []clientv3.OpOption{clientv3.WithLease(clientv3.LeaseID(lcr.ID))}, nil
|
||||
}
|
||||
|
||||
func keyWithPrefix(prefix, key string) string {
|
||||
if strings.HasPrefix(key, prefix) {
|
||||
return key
|
||||
}
|
||||
return path.Join(prefix, key)
|
||||
}
|
||||
|
||||
// decode decodes value of bytes into object. It will also set the object resource version to rev.
|
||||
// On success, objPtr would be set to the object.
|
||||
func decode(codec runtime.Codec, versioner storage.Versioner, value []byte, objPtr runtime.Object, rev int64) error {
|
||||
if _, err := conversion.EnforcePtr(objPtr); err != nil {
|
||||
panic("unable to convert output object to pointer")
|
||||
}
|
||||
_, _, err := codec.Decode(value, nil, objPtr)
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
// being unable to set the version does not prevent the object from being extracted
|
||||
versioner.UpdateObject(objPtr, uint64(rev))
|
||||
return nil
|
||||
}
|
||||
|
||||
// decodeList decodes a list of values into a list of objects, with resource version set to corresponding rev.
|
||||
// On success, ListPtr would be set to the list of objects.
|
||||
func decodeList(elems []*elemForDecode, filter storage.Filter, ListPtr interface{}, codec runtime.Codec, versioner storage.Versioner) error {
|
||||
v, err := conversion.EnforcePtr(ListPtr)
|
||||
if err != nil || v.Kind() != reflect.Slice {
|
||||
panic("need ptr to slice")
|
||||
}
|
||||
for _, elem := range elems {
|
||||
obj, _, err := codec.Decode(elem.data, nil, reflect.New(v.Type().Elem()).Interface().(runtime.Object))
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
// being unable to set the version does not prevent the object from being extracted
|
||||
versioner.UpdateObject(obj, elem.rev)
|
||||
if filter.Filter(obj) {
|
||||
v.Set(reflect.Append(v, reflect.ValueOf(obj).Elem()))
|
||||
}
|
||||
}
|
||||
return nil
|
||||
}
|
||||
|
||||
func checkPreconditions(key string, preconditions *storage.Preconditions, out runtime.Object) error {
|
||||
if preconditions == nil {
|
||||
return nil
|
||||
}
|
||||
objMeta, err := api.ObjectMetaFor(out)
|
||||
if err != nil {
|
||||
return storage.NewInternalErrorf("can't enforce preconditions %v on un-introspectable object %v, got error: %v", *preconditions, out, err)
|
||||
}
|
||||
if preconditions.UID != nil && *preconditions.UID != objMeta.UID {
|
||||
errMsg := fmt.Sprintf("Precondition failed: UID in precondition: %v, UID in object meta: %v", preconditions.UID, objMeta.UID)
|
||||
return storage.NewInvalidObjError(key, errMsg)
|
||||
}
|
||||
return nil
|
||||
}
|
||||
|
||||
func notFound(key string) clientv3.Cmp {
|
||||
return clientv3.Compare(clientv3.ModRevision(key), "=", 0)
|
||||
}
|
||||
347
vendor/k8s.io/kubernetes/pkg/storage/etcd3/watcher.go
generated
vendored
Normal file
347
vendor/k8s.io/kubernetes/pkg/storage/etcd3/watcher.go
generated
vendored
Normal file
|
|
@ -0,0 +1,347 @@
|
|||
/*
|
||||
Copyright 2016 The Kubernetes 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 etcd3
|
||||
|
||||
import (
|
||||
"fmt"
|
||||
"net/http"
|
||||
"strings"
|
||||
"sync"
|
||||
|
||||
"k8s.io/kubernetes/pkg/api/unversioned"
|
||||
"k8s.io/kubernetes/pkg/runtime"
|
||||
"k8s.io/kubernetes/pkg/storage"
|
||||
"k8s.io/kubernetes/pkg/watch"
|
||||
|
||||
"github.com/coreos/etcd/clientv3"
|
||||
etcdrpc "github.com/coreos/etcd/etcdserver/api/v3rpc/rpctypes"
|
||||
"github.com/golang/glog"
|
||||
"golang.org/x/net/context"
|
||||
"google.golang.org/grpc"
|
||||
"google.golang.org/grpc/codes"
|
||||
)
|
||||
|
||||
const (
|
||||
// We have set a buffer in order to reduce times of context switches.
|
||||
incomingBufSize = 100
|
||||
outgoingBufSize = 100
|
||||
)
|
||||
|
||||
type watcher struct {
|
||||
client *clientv3.Client
|
||||
codec runtime.Codec
|
||||
versioner storage.Versioner
|
||||
}
|
||||
|
||||
// watchChan implements watch.Interface.
|
||||
type watchChan struct {
|
||||
watcher *watcher
|
||||
key string
|
||||
initialRev int64
|
||||
recursive bool
|
||||
filter storage.Filter
|
||||
ctx context.Context
|
||||
cancel context.CancelFunc
|
||||
incomingEventChan chan *event
|
||||
resultChan chan watch.Event
|
||||
errChan chan error
|
||||
}
|
||||
|
||||
func newWatcher(client *clientv3.Client, codec runtime.Codec, versioner storage.Versioner) *watcher {
|
||||
return &watcher{
|
||||
client: client,
|
||||
codec: codec,
|
||||
versioner: versioner,
|
||||
}
|
||||
}
|
||||
|
||||
// Watch watches on a key and returns a watch.Interface that transfers relevant notifications.
|
||||
// If rev is zero, it will return the existing object(s) and then start watching from
|
||||
// the maximum revision+1 from returned objects.
|
||||
// If rev is non-zero, it will watch events happened after given revision.
|
||||
// If recursive is false, it watches on given key.
|
||||
// If recursive is true, it watches any children and directories under the key, excluding the root key itself.
|
||||
// filter must be non-nil. Only if filter returns true will the changes be returned.
|
||||
func (w *watcher) Watch(ctx context.Context, key string, rev int64, recursive bool, filter storage.Filter) (watch.Interface, error) {
|
||||
if recursive && !strings.HasSuffix(key, "/") {
|
||||
key += "/"
|
||||
}
|
||||
wc := w.createWatchChan(ctx, key, rev, recursive, filter)
|
||||
go wc.run()
|
||||
return wc, nil
|
||||
}
|
||||
|
||||
func (w *watcher) createWatchChan(ctx context.Context, key string, rev int64, recursive bool, filter storage.Filter) *watchChan {
|
||||
wc := &watchChan{
|
||||
watcher: w,
|
||||
key: key,
|
||||
initialRev: rev,
|
||||
recursive: recursive,
|
||||
filter: filter,
|
||||
incomingEventChan: make(chan *event, incomingBufSize),
|
||||
resultChan: make(chan watch.Event, outgoingBufSize),
|
||||
errChan: make(chan error, 1),
|
||||
}
|
||||
wc.ctx, wc.cancel = context.WithCancel(ctx)
|
||||
return wc
|
||||
}
|
||||
|
||||
func (wc *watchChan) run() {
|
||||
go wc.startWatching()
|
||||
|
||||
var resultChanWG sync.WaitGroup
|
||||
resultChanWG.Add(1)
|
||||
go wc.processEvent(&resultChanWG)
|
||||
|
||||
select {
|
||||
case err := <-wc.errChan:
|
||||
errResult := parseError(err)
|
||||
if errResult != nil {
|
||||
// error result is guaranteed to be received by user before closing ResultChan.
|
||||
select {
|
||||
case wc.resultChan <- *errResult:
|
||||
case <-wc.ctx.Done(): // user has given up all results
|
||||
}
|
||||
}
|
||||
wc.cancel()
|
||||
case <-wc.ctx.Done():
|
||||
}
|
||||
// we need to wait until resultChan wouldn't be sent to anymore
|
||||
resultChanWG.Wait()
|
||||
close(wc.resultChan)
|
||||
}
|
||||
|
||||
func (wc *watchChan) Stop() {
|
||||
wc.cancel()
|
||||
}
|
||||
|
||||
func (wc *watchChan) ResultChan() <-chan watch.Event {
|
||||
return wc.resultChan
|
||||
}
|
||||
|
||||
// sync tries to retrieve existing data and send them to process.
|
||||
// The revision to watch will be set to the revision in response.
|
||||
func (wc *watchChan) sync() error {
|
||||
opts := []clientv3.OpOption{}
|
||||
if wc.recursive {
|
||||
opts = append(opts, clientv3.WithPrefix())
|
||||
}
|
||||
getResp, err := wc.watcher.client.Get(wc.ctx, wc.key, opts...)
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
wc.initialRev = getResp.Header.Revision
|
||||
|
||||
for _, kv := range getResp.Kvs {
|
||||
wc.sendEvent(parseKV(kv))
|
||||
}
|
||||
return nil
|
||||
}
|
||||
|
||||
// startWatching does:
|
||||
// - get current objects if initialRev=0; set initialRev to current rev
|
||||
// - watch on given key and send events to process.
|
||||
func (wc *watchChan) startWatching() {
|
||||
if wc.initialRev == 0 {
|
||||
if err := wc.sync(); err != nil {
|
||||
wc.sendError(err)
|
||||
return
|
||||
}
|
||||
}
|
||||
opts := []clientv3.OpOption{clientv3.WithRev(wc.initialRev + 1)}
|
||||
if wc.recursive {
|
||||
opts = append(opts, clientv3.WithPrefix())
|
||||
}
|
||||
wch := wc.watcher.client.Watch(wc.ctx, wc.key, opts...)
|
||||
for wres := range wch {
|
||||
if wres.Err() != nil {
|
||||
// If there is an error on server (e.g. compaction), the channel will return it before closed.
|
||||
wc.sendError(wres.Err())
|
||||
return
|
||||
}
|
||||
for _, e := range wres.Events {
|
||||
wc.sendEvent(parseEvent(e))
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// processEvent processes events from etcd watcher and sends results to resultChan.
|
||||
func (wc *watchChan) processEvent(wg *sync.WaitGroup) {
|
||||
defer wg.Done()
|
||||
|
||||
for {
|
||||
select {
|
||||
case e := <-wc.incomingEventChan:
|
||||
res := wc.transform(e)
|
||||
if res == nil {
|
||||
continue
|
||||
}
|
||||
// If user couldn't receive results fast enough, we also block incoming events from watcher.
|
||||
// Because storing events in local will cause more memory usage.
|
||||
// The worst case would be closing the fast watcher.
|
||||
select {
|
||||
case wc.resultChan <- *res:
|
||||
case <-wc.ctx.Done():
|
||||
return
|
||||
}
|
||||
case <-wc.ctx.Done():
|
||||
return
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// transform transforms an event into a result for user if not filtered.
|
||||
// TODO (Optimization):
|
||||
// - Save remote round-trip.
|
||||
// Currently, DELETE and PUT event don't contain the previous value.
|
||||
// We need to do another Get() in order to get previous object and have logic upon it.
|
||||
// We could potentially do some optimizations:
|
||||
// - For PUT, we can save current and previous objects into the value.
|
||||
// - For DELETE, See https://github.com/coreos/etcd/issues/4620
|
||||
func (wc *watchChan) transform(e *event) (res *watch.Event) {
|
||||
curObj, oldObj, err := prepareObjs(wc.ctx, e, wc.watcher.client, wc.watcher.codec, wc.watcher.versioner)
|
||||
if err != nil {
|
||||
wc.sendError(err)
|
||||
return nil
|
||||
}
|
||||
|
||||
switch {
|
||||
case e.isDeleted:
|
||||
if !wc.filter.Filter(oldObj) {
|
||||
return nil
|
||||
}
|
||||
res = &watch.Event{
|
||||
Type: watch.Deleted,
|
||||
Object: oldObj,
|
||||
}
|
||||
case e.isCreated:
|
||||
if !wc.filter.Filter(curObj) {
|
||||
return nil
|
||||
}
|
||||
res = &watch.Event{
|
||||
Type: watch.Added,
|
||||
Object: curObj,
|
||||
}
|
||||
default:
|
||||
curObjPasses := wc.filter.Filter(curObj)
|
||||
oldObjPasses := wc.filter.Filter(oldObj)
|
||||
switch {
|
||||
case curObjPasses && oldObjPasses:
|
||||
res = &watch.Event{
|
||||
Type: watch.Modified,
|
||||
Object: curObj,
|
||||
}
|
||||
case curObjPasses && !oldObjPasses:
|
||||
res = &watch.Event{
|
||||
Type: watch.Added,
|
||||
Object: curObj,
|
||||
}
|
||||
case !curObjPasses && oldObjPasses:
|
||||
res = &watch.Event{
|
||||
Type: watch.Deleted,
|
||||
Object: oldObj,
|
||||
}
|
||||
}
|
||||
}
|
||||
return res
|
||||
}
|
||||
|
||||
func parseError(err error) *watch.Event {
|
||||
var status *unversioned.Status
|
||||
switch {
|
||||
case err == etcdrpc.ErrCompacted:
|
||||
status = &unversioned.Status{
|
||||
Status: unversioned.StatusFailure,
|
||||
Message: err.Error(),
|
||||
Code: http.StatusGone,
|
||||
Reason: unversioned.StatusReasonExpired,
|
||||
}
|
||||
default:
|
||||
status = &unversioned.Status{
|
||||
Status: unversioned.StatusFailure,
|
||||
Message: err.Error(),
|
||||
Code: http.StatusInternalServerError,
|
||||
Reason: unversioned.StatusReasonInternalError,
|
||||
}
|
||||
}
|
||||
|
||||
return &watch.Event{
|
||||
Type: watch.Error,
|
||||
Object: status,
|
||||
}
|
||||
}
|
||||
|
||||
func (wc *watchChan) sendError(err error) {
|
||||
// Context.canceled is an expected behavior.
|
||||
// We should just stop all goroutines in watchChan without returning error.
|
||||
// TODO: etcd client should return context.Canceled instead of grpc specific error.
|
||||
if grpc.Code(err) == codes.Canceled || err == context.Canceled {
|
||||
return
|
||||
}
|
||||
select {
|
||||
case wc.errChan <- err:
|
||||
case <-wc.ctx.Done():
|
||||
}
|
||||
}
|
||||
|
||||
func (wc *watchChan) sendEvent(e *event) {
|
||||
if len(wc.incomingEventChan) == incomingBufSize {
|
||||
glog.V(2).Infof("Fast watcher, slow processing. Number of buffered events: %d."+
|
||||
"Probably caused by slow decoding, user not receiving fast, or other processing logic",
|
||||
incomingBufSize)
|
||||
}
|
||||
select {
|
||||
case wc.incomingEventChan <- e:
|
||||
case <-wc.ctx.Done():
|
||||
}
|
||||
}
|
||||
|
||||
func prepareObjs(ctx context.Context, e *event, client *clientv3.Client, codec runtime.Codec, versioner storage.Versioner) (curObj runtime.Object, oldObj runtime.Object, err error) {
|
||||
if !e.isDeleted {
|
||||
curObj, err = decodeObj(codec, versioner, e.value, e.rev)
|
||||
if err != nil {
|
||||
return nil, nil, err
|
||||
}
|
||||
}
|
||||
if e.isDeleted || !e.isCreated {
|
||||
getResp, err := client.Get(ctx, e.key, clientv3.WithRev(e.rev-1))
|
||||
if err != nil {
|
||||
return nil, nil, err
|
||||
}
|
||||
// Note that this sends the *old* object with the etcd revision for the time at
|
||||
// which it gets deleted.
|
||||
// We assume old object is returned only in Deleted event. Users (e.g. cacher) need
|
||||
// to have larger than previous rev to tell the ordering.
|
||||
oldObj, err = decodeObj(codec, versioner, getResp.Kvs[0].Value, e.rev)
|
||||
if err != nil {
|
||||
return nil, nil, err
|
||||
}
|
||||
}
|
||||
return curObj, oldObj, nil
|
||||
}
|
||||
|
||||
func decodeObj(codec runtime.Codec, versioner storage.Versioner, data []byte, rev int64) (runtime.Object, error) {
|
||||
obj, err := runtime.Decode(codec, []byte(data))
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
// ensure resource version is set on the object we load from etcd
|
||||
if err := versioner.UpdateObject(obj, uint64(rev)); err != nil {
|
||||
return nil, fmt.Errorf("failure to version api object (%d) %#v: %v", rev, obj, err)
|
||||
}
|
||||
return obj, nil
|
||||
}
|
||||
Loading…
Add table
Add a link
Reference in a new issue