Check in the vendor directory

Travis seems to be having issues pulling deps, so we'll have to check in
the vendor directory and prevent the makefile from trying to regenerate
it normally.

vendor/k8s.io/apiserver/pkg/storage/etcd3/OWNERS

@@ -0,0 +1,5 @@
+reviewers:
+- wojtek-t
+- timothysc
+- madhusudancs
+- hongchaodeng

vendor/k8s.io/apiserver/pkg/storage/etcd3/compact.go

@@ -0,0 +1,162 @@
+/*
+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 (
+	"context"
+	"strconv"
+	"sync"
+	"time"
+
+	"github.com/coreos/etcd/clientv3"
+	"github.com/golang/glog"
+)
+
+const (
+	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, compactInterval time.Duration) {
+	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{}{}
+	}
+
+	if compactInterval != 0 {
+		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.V(4).Infof("etcd: compacted rev (%d), endpoints (%v)", rev, client.Endpoints())
+	return curTime, curRev, nil
+}

vendor/k8s.io/apiserver/pkg/storage/etcd3/errors.go

@@ -0,0 +1,42 @@
+/*
+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 (
+	"k8s.io/apimachinery/pkg/api/errors"
+
+	etcdrpc "github.com/coreos/etcd/etcdserver/api/v3rpc/rpctypes"
+)
+
+func interpretWatchError(err error) error {
+	switch {
+	case err == etcdrpc.ErrCompacted:
+		return errors.NewResourceExpired("The resourceVersion for the provided watch is too old.")
+	}
+	return err
+}
+
+func interpretListError(err error, paging bool) error {
+	switch {
+	case err == etcdrpc.ErrCompacted:
+		if paging {
+			return errors.NewResourceExpired("The provided from parameter is too old to display a consistent list result. You must start a new list without the from.")
+		}
+		return errors.NewResourceExpired("The resourceVersion for the provided list is too old.")
+	}
+	return err
+}

vendor/k8s.io/apiserver/pkg/storage/etcd3/event.go

@@ -0,0 +1,57 @@
+/*
+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
+	prevValue []byte
+	rev       int64
+	isDeleted bool
+	isCreated bool
+}
+
+// parseKV converts a KeyValue retrieved from an initial sync() listing to a synthetic isCreated event.
+func parseKV(kv *mvccpb.KeyValue) *event {
+	return &event{
+		key:       string(kv.Key),
+		value:     kv.Value,
+		prevValue: nil,
+		rev:       kv.ModRevision,
+		isDeleted: false,
+		isCreated: true,
+	}
+}
+
+func parseEvent(e *clientv3.Event) *event {
+	ret := &event{
+		key:       string(e.Kv.Key),
+		value:     e.Kv.Value,
+		rev:       e.Kv.ModRevision,
+		isDeleted: e.Type == clientv3.EventTypeDelete,
+		isCreated: e.IsCreate(),
+	}
+	if e.PrevKv != nil {
+		ret.prevValue = e.PrevKv.Value
+	}
+	return ret
+}

vendor/k8s.io/apiserver/pkg/storage/etcd3/preflight/checks.go

@@ -0,0 +1,70 @@
+/*
+Copyright 2017 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 preflight
+
+import (
+	"fmt"
+	"math/rand"
+	"net"
+	"net/url"
+	"time"
+)
+
+const connectionTimeout = 1 * time.Second
+
+// EtcdConnection holds the Etcd server list
+type EtcdConnection struct {
+	ServerList []string
+}
+
+func (EtcdConnection) serverReachable(connURL *url.URL) bool {
+	scheme := connURL.Scheme
+	if scheme == "http" || scheme == "https" || scheme == "tcp" {
+		scheme = "tcp"
+	}
+	if conn, err := net.DialTimeout(scheme, connURL.Host, connectionTimeout); err == nil {
+		defer conn.Close()
+		return true
+	}
+	return false
+}
+
+func parseServerURI(serverURI string) (*url.URL, error) {
+	connURL, err := url.Parse(serverURI)
+	if err != nil {
+		return &url.URL{}, fmt.Errorf("unable to parse etcd url: %v", err)
+	}
+	return connURL, nil
+}
+
+// CheckEtcdServers will attempt to reach all etcd servers once. If any
+// can be reached, return true.
+func (con EtcdConnection) CheckEtcdServers() (done bool, err error) {
+	// Attempt to reach every Etcd server randomly.
+	serverNumber := len(con.ServerList)
+	serverPerms := rand.Perm(serverNumber)
+	for _, index := range serverPerms {
+		host, err := parseServerURI(con.ServerList[index])
+		if err != nil {
+			return false, err
+		}
+		if con.serverReachable(host) {
+			return true, nil
+		}
+	}
+	return false, nil
+}

vendor/k8s.io/apiserver/pkg/storage/etcd3/store.go

@@ -0,0 +1,816 @@
+/*
+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"
+	"context"
+	"encoding/base64"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"path"
+	"reflect"
+	"strings"
+	"time"
+
+	"github.com/coreos/etcd/clientv3"
+	"github.com/golang/glog"
+
+	apierrors "k8s.io/apimachinery/pkg/api/errors"
+	"k8s.io/apimachinery/pkg/api/meta"
+	"k8s.io/apimachinery/pkg/conversion"
+	"k8s.io/apimachinery/pkg/runtime"
+	utilruntime "k8s.io/apimachinery/pkg/util/runtime"
+	"k8s.io/apimachinery/pkg/watch"
+	"k8s.io/apiserver/pkg/storage"
+	"k8s.io/apiserver/pkg/storage/etcd"
+	"k8s.io/apiserver/pkg/storage/value"
+	utiltrace "k8s.io/apiserver/pkg/util/trace"
+)
+
+// authenticatedDataString satisfies the value.Context interface. It uses the key to
+// authenticate the stored data. This does not defend against reuse of previously
+// encrypted values under the same key, but will prevent an attacker from using an
+// encrypted value from a different key. A stronger authenticated data segment would
+// include the etcd3 Version field (which is incremented on each write to a key and
+// reset when the key is deleted), but an attacker with write access to etcd can
+// force deletion and recreation of keys to weaken that angle.
+type authenticatedDataString string
+
+// AuthenticatedData implements the value.Context interface.
+func (d authenticatedDataString) AuthenticatedData() []byte {
+	return []byte(string(d))
+}
+
+var _ value.Context = authenticatedDataString("")
+
+type store struct {
+	client *clientv3.Client
+	// getOpts contains additional options that should be passed
+	// to all Get() calls.
+	getOps        []clientv3.OpOption
+	codec         runtime.Codec
+	versioner     storage.Versioner
+	transformer   value.Transformer
+	pathPrefix    string
+	watcher       *watcher
+	pagingEnabled bool
+}
+
+type elemForDecode struct {
+	data []byte
+	rev  uint64
+}
+
+type objState struct {
+	obj   runtime.Object
+	meta  *storage.ResponseMeta
+	rev   int64
+	data  []byte
+	stale bool
+}
+
+// New returns an etcd3 implementation of storage.Interface.
+func New(c *clientv3.Client, codec runtime.Codec, prefix string, transformer value.Transformer, pagingEnabled bool) storage.Interface {
+	return newStore(c, true, pagingEnabled, codec, prefix, transformer)
+}
+
+// NewWithNoQuorumRead returns etcd3 implementation of storage.Interface
+// where Get operations don't require quorum read.
+func NewWithNoQuorumRead(c *clientv3.Client, codec runtime.Codec, prefix string, transformer value.Transformer, pagingEnabled bool) storage.Interface {
+	return newStore(c, false, pagingEnabled, codec, prefix, transformer)
+}
+
+func newStore(c *clientv3.Client, quorumRead, pagingEnabled bool, codec runtime.Codec, prefix string, transformer value.Transformer) *store {
+	versioner := etcd.APIObjectVersioner{}
+	result := &store{
+		client:        c,
+		codec:         codec,
+		versioner:     versioner,
+		transformer:   transformer,
+		pagingEnabled: pagingEnabled,
+		// for compatibility with etcd2 impl.
+		// no-op for default prefix of '/registry'.
+		// keeps compatibility with etcd2 impl for custom prefixes that don't start with '/'
+		pathPrefix: path.Join("/", prefix),
+		watcher:    newWatcher(c, codec, versioner, transformer),
+	}
+	if !quorumRead {
+		// In case of non-quorum reads, we can set WithSerializable()
+		// options for all Get operations.
+		result.getOps = append(result.getOps, clientv3.WithSerializable())
+	}
+	return result
+}
+
+// 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, resourceVersion string, out runtime.Object, ignoreNotFound bool) error {
+	key = path.Join(s.pathPrefix, key)
+	getResp, err := s.client.KV.Get(ctx, key, s.getOps...)
+	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]
+
+	data, _, err := s.transformer.TransformFromStorage(kv.Value, authenticatedDataString(key))
+	if err != nil {
+		return storage.NewInternalError(err.Error())
+	}
+
+	return decode(s.codec, s.versioner, data, 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")
+	}
+	if err := s.versioner.PrepareObjectForStorage(obj); err != nil {
+		return fmt.Errorf("PrepareObjectForStorage failed: %v", err)
+	}
+	data, err := runtime.Encode(s.codec, obj)
+	if err != nil {
+		return err
+	}
+	key = path.Join(s.pathPrefix, key)
+
+	opts, err := s.ttlOpts(ctx, int64(ttl))
+	if err != nil {
+		return err
+	}
+
+	newData, err := s.transformer.TransformToStorage(data, authenticatedDataString(key))
+	if err != nil {
+		return storage.NewInternalError(err.Error())
+	}
+
+	txnResp, err := s.client.KV.Txn(ctx).If(
+		notFound(key),
+	).Then(
+		clientv3.OpPut(key, string(newData), 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, preconditions *storage.Preconditions) error {
+	v, err := conversion.EnforcePtr(out)
+	if err != nil {
+		panic("unable to convert output object to pointer")
+	}
+	key = path.Join(s.pathPrefix, key)
+	if preconditions == nil {
+		return s.unconditionalDelete(ctx, key, out)
+	}
+	return s.conditionalDelete(ctx, key, out, v, preconditions)
+}
+
+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]
+	data, _, err := s.transformer.TransformFromStorage(kv.Value, authenticatedDataString(key))
+	if err != nil {
+		return storage.NewInternalError(err.Error())
+	}
+	return decode(s.codec, s.versioner, data, out, kv.ModRevision)
+}
+
+func (s *store) conditionalDelete(ctx context.Context, key string, out runtime.Object, v reflect.Value, preconditions *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, preconditions, 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,
+	preconditions *storage.Preconditions, tryUpdate storage.UpdateFunc, suggestion ...runtime.Object) error {
+	trace := utiltrace.New(fmt.Sprintf("GuaranteedUpdate etcd3: %s", reflect.TypeOf(out).String()))
+	defer trace.LogIfLong(500 * time.Millisecond)
+
+	v, err := conversion.EnforcePtr(out)
+	if err != nil {
+		panic("unable to convert output object to pointer")
+	}
+	key = path.Join(s.pathPrefix, key)
+
+	getCurrentState := func() (*objState, error) {
+		getResp, err := s.client.KV.Get(ctx, key, s.getOps...)
+		if err != nil {
+			return nil, err
+		}
+		return s.getState(getResp, key, v, ignoreNotFound)
+	}
+
+	var origState *objState
+	var mustCheckData bool
+	if len(suggestion) == 1 && suggestion[0] != nil {
+		origState, err = s.getStateFromObject(suggestion[0])
+		if err != nil {
+			return err
+		}
+		mustCheckData = true
+	} else {
+		origState, err = getCurrentState()
+		if err != nil {
+			return err
+		}
+	}
+	trace.Step("initial value restored")
+
+	transformContext := authenticatedDataString(key)
+	for {
+		if err := checkPreconditions(key, preconditions, origState.obj); err != nil {
+			return err
+		}
+
+		ret, ttl, err := s.updateState(origState, tryUpdate)
+		if err != nil {
+			// It's possible we were working with stale data
+			if mustCheckData && apierrors.IsConflict(err) {
+				// Actually fetch
+				origState, err = getCurrentState()
+				if err != nil {
+					return err
+				}
+				mustCheckData = false
+				// Retry
+				continue
+			}
+
+			return err
+		}
+
+		data, err := runtime.Encode(s.codec, ret)
+		if err != nil {
+			return err
+		}
+		if !origState.stale && bytes.Equal(data, origState.data) {
+			// if we skipped the original Get in this loop, we must refresh from
+			// etcd in order to be sure the data in the store is equivalent to
+			// our desired serialization
+			if mustCheckData {
+				origState, err = getCurrentState()
+				if err != nil {
+					return err
+				}
+				mustCheckData = false
+				if !bytes.Equal(data, origState.data) {
+					// original data changed, restart loop
+					continue
+				}
+			}
+			// recheck that the data from etcd is not stale before short-circuiting a write
+			if !origState.stale {
+				return decode(s.codec, s.versioner, origState.data, out, origState.rev)
+			}
+		}
+
+		newData, err := s.transformer.TransformToStorage(data, transformContext)
+		if err != nil {
+			return storage.NewInternalError(err.Error())
+		}
+
+		opts, err := s.ttlOpts(ctx, int64(ttl))
+		if err != nil {
+			return err
+		}
+		trace.Step("Transaction prepared")
+
+		txnResp, err := s.client.KV.Txn(ctx).If(
+			clientv3.Compare(clientv3.ModRevision(key), "=", origState.rev),
+		).Then(
+			clientv3.OpPut(key, string(newData), opts...),
+		).Else(
+			clientv3.OpGet(key),
+		).Commit()
+		if err != nil {
+			return err
+		}
+		trace.Step("Transaction committed")
+		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)
+			origState, err = s.getState(getResp, key, v, ignoreNotFound)
+			if err != nil {
+				return err
+			}
+			trace.Step("Retry value restored")
+			mustCheckData = false
+			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, resourceVersion string, pred storage.SelectionPredicate, listObj runtime.Object) error {
+	listPtr, err := meta.GetItemsPtr(listObj)
+	if err != nil {
+		return err
+	}
+	v, err := conversion.EnforcePtr(listPtr)
+	if err != nil || v.Kind() != reflect.Slice {
+		panic("need ptr to slice")
+	}
+
+	key = path.Join(s.pathPrefix, key)
+	getResp, err := s.client.KV.Get(ctx, key, s.getOps...)
+	if err != nil {
+		return err
+	}
+
+	if len(getResp.Kvs) > 0 {
+		data, _, err := s.transformer.TransformFromStorage(getResp.Kvs[0].Value, authenticatedDataString(key))
+		if err != nil {
+			return storage.NewInternalError(err.Error())
+		}
+		if err := appendListItem(v, data, uint64(getResp.Kvs[0].ModRevision), pred, s.codec, s.versioner); err != nil {
+			return err
+		}
+	}
+	// update version with cluster level revision
+	return s.versioner.UpdateList(listObj, uint64(getResp.Header.Revision), "")
+}
+
+func (s *store) Count(key string) (int64, error) {
+	key = path.Join(s.pathPrefix, key)
+	getResp, err := s.client.KV.Get(context.Background(), key, clientv3.WithRange(clientv3.GetPrefixRangeEnd(key)), clientv3.WithCountOnly())
+	if err != nil {
+		return 0, err
+	}
+	return getResp.Count, nil
+}
+
+// continueToken is a simple structured object for encoding the state of a continue token.
+// TODO: if we change the version of the encoded from, we can't start encoding the new version
+// until all other servers are upgraded (i.e. we need to support rolling schema)
+// This is a public API struct and cannot change.
+type continueToken struct {
+	APIVersion      string `json:"v"`
+	ResourceVersion int64  `json:"rv"`
+	StartKey        string `json:"start"`
+}
+
+// parseFrom transforms an encoded predicate from into a versioned struct.
+// TODO: return a typed error that instructs clients that they must relist
+func decodeContinue(continueValue, keyPrefix string) (fromKey string, rv int64, err error) {
+	data, err := base64.RawURLEncoding.DecodeString(continueValue)
+	if err != nil {
+		return "", 0, fmt.Errorf("continue key is not valid: %v", err)
+	}
+	var c continueToken
+	if err := json.Unmarshal(data, &c); err != nil {
+		return "", 0, fmt.Errorf("continue key is not valid: %v", err)
+	}
+	switch c.APIVersion {
+	case "meta.k8s.io/v1":
+		if c.ResourceVersion == 0 {
+			return "", 0, fmt.Errorf("continue key is not valid: incorrect encoded start resourceVersion (version meta.k8s.io/v1)")
+		}
+		if len(c.StartKey) == 0 {
+			return "", 0, fmt.Errorf("continue key is not valid: encoded start key empty (version meta.k8s.io/v1)")
+		}
+		// defend against path traversal attacks by clients - path.Clean will ensure that startKey cannot
+		// be at a higher level of the hierarchy, and so when we append the key prefix we will end up with
+		// continue start key that is fully qualified and cannot range over anything less specific than
+		// keyPrefix.
+		key := c.StartKey
+		if !strings.HasPrefix(key, "/") {
+			key = "/" + key
+		}
+		cleaned := path.Clean(key)
+		if cleaned != key {
+			return "", 0, fmt.Errorf("continue key is not valid: %s", c.StartKey)
+		}
+		return keyPrefix + cleaned[1:], c.ResourceVersion, nil
+	default:
+		return "", 0, fmt.Errorf("continue key is not valid: server does not recognize this encoded version %q", c.APIVersion)
+	}
+}
+
+// encodeContinue returns a string representing the encoded continuation of the current query.
+func encodeContinue(key, keyPrefix string, resourceVersion int64) (string, error) {
+	nextKey := strings.TrimPrefix(key, keyPrefix)
+	if nextKey == key {
+		return "", fmt.Errorf("unable to encode next field: the key and key prefix do not match")
+	}
+	out, err := json.Marshal(&continueToken{APIVersion: "meta.k8s.io/v1", ResourceVersion: resourceVersion, StartKey: nextKey})
+	if err != nil {
+		return "", err
+	}
+	return base64.RawURLEncoding.EncodeToString(out), nil
+}
+
+// List implements storage.Interface.List.
+func (s *store) List(ctx context.Context, key, resourceVersion string, pred storage.SelectionPredicate, listObj runtime.Object) error {
+	listPtr, err := meta.GetItemsPtr(listObj)
+	if err != nil {
+		return err
+	}
+	v, err := conversion.EnforcePtr(listPtr)
+	if err != nil || v.Kind() != reflect.Slice {
+		panic("need ptr to slice")
+	}
+
+	if s.pathPrefix != "" {
+		key = path.Join(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 += "/"
+	}
+	keyPrefix := key
+
+	// set the appropriate clientv3 options to filter the returned data set
+	var paging bool
+	options := make([]clientv3.OpOption, 0, 4)
+	if s.pagingEnabled && pred.Limit > 0 {
+		paging = true
+		options = append(options, clientv3.WithLimit(pred.Limit))
+	}
+
+	var returnedRV int64
+	switch {
+	case s.pagingEnabled && len(pred.Continue) > 0:
+		continueKey, continueRV, err := decodeContinue(pred.Continue, keyPrefix)
+		if err != nil {
+			return apierrors.NewBadRequest(fmt.Sprintf("invalid continue token: %v", err))
+		}
+
+		if len(resourceVersion) > 0 && resourceVersion != "0" {
+			return apierrors.NewBadRequest("specifying resource version is not allowed when using continue")
+		}
+
+		rangeEnd := clientv3.GetPrefixRangeEnd(keyPrefix)
+		options = append(options, clientv3.WithRange(rangeEnd))
+		key = continueKey
+
+		options = append(options, clientv3.WithRev(continueRV))
+		returnedRV = continueRV
+
+	case s.pagingEnabled && pred.Limit > 0:
+		if len(resourceVersion) > 0 {
+			fromRV, err := s.versioner.ParseListResourceVersion(resourceVersion)
+			if err != nil {
+				return apierrors.NewBadRequest(fmt.Sprintf("invalid resource version: %v", err))
+			}
+			if fromRV > 0 {
+				options = append(options, clientv3.WithRev(int64(fromRV)))
+			}
+			returnedRV = int64(fromRV)
+		}
+
+		rangeEnd := clientv3.GetPrefixRangeEnd(keyPrefix)
+		options = append(options, clientv3.WithRange(rangeEnd))
+
+	default:
+		if len(resourceVersion) > 0 {
+			fromRV, err := s.versioner.ParseListResourceVersion(resourceVersion)
+			if err != nil {
+				return apierrors.NewBadRequest(fmt.Sprintf("invalid resource version: %v", err))
+			}
+			if fromRV > 0 {
+				options = append(options, clientv3.WithRev(int64(fromRV)))
+			}
+			returnedRV = int64(fromRV)
+		}
+
+		options = append(options, clientv3.WithPrefix())
+	}
+
+	// loop until we have filled the requested limit from etcd or there are no more results
+	var lastKey []byte
+	var hasMore bool
+	for {
+		getResp, err := s.client.KV.Get(ctx, key, options...)
+		if err != nil {
+			return interpretListError(err, len(pred.Continue) > 0)
+		}
+		hasMore = getResp.More
+
+		if len(getResp.Kvs) == 0 && getResp.More {
+			return fmt.Errorf("no results were found, but etcd indicated there were more values remaining")
+		}
+
+		// avoid small allocations for the result slice, since this can be called in many
+		// different contexts and we don't know how significantly the result will be filtered
+		if pred.Empty() {
+			growSlice(v, len(getResp.Kvs))
+		} else {
+			growSlice(v, 2048, len(getResp.Kvs))
+		}
+
+		// take items from the response until the bucket is full, filtering as we go
+		for _, kv := range getResp.Kvs {
+			if paging && int64(v.Len()) >= pred.Limit {
+				hasMore = true
+				break
+			}
+			lastKey = kv.Key
+
+			data, _, err := s.transformer.TransformFromStorage(kv.Value, authenticatedDataString(kv.Key))
+			if err != nil {
+				utilruntime.HandleError(fmt.Errorf("unable to transform key %q: %v", kv.Key, err))
+				continue
+			}
+
+			if err := appendListItem(v, data, uint64(kv.ModRevision), pred, s.codec, s.versioner); err != nil {
+				return err
+			}
+		}
+
+		// indicate to the client which resource version was returned
+		if returnedRV == 0 {
+			returnedRV = getResp.Header.Revision
+		}
+
+		// no more results remain or we didn't request paging
+		if !hasMore || !paging {
+			break
+		}
+		// we're paging but we have filled our bucket
+		if int64(v.Len()) >= pred.Limit {
+			break
+		}
+		key = string(lastKey) + "\x00"
+	}
+
+	// instruct the client to begin querying from immediately after the last key we returned
+	// we never return a key that the client wouldn't be allowed to see
+	if hasMore {
+		// we want to start immediately after the last key
+		next, err := encodeContinue(string(lastKey)+"\x00", keyPrefix, returnedRV)
+		if err != nil {
+			return err
+		}
+		return s.versioner.UpdateList(listObj, uint64(returnedRV), next)
+	}
+
+	// no continuation
+	return s.versioner.UpdateList(listObj, uint64(returnedRV), "")
+}
+
+// growSlice takes a slice value and grows its capacity up
+// to the maximum of the passed sizes or maxCapacity, whichever
+// is smaller. Above maxCapacity decisions about allocation are left
+// to the Go runtime on append. This allows a caller to make an
+// educated guess about the potential size of the total list while
+// still avoiding overly aggressive initial allocation. If sizes
+// is empty maxCapacity will be used as the size to grow.
+func growSlice(v reflect.Value, maxCapacity int, sizes ...int) {
+	cap := v.Cap()
+	max := cap
+	for _, size := range sizes {
+		if size > max {
+			max = size
+		}
+	}
+	if len(sizes) == 0 || max > maxCapacity {
+		max = maxCapacity
+	}
+	if max <= cap {
+		return
+	}
+	if v.Len() > 0 {
+		extra := reflect.MakeSlice(v.Type(), 0, max)
+		reflect.Copy(extra, v)
+		v.Set(extra)
+	} else {
+		extra := reflect.MakeSlice(v.Type(), 0, max)
+		v.Set(extra)
+	}
+}
+
+// Watch implements storage.Interface.Watch.
+func (s *store) Watch(ctx context.Context, key string, resourceVersion string, pred storage.SelectionPredicate) (watch.Interface, error) {
+	return s.watch(ctx, key, resourceVersion, pred, false)
+}
+
+// WatchList implements storage.Interface.WatchList.
+func (s *store) WatchList(ctx context.Context, key string, resourceVersion string, pred storage.SelectionPredicate) (watch.Interface, error) {
+	return s.watch(ctx, key, resourceVersion, pred, true)
+}
+
+func (s *store) watch(ctx context.Context, key string, rv string, pred storage.SelectionPredicate, recursive bool) (watch.Interface, error) {
+	rev, err := s.versioner.ParseWatchResourceVersion(rv)
+	if err != nil {
+		return nil, err
+	}
+	key = path.Join(s.pathPrefix, key)
+	return s.watcher.Watch(ctx, key, int64(rev), recursive, pred)
+}
+
+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 {
+		data, stale, err := s.transformer.TransformFromStorage(getResp.Kvs[0].Value, authenticatedDataString(key))
+		if err != nil {
+			return nil, storage.NewInternalError(err.Error())
+		}
+		state.rev = getResp.Kvs[0].ModRevision
+		state.meta.ResourceVersion = uint64(state.rev)
+		state.data = data
+		state.stale = stale
+		if err := decode(s.codec, s.versioner, state.data, state.obj, state.rev); err != nil {
+			return nil, err
+		}
+	}
+	return state, nil
+}
+
+func (s *store) getStateFromObject(obj runtime.Object) (*objState, error) {
+	state := &objState{
+		obj:  obj,
+		meta: &storage.ResponseMeta{},
+	}
+
+	rv, err := s.versioner.ObjectResourceVersion(obj)
+	if err != nil {
+		return nil, fmt.Errorf("couldn't get resource version: %v", err)
+	}
+	state.rev = int64(rv)
+	state.meta.ResourceVersion = uint64(state.rev)
+
+	// Compute the serialized form - for that we need to temporarily clean
+	// its resource version field (those are not stored in etcd).
+	if err := s.versioner.PrepareObjectForStorage(obj); err != nil {
+		return nil, fmt.Errorf("PrepareObjectForStorage failed: %v", err)
+	}
+	state.data, err = runtime.Encode(s.codec, obj)
+	if err != nil {
+		return nil, err
+	}
+	s.versioner.UpdateObject(state.obj, uint64(rv))
+	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
+	}
+
+	if err := s.versioner.PrepareObjectForStorage(ret); err != nil {
+		return nil, 0, fmt.Errorf("PrepareObjectForStorage 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
+}
+
+// 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
+}
+
+// appendListItem decodes and appends the object (if it passes filter) to v, which must be a slice.
+func appendListItem(v reflect.Value, data []byte, rev uint64, pred storage.SelectionPredicate, codec runtime.Codec, versioner storage.Versioner) error {
+	obj, _, err := codec.Decode(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, rev)
+	if matched, err := pred.Matches(obj); err == nil && matched {
+		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 := meta.Accessor(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.GetUID() {
+		errMsg := fmt.Sprintf("Precondition failed: UID in precondition: %v, UID in object meta: %v", *preconditions.UID, objMeta.GetUID())
+		return storage.NewInvalidObjError(key, errMsg)
+	}
+	return nil
+}
+
+func notFound(key string) clientv3.Cmp {
+	return clientv3.Compare(clientv3.ModRevision(key), "=", 0)
+}

vendor/k8s.io/apiserver/pkg/storage/etcd3/watcher.go

@@ -0,0 +1,402 @@
+/*
+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 (
+	"context"
+	"errors"
+	"fmt"
+	"os"
+	"strconv"
+	"strings"
+	"sync"
+
+	apierrs "k8s.io/apimachinery/pkg/api/errors"
+	"k8s.io/apimachinery/pkg/runtime"
+	"k8s.io/apimachinery/pkg/watch"
+	"k8s.io/apiserver/pkg/storage"
+	"k8s.io/apiserver/pkg/storage/value"
+
+	"github.com/coreos/etcd/clientv3"
+	"github.com/golang/glog"
+)
+
+const (
+	// We have set a buffer in order to reduce times of context switches.
+	incomingBufSize = 100
+	outgoingBufSize = 100
+)
+
+// fatalOnDecodeError is used during testing to panic the server if watcher encounters a decoding error
+var fatalOnDecodeError = false
+
+// errTestingDecode is the only error that testingDeferOnDecodeError catches during a panic
+var errTestingDecode = errors.New("sentinel error only used during testing to indicate watch decoding error")
+
+// testingDeferOnDecodeError is used during testing to recover from a panic caused by errTestingDecode, all other values continue to panic
+func testingDeferOnDecodeError() {
+	if r := recover(); r != nil && r != errTestingDecode {
+		panic(r)
+	}
+}
+
+func init() {
+	// check to see if we are running in a test environment
+	fatalOnDecodeError, _ = strconv.ParseBool(os.Getenv("KUBE_PANIC_WATCH_DECODE_ERROR"))
+}
+
+type watcher struct {
+	client      *clientv3.Client
+	codec       runtime.Codec
+	versioner   storage.Versioner
+	transformer value.Transformer
+}
+
+// watchChan implements watch.Interface.
+type watchChan struct {
+	watcher           *watcher
+	key               string
+	initialRev        int64
+	recursive         bool
+	internalPred      storage.SelectionPredicate
+	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, transformer value.Transformer) *watcher {
+	return &watcher{
+		client:      client,
+		codec:       codec,
+		versioner:   versioner,
+		transformer: transformer,
+	}
+}
+
+// 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.
+// pred must be non-nil. Only if pred matches the change, it will be returned.
+func (w *watcher) Watch(ctx context.Context, key string, rev int64, recursive bool, pred storage.SelectionPredicate) (watch.Interface, error) {
+	if recursive && !strings.HasSuffix(key, "/") {
+		key += "/"
+	}
+	wc := w.createWatchChan(ctx, key, rev, recursive, pred)
+	go wc.run()
+	return wc, nil
+}
+
+func (w *watcher) createWatchChan(ctx context.Context, key string, rev int64, recursive bool, pred storage.SelectionPredicate) *watchChan {
+	wc := &watchChan{
+		watcher:           w,
+		key:               key,
+		initialRev:        rev,
+		recursive:         recursive,
+		internalPred:      pred,
+		incomingEventChan: make(chan *event, incomingBufSize),
+		resultChan:        make(chan watch.Event, outgoingBufSize),
+		errChan:           make(chan error, 1),
+	}
+	if pred.Empty() {
+		// The filter doesn't filter out any object.
+		wc.internalPred = storage.Everything
+	}
+	wc.ctx, wc.cancel = context.WithCancel(ctx)
+	return wc
+}
+
+func (wc *watchChan) run() {
+	watchClosedCh := make(chan struct{})
+	go wc.startWatching(watchClosedCh)
+
+	var resultChanWG sync.WaitGroup
+	resultChanWG.Add(1)
+	go wc.processEvent(&resultChanWG)
+
+	select {
+	case err := <-wc.errChan:
+		if err == context.Canceled {
+			break
+		}
+		errResult := transformErrorToEvent(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
+			}
+		}
+	case <-watchClosedCh:
+	case <-wc.ctx.Done(): // user cancel
+	}
+
+	// We use wc.ctx to reap all goroutines. Under whatever condition, we should stop them all.
+	// It's fine to double cancel.
+	wc.cancel()
+
+	// we need to wait until resultChan wouldn't be used 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.
+// All events sent will have isCreated=true
+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(watchClosedCh chan struct{}) {
+	if wc.initialRev == 0 {
+		if err := wc.sync(); err != nil {
+			glog.Errorf("failed to sync with latest state: %v", err)
+			wc.sendError(err)
+			return
+		}
+	}
+	opts := []clientv3.OpOption{clientv3.WithRev(wc.initialRev + 1), clientv3.WithPrevKV()}
+	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 {
+			err := wres.Err()
+			// If there is an error on server (e.g. compaction), the channel will return it before closed.
+			glog.Errorf("watch chan error: %v", err)
+			wc.sendError(err)
+			return
+		}
+		for _, e := range wres.Events {
+			wc.sendEvent(parseEvent(e))
+		}
+	}
+	// When we come to this point, it's only possible that client side ends the watch.
+	// e.g. cancel the context, close the client.
+	// If this watch chan is broken and context isn't cancelled, other goroutines will still hang.
+	// We should notify the main thread that this goroutine has exited.
+	close(watchClosedCh)
+}
+
+// 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 len(wc.resultChan) == outgoingBufSize {
+				glog.Warningf("Fast watcher, slow processing. Number of buffered events: %d."+
+					"Probably caused by slow dispatching events to watchers", outgoingBufSize)
+			}
+			// 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
+		}
+	}
+}
+
+func (wc *watchChan) filter(obj runtime.Object) bool {
+	if wc.internalPred.Empty() {
+		return true
+	}
+	matched, err := wc.internalPred.Matches(obj)
+	return err == nil && matched
+}
+
+func (wc *watchChan) acceptAll() bool {
+	return wc.internalPred.Empty()
+}
+
+// transform transforms an event into a result for user if not filtered.
+func (wc *watchChan) transform(e *event) (res *watch.Event) {
+	curObj, oldObj, err := wc.prepareObjs(e)
+	if err != nil {
+		glog.Errorf("failed to prepare current and previous objects: %v", err)
+		wc.sendError(err)
+		return nil
+	}
+
+	switch {
+	case e.isDeleted:
+		if !wc.filter(oldObj) {
+			return nil
+		}
+		res = &watch.Event{
+			Type:   watch.Deleted,
+			Object: oldObj,
+		}
+	case e.isCreated:
+		if !wc.filter(curObj) {
+			return nil
+		}
+		res = &watch.Event{
+			Type:   watch.Added,
+			Object: curObj,
+		}
+	default:
+		if wc.acceptAll() {
+			res = &watch.Event{
+				Type:   watch.Modified,
+				Object: curObj,
+			}
+			return res
+		}
+		curObjPasses := wc.filter(curObj)
+		oldObjPasses := wc.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 transformErrorToEvent(err error) *watch.Event {
+	err = interpretWatchError(err)
+	if _, ok := err.(apierrs.APIStatus); !ok {
+		err = apierrs.NewInternalError(err)
+	}
+	status := err.(apierrs.APIStatus).Status()
+	return &watch.Event{
+		Type:   watch.Error,
+		Object: &status,
+	}
+}
+
+func (wc *watchChan) sendError(err error) {
+	select {
+	case wc.errChan <- err:
+	case <-wc.ctx.Done():
+	}
+}
+
+func (wc *watchChan) sendEvent(e *event) {
+	if len(wc.incomingEventChan) == incomingBufSize {
+		glog.Warningf("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 (wc *watchChan) prepareObjs(e *event) (curObj runtime.Object, oldObj runtime.Object, err error) {
+	if !e.isDeleted {
+		data, _, err := wc.watcher.transformer.TransformFromStorage(e.value, authenticatedDataString(e.key))
+		if err != nil {
+			return nil, nil, err
+		}
+		curObj, err = decodeObj(wc.watcher.codec, wc.watcher.versioner, data, e.rev)
+		if err != nil {
+			return nil, nil, err
+		}
+	}
+	// We need to decode prevValue, only if this is deletion event or
+	// the underlying filter doesn't accept all objects (otherwise we
+	// know that the filter for previous object will return true and
+	// we need the object only to compute whether it was filtered out
+	// before).
+	if len(e.prevValue) > 0 && (e.isDeleted || !wc.acceptAll()) {
+		data, _, err := wc.watcher.transformer.TransformFromStorage(e.prevValue, authenticatedDataString(e.key))
+		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.
+		oldObj, err = decodeObj(wc.watcher.codec, wc.watcher.versioner, data, 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, err error) {
+	obj, err := runtime.Decode(codec, []byte(data))
+	if err != nil {
+		if fatalOnDecodeError {
+			// catch watch decode error iff we caused it on
+			// purpose during a unit test
+			defer testingDeferOnDecodeError()
+			// we are running in a test environment and thus an
+			// error here is due to a coder mistake if the defer
+			// does not catch it
+			panic(err)
+		}
+		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
+}