// Dispatch the command to leader
func dispatch(c Command, w http.ResponseWriter, req *http.Request, etcd bool) error {
if r.State() == raft.Leader {
if body, err := r.Do(c); err != nil {
return err
} else {
if body == nil {
return etcdErr.NewError(300, "Empty result from raft")
} else {
body, _ := body.([]byte)
w.WriteHeader(http.StatusOK)
w.Write(body)
return nil
}
}
} else {
leader := r.Leader()
// current no leader
if leader == "" {
return etcdErr.NewError(300, "")
}
redirect(leader, etcd, w, req)
return nil
}
return etcdErr.NewError(300, "")
}
// Set the value of the key to the value if the given prevValue is equal to the value of the key
func (s *Store) TestAndSet(key string, prevValue string, value string, expireTime time.Time, index uint64) ([]byte, error) {
s.mutex.Lock()
defer s.mutex.Unlock()
// Update stats
s.BasicStats.TestAndSets++
resp := s.internalGet(key)
if resp == nil {
if prevValue != "" {
errmsg := fmt.Sprintf("TestAndSet: key not found and previousValue is not empty %s:%s ", key, prevValue)
return nil, etcdErr.NewError(100, errmsg)
}
return s.internalSet(key, value, expireTime, index)
}
if resp.Value == prevValue {
// If test succeed, do set
return s.internalSet(key, value, expireTime, index)
} else {
// If fails, return err
return nil, etcdErr.NewError(101, fmt.Sprintf("TestAndSet: %s!=%s",
resp.Value, prevValue))
}
}
// Set the value of the key to the value if the given prevValue is equal to the value of the key
func (s *Store) TestAndSet(key string, prevValue string, value string, expireTime time.Time, index uint64) ([]byte, error) {
s.mutex.Lock()
defer s.mutex.Unlock()
// Update stats
s.BasicStats.TestAndSets++
resp := s.internalGet(key)
if resp == nil {
return nil, etcdErr.NewError(100, "testandset: "+key)
}
if resp.Value == prevValue {
// If test success, do set
return s.internalSet(key, value, expireTime, index)
} else {
// If fails, return err
return nil, etcdErr.NewError(101, fmt.Sprintf("TestAndSet: %s!=%s",
resp.Value, prevValue))
}
}
// InternalGet gets the node of the given nodePath.
func (s *store) internalGet(nodePath string) (*node, *etcdErr.Error) {
nodePath = path.Clean(path.Join("/", nodePath))
walkFunc := func(parent *node, name string) (*node, *etcdErr.Error) {
if !parent.IsDir() {
err := etcdErr.NewError(etcdErr.EcodeNotDir, parent.Path, s.CurrentIndex)
return nil, err
}
child, ok := parent.Children[name]
if ok {
return child, nil
}
return nil, etcdErr.NewError(etcdErr.EcodeKeyNotFound, path.Join(parent.Path, name), s.CurrentIndex)
}
f, err := s.walk(nodePath, walkFunc)
if err != nil {
return nil, err
}
return f, nil
}
// Watches a given key prefix for changes.
func WatchKeyHandler(w http.ResponseWriter, req *http.Request, s Server) error {
var err error
vars := mux.Vars(req)
key := "/" + vars["key"]
// Create a command to watch from a given index (default 0).
var sinceIndex uint64 = 0
if req.Method == "POST" {
sinceIndex, err = strconv.ParseUint(string(req.FormValue("index")), 10, 64)
if err != nil {
return etcdErr.NewError(203, "Watch From Index", s.Store().Index())
}
}
// Start the watcher on the store.
watcher, err := s.Store().Watch(key, false, false, sinceIndex)
if err != nil {
return etcdErr.NewError(500, key, s.Store().Index())
}
event := <-watcher.EventChan
// Convert event to a response and write to client.
w.WriteHeader(http.StatusOK)
if req.Method == "HEAD" {
return nil
}
b, _ := json.Marshal(event.Response(s.Store().Index()))
w.Write(b)
return nil
}
// Update updates the value/ttl of the node.
// If the node is a file, the value and the ttl can be updated.
// If the node is a directory, only the ttl can be updated.
func (s *store) Update(nodePath string, newValue string, expireTime time.Time) (*Event, error) {
var err *etcdErr.Error
s.worldLock.Lock()
defer s.worldLock.Unlock()
defer func() {
if err == nil {
s.Stats.Inc(UpdateSuccess)
reportWriteSuccess(Update)
return
}
s.Stats.Inc(UpdateFail)
reportWriteFailure(Update)
}()
nodePath = path.Clean(path.Join("/", nodePath))
// we do not allow the user to change "/"
if s.readonlySet.Contains(nodePath) {
return nil, etcdErr.NewError(etcdErr.EcodeRootROnly, "/", s.CurrentIndex)
}
currIndex, nextIndex := s.CurrentIndex, s.CurrentIndex+1
n, err := s.internalGet(nodePath)
if err != nil { // if the node does not exist, return error
return nil, err
}
if n.IsDir() && len(newValue) != 0 {
// if the node is a directory, we cannot update value to non-empty
return nil, etcdErr.NewError(etcdErr.EcodeNotFile, nodePath, currIndex)
}
e := newEvent(Update, nodePath, nextIndex, n.CreatedIndex)
e.EtcdIndex = nextIndex
e.PrevNode = n.Repr(false, false, s.clock)
eNode := e.Node
n.Write(newValue, nextIndex)
if n.IsDir() {
eNode.Dir = true
} else {
// copy the value for safety
newValueCopy := newValue
eNode.Value = &newValueCopy
}
// update ttl
n.UpdateTTL(expireTime)
eNode.Expiration, eNode.TTL = n.expirationAndTTL(s.clock)
s.WatcherHub.notify(e)
s.CurrentIndex = nextIndex
return e, nil
}
// Sets the value for a given key.
func SetKeyHandler(w http.ResponseWriter, req *http.Request, s Server) error {
vars := mux.Vars(req)
key := "/" + vars["key"]
req.ParseForm()
// Parse non-blank value.
value := req.Form.Get("value")
if len(value) == 0 {
return etcdErr.NewError(200, "Set", s.Store().Index())
}
// Convert time-to-live to an expiration time.
expireTime, err := store.TTL(req.Form.Get("ttl"))
if err != nil {
return etcdErr.NewError(202, "Set", s.Store().Index())
}
// If the "prevValue" is specified then test-and-set. Otherwise create a new key.
var c raft.Command
if prevValueArr, ok := req.Form["prevValue"]; ok {
if len(prevValueArr[0]) > 0 {
// test against previous value
c = s.Store().CommandFactory().CreateCompareAndSwapCommand(key, value, prevValueArr[0], 0, expireTime)
} else {
// test against existence
c = s.Store().CommandFactory().CreateCreateCommand(key, value, expireTime, false)
}
} else {
c = s.Store().CommandFactory().CreateSetCommand(key, value, expireTime)
}
return s.Dispatch(c, w, req)
}
// applyJoin attempts to join a machine to the cluster.
func applyJoin(c *JoinCommand, context raft.Context) (uint64, error) {
ps, _ := context.Server().Context().(*PeerServer)
commitIndex := context.CommitIndex()
// Make sure we're not getting a cached value from the registry.
ps.registry.Invalidate(c.Name)
// Check if the join command is from a previous peer, who lost all its previous log.
if peerURL, ok := ps.registry.PeerURL(c.Name); ok {
// If previous node restarts with different peer URL,
// update its information.
if peerURL != c.RaftURL {
log.Infof("Rejoin with %v instead of %v from %v", c.RaftURL, peerURL, c.Name)
if err := updatePeerURL(c, ps); err != nil {
return 0, err
}
}
if c.Name == context.Server().Name() {
ps.removedInLog = false
}
return commitIndex, nil
}
// Check if the join command adds an instance that collides with existing one on peer URL.
peerURLs := ps.registry.PeerURLs(ps.raftServer.Leader(), c.Name)
for _, peerURL := range peerURLs {
if peerURL == c.RaftURL {
log.Warnf("%v tries to join the cluster with existing URL %v", c.Name, c.EtcdURL)
return 0, etcdErr.NewError(etcdErr.EcodeExistingPeerAddr, c.EtcdURL, context.CommitIndex())
}
}
// Check peer number in the cluster
count := ps.registry.Count()
// ClusterConfig doesn't init until first machine is added
if count > 0 && count >= ps.ClusterConfig().ActiveSize {
log.Debug("Reject join request from ", c.Name)
return 0, etcdErr.NewError(etcdErr.EcodeNoMorePeer, "", context.CommitIndex())
}
// Add to shared peer registry.
ps.registry.Register(c.Name, c.RaftURL, c.EtcdURL)
// Add peer in raft
if err := context.Server().AddPeer(c.Name, ""); err != nil {
return 0, err
}
// Add peer stats
if c.Name != ps.RaftServer().Name() {
ps.followersStats.Followers[c.Name] = &raftFollowerStats{}
ps.followersStats.Followers[c.Name].Latency.Minimum = 1 << 63
}
if c.Name == context.Server().Name() {
ps.removedInLog = false
}
return commitIndex, nil
}
func (s *store) CompareAndSwap(nodePath string, prevValue string, prevIndex uint64,
value string, expireTime time.Time) (*Event, error) {
s.worldLock.Lock()
defer s.worldLock.Unlock()
nodePath = path.Clean(path.Join("/", nodePath))
// we do not allow the user to change "/"
if s.readonlySet.Contains(nodePath) {
return nil, etcdErr.NewError(etcdErr.EcodeRootROnly, "/", s.CurrentIndex)
}
n, err := s.internalGet(nodePath)
if err != nil {
s.Stats.Inc(CompareAndSwapFail)
reportWriteFailure(CompareAndSwap)
return nil, err
}
if n.IsDir() { // can only compare and swap file
s.Stats.Inc(CompareAndSwapFail)
reportWriteFailure(CompareAndSwap)
return nil, etcdErr.NewError(etcdErr.EcodeNotFile, nodePath, s.CurrentIndex)
}
// If both of the prevValue and prevIndex are given, we will test both of them.
// Command will be executed, only if both of the tests are successful.
if ok, which := n.Compare(prevValue, prevIndex); !ok {
cause := getCompareFailCause(n, which, prevValue, prevIndex)
s.Stats.Inc(CompareAndSwapFail)
reportWriteFailure(CompareAndSwap)
return nil, etcdErr.NewError(etcdErr.EcodeTestFailed, cause, s.CurrentIndex)
}
// update etcd index
s.CurrentIndex++
e := newEvent(CompareAndSwap, nodePath, s.CurrentIndex, n.CreatedIndex)
e.EtcdIndex = s.CurrentIndex
e.PrevNode = n.Repr(false, false, s.clock)
eNode := e.Node
// if test succeed, write the value
n.Write(value, s.CurrentIndex)
n.UpdateTTL(expireTime)
// copy the value for safety
valueCopy := value
eNode.Value = &valueCopy
eNode.Expiration, eNode.TTL = n.expirationAndTTL(s.clock)
s.WatcherHub.notify(e)
s.Stats.Inc(CompareAndSwapSuccess)
reportWriteSuccess(CompareAndSwap)
return e, nil
}
// Remove function remove the node.
func (n *node) Remove(dir, recursive bool, callback func(path string)) *etcdErr.Error {
if n.IsDir() {
if !dir {
// cannot delete a directory without recursive set to true
return etcdErr.NewError(etcdErr.EcodeNotFile, n.Path, n.store.CurrentIndex)
}
if len(n.Children) != 0 && !recursive {
// cannot delete a directory if it is not empty and the operation
// is not recursive
return etcdErr.NewError(etcdErr.EcodeDirNotEmpty, n.Path, n.store.CurrentIndex)
}
}
if !n.IsDir() { // key-value pair
_, name := path.Split(n.Path)
// find its parent and remove the node from the map
if n.Parent != nil && n.Parent.Children[name] == n {
delete(n.Parent.Children, name)
}
if callback != nil {
callback(n.Path)
}
if !n.IsPermanent() {
n.store.ttlKeyHeap.remove(n)
}
return nil
}
for _, child := range n.Children { // delete all children
child.Remove(true, true, callback)
}
// delete self
_, name := path.Split(n.Path)
if n.Parent != nil && n.Parent.Children[name] == n {
delete(n.Parent.Children, name)
if callback != nil {
callback(n.Path)
}
if !n.IsPermanent() {
n.store.ttlKeyHeap.remove(n)
}
}
return nil
}
// acquireHandler attempts to acquire a lock on the given key.
// The "key" parameter specifies the resource to lock.
// The "value" parameter specifies a value to associate with the lock.
// The "ttl" parameter specifies how long the lock will persist for.
// The "timeout" parameter specifies how long the request should wait for the lock.
func (h *handler) acquireHandler(w http.ResponseWriter, req *http.Request) error {
h.client.SyncCluster()
// Setup connection watcher.
closeNotifier, _ := w.(http.CloseNotifier)
closeChan := closeNotifier.CloseNotify()
stopChan := make(chan bool)
// Parse the lock "key".
vars := mux.Vars(req)
keypath := path.Join(prefix, vars["key"])
value := req.FormValue("value")
// Parse "timeout" parameter.
var timeout int
var err error
if req.FormValue("timeout") == "" {
timeout = -1
} else if timeout, err = strconv.Atoi(req.FormValue("timeout")); err != nil {
return etcdErr.NewError(etcdErr.EcodeTimeoutNaN, "Acquire", 0)
}
timeout = timeout + 1
// Parse TTL.
ttl, err := strconv.Atoi(req.FormValue("ttl"))
if err != nil {
return etcdErr.NewError(etcdErr.EcodeTTLNaN, "Acquire", 0)
}
// If node exists then just watch it. Otherwise create the node and watch it.
node, index, pos := h.findExistingNode(keypath, value)
if index > 0 {
if pos == 0 {
// If lock is already acquired then update the TTL.
h.client.Update(node.Key, node.Value, uint64(ttl))
} else {
// Otherwise watch until it becomes acquired (or errors).
err = h.watch(keypath, index, nil)
}
} else {
index, err = h.createNode(keypath, value, ttl, closeChan, stopChan)
}
// Stop all goroutines.
close(stopChan)
// Check for an error.
if err != nil {
return err
}
// Write response.
w.Write([]byte(strconv.Itoa(index)))
return nil
}
func (s *store) CompareAndDelete(nodePath string, prevValue string, prevIndex uint64) (*Event, error) {
var err *etcdErr.Error
s.worldLock.Lock()
defer s.worldLock.Unlock()
defer func() {
if err == nil {
s.Stats.Inc(CompareAndDeleteSuccess)
reportWriteSuccess(CompareAndDelete)
return
}
s.Stats.Inc(CompareAndDeleteFail)
reportWriteFailure(CompareAndDelete)
}()
nodePath = path.Clean(path.Join("/", nodePath))
n, err := s.internalGet(nodePath)
if err != nil { // if the node does not exist, return error
return nil, err
}
if n.IsDir() { // can only compare and delete file
return nil, etcdErr.NewError(etcdErr.EcodeNotFile, nodePath, s.CurrentIndex)
}
// If both of the prevValue and prevIndex are given, we will test both of them.
// Command will be executed, only if both of the tests are successful.
if ok, which := n.Compare(prevValue, prevIndex); !ok {
cause := getCompareFailCause(n, which, prevValue, prevIndex)
return nil, etcdErr.NewError(etcdErr.EcodeTestFailed, cause, s.CurrentIndex)
}
// update etcd index
s.CurrentIndex++
e := newEvent(CompareAndDelete, nodePath, s.CurrentIndex, n.CreatedIndex)
e.EtcdIndex = s.CurrentIndex
e.PrevNode = n.Repr(false, false, s.clock)
callback := func(path string) { // notify function
// notify the watchers with deleted set true
s.WatcherHub.notifyWatchers(e, path, true)
}
err = n.Remove(false, false, callback)
if err != nil {
return nil, err
}
s.WatcherHub.notify(e)
return e, nil
}
func (s *store) CompareAndSwap(nodePath string, prevValue string, prevIndex uint64,
value string, expireTime time.Time) (*Event, error) {
nodePath = path.Clean(path.Join("/", nodePath))
// we do not allow the user to change "/"
if nodePath == "/" {
return nil, etcdErr.NewError(etcdErr.EcodeRootROnly, "/", s.CurrentIndex)
}
s.worldLock.Lock()
defer s.worldLock.Unlock()
n, err := s.internalGet(nodePath)
if err != nil {
s.Stats.Inc(CompareAndSwapFail)
return nil, err
}
if n.IsDir() { // can only compare and swap file
s.Stats.Inc(CompareAndSwapFail)
return nil, etcdErr.NewError(etcdErr.EcodeNotFile, nodePath, s.CurrentIndex)
}
// If both of the prevValue and prevIndex are given, we will test both of them.
// Command will be executed, only if both of the tests are successful.
if !n.Compare(prevValue, prevIndex) {
cause := fmt.Sprintf("[%v != %v] [%v != %v]", prevValue, n.Value, prevIndex, n.ModifiedIndex)
s.Stats.Inc(CompareAndSwapFail)
return nil, etcdErr.NewError(etcdErr.EcodeTestFailed, cause, s.CurrentIndex)
}
// update etcd index
s.CurrentIndex++
e := newEvent(CompareAndSwap, nodePath, s.CurrentIndex, n.CreatedIndex)
e.PrevNode = n.Repr(false, false)
eNode := e.Node
// if test succeed, write the value
n.Write(value, s.CurrentIndex)
n.UpdateTTL(expireTime)
// copy the value for safety
valueCopy := ustrings.Clone(value)
eNode.Value = &valueCopy
eNode.Expiration, eNode.TTL = n.ExpirationAndTTL()
s.WatcherHub.notify(e)
s.Stats.Inc(CompareAndSwapSuccess)
return e, nil
}
// Update function updates the value/ttl of the node.
// If the node is a file, the value and the ttl can be updated.
// If the node is a directory, only the ttl can be updated.
func (s *store) Update(nodePath string, newValue string, expireTime time.Time) (*Event, error) {
nodePath = path.Clean(path.Join("/", nodePath))
// we do not allow the user to change "/"
if nodePath == "/" {
return nil, etcdErr.NewError(etcdErr.EcodeRootROnly, "/", s.CurrentIndex)
}
s.worldLock.Lock()
defer s.worldLock.Unlock()
currIndex, nextIndex := s.CurrentIndex, s.CurrentIndex+1
n, err := s.internalGet(nodePath)
if err != nil { // if the node does not exist, return error
s.Stats.Inc(UpdateFail)
return nil, err
}
e := newEvent(Update, nodePath, nextIndex, n.CreatedIndex)
eNode := e.Node
if len(newValue) != 0 {
if n.IsDir() {
// if the node is a directory, we cannot update value
s.Stats.Inc(UpdateFail)
return nil, etcdErr.NewError(etcdErr.EcodeNotFile, nodePath, currIndex)
}
eNode.PrevValue = n.Value
n.Write(newValue, nextIndex)
eNode.Value = newValue
} else {
// do not update value
eNode.Value = n.Value
}
// update ttl
n.UpdateTTL(expireTime)
eNode.Expiration, eNode.TTL = n.ExpirationAndTTL()
s.WatcherHub.notify(e)
s.Stats.Inc(UpdateSuccess)
s.CurrentIndex = nextIndex
return e, nil
}
// Join a server to the cluster
func (c *JoinCommandV1) Apply(context raft.Context) (interface{}, error) {
ps, _ := context.Server().Context().(*PeerServer)
b := make([]byte, 8)
binary.PutUvarint(b, context.CommitIndex())
// Make sure we're not getting a cached value from the registry.
ps.registry.Invalidate(c.Name)
// Check if the join command is from a previous peer, who lost all its previous log.
if peerURL, ok := ps.registry.PeerURL(c.Name); ok {
// If previous node restarts with different peer URL,
// update its information.
if peerURL != c.RaftURL {
log.Infof("Rejoin with %v instead of %v from %v", c.RaftURL, peerURL, c.Name)
if err := c.updatePeerURL(ps); err != nil {
return []byte{0}, err
}
}
return b, nil
}
// Check if the join command adds an instance that collides with existing one on peer URL.
peerURLs := ps.registry.PeerURLs(ps.raftServer.Leader(), c.Name)
for _, peerURL := range peerURLs {
if peerURL == c.RaftURL {
log.Warnf("%v tries to join the cluster with existing URL %v", c.Name, c.EtcdURL)
return []byte{0}, etcdErr.NewError(etcdErr.EcodeExistingPeerAddr, c.EtcdURL, context.CommitIndex())
}
}
// Check peer number in the cluster
if ps.registry.PeerCount() >= ps.ClusterConfig().ActiveSize {
log.Debug("Reject join request from ", c.Name)
return []byte{0}, etcdErr.NewError(etcdErr.EcodeNoMorePeer, "", context.CommitIndex())
}
// Add to shared peer registry.
ps.registry.RegisterPeer(c.Name, c.RaftURL, c.EtcdURL)
// Add peer in raft
err := context.Server().AddPeer(c.Name, "")
// Add peer stats
if c.Name != ps.RaftServer().Name() {
ps.followersStats.Followers[c.Name] = &raftFollowerStats{}
ps.followersStats.Followers[c.Name].Latency.Minimum = 1 << 63
}
return b, err
}
// renewLockHandler attempts to update the TTL on an existing lock.
// Returns a 200 OK if successful. Returns non-200 on error.
func (h *handler) renewLockHandler(w http.ResponseWriter, req *http.Request) error {
h.client.SyncCluster()
// Read the lock path.
vars := mux.Vars(req)
keypath := path.Join(prefix, vars["key"])
// Parse new TTL parameter.
ttl, err := strconv.Atoi(req.FormValue("ttl"))
if err != nil {
return etcdErr.NewError(etcdErr.EcodeTTLNaN, "Renew", 0)
}
// Read and set defaults for index and value.
index := req.FormValue("index")
value := req.FormValue("value")
if len(index) == 0 && len(value) == 0 {
return etcdErr.NewError(etcdErr.EcodeIndexOrValueRequired, "Renew", 0)
}
if len(index) == 0 {
// If index is not specified then look it up by value.
resp, err := h.client.Get(keypath, true, true)
if err != nil {
return err
}
nodes := lockNodes{resp.Node.Nodes}
node, _ := nodes.FindByValue(value)
if node == nil {
return etcdErr.NewError(etcdErr.EcodeKeyNotFound, "Renew", 0)
}
index = path.Base(node.Key)
} else if len(value) == 0 {
// If value is not specified then default it to the previous value.
resp, err := h.client.Get(path.Join(keypath, index), true, false)
if err != nil {
return err
}
value = resp.Node.Value
}
// Renew the lock, if it exists.
if _, err = h.client.Update(path.Join(keypath, index), value, uint64(ttl)); err != nil {
return err
}
return nil
}
// Add function adds a node to the receiver node.
// If the receiver is not a directory, a "Not A Directory" error will be returned.
// If there is an existing node with the same name under the directory, a "Already Exist"
// error will be returned
func (n *node) Add(child *node) *etcdErr.Error {
if !n.IsDir() {
return etcdErr.NewError(etcdErr.EcodeNotDir, "", n.store.CurrentIndex)
}
_, name := path.Split(child.Path)
if _, ok := n.Children[name]; ok {
return etcdErr.NewError(etcdErr.EcodeNodeExist, "", n.store.CurrentIndex)
}
n.Children[name] = child
return nil
}
// deleteHandler remove a given leader.
func (h *handler) deleteHandler(w http.ResponseWriter, req *http.Request) error {
vars := mux.Vars(req)
name := req.FormValue("name")
if name == "" {
return etcdErr.NewError(etcdErr.EcodeNameRequired, "Delete", 0)
}
// Proxy the request to the the lock service.
u, err := url.Parse(fmt.Sprintf("%s/mod/v2/lock/%s", h.addr, vars["key"]))
if err != nil {
return err
}
q := u.Query()
q.Set("value", name)
u.RawQuery = q.Encode()
r, err := http.NewRequest("DELETE", u.String(), nil)
if err != nil {
return err
}
// Read from the leader lock.
resp, err := h.client.Do(r)
if err != nil {
return err
}
defer resp.Body.Close()
w.WriteHeader(resp.StatusCode)
io.Copy(w, resp.Body)
return nil
}
// Handler to return the basic stats of etcd
func StatsHttpHandler(w http.ResponseWriter, req *http.Request) error {
option := req.URL.Path[len("/v1/stats/"):]
switch option {
case "self":
w.WriteHeader(http.StatusOK)
w.Write(r.Stats())
case "leader":
if r.State() == raft.Leader {
w.Write(r.PeerStats())
} else {
leader := r.Leader()
// current no leader
if leader == "" {
return etcdErr.NewError(300, "")
}
redirect(leader, true, w, req)
}
case "store":
w.WriteHeader(http.StatusOK)
w.Write(etcdStore.Stats())
}
return nil
}
// Adds a server handler to the router.
func (s *Server) handleFunc(r *mux.Router, path string, f func(http.ResponseWriter, *http.Request) error) *mux.Route {
// Wrap the standard HandleFunc interface to pass in the server reference.
return r.HandleFunc(path, func(w http.ResponseWriter, req *http.Request) {
if req.Method == "HEAD" {
w = &HEADResponseWriter{w}
}
// Log request.
log.Debugf("[recv] %s %s %s [%s]", req.Method, s.URL(), req.URL.Path, req.RemoteAddr)
// Forward request along if the server is a standby.
if s.peerServer.Mode() == StandbyMode {
if s.peerServer.standbyClientURL == "" {
w.Header().Set("Content-Type", "application/json")
etcdErr.NewError(402, "", 0).Write(w)
return
}
uhttp.Redirect(s.peerServer.standbyClientURL, w, req)
return
}
// Execute handler function and return error if necessary.
if err := f(w, req); err != nil {
if etcdErr, ok := err.(*etcdErr.Error); ok {
log.Debug("Return error: ", (*etcdErr).Error())
w.Header().Set("Content-Type", "application/json")
etcdErr.Write(w)
} else {
http.Error(w, err.Error(), http.StatusInternalServerError)
}
}
})
}
// Delete deletes the node at the given path.
// If the node is a directory, recursive must be true to delete it.
func (s *store) Delete(nodePath string, dir, recursive bool) (*Event, error) {
var err *etcdErr.Error
s.worldLock.Lock()
defer s.worldLock.Unlock()
defer func() {
if err == nil {
s.Stats.Inc(DeleteSuccess)
reportWriteSuccess(Delete)
return
}
s.Stats.Inc(DeleteFail)
reportWriteFailure(Delete)
}()
nodePath = path.Clean(path.Join("/", nodePath))
// we do not allow the user to change "/"
if s.readonlySet.Contains(nodePath) {
return nil, etcdErr.NewError(etcdErr.EcodeRootROnly, "/", s.CurrentIndex)
}
// recursive implies dir
if recursive == true {
dir = true
}
n, err := s.internalGet(nodePath)
if err != nil { // if the node does not exist, return error
return nil, err
}
nextIndex := s.CurrentIndex + 1
e := newEvent(Delete, nodePath, nextIndex, n.CreatedIndex)
e.EtcdIndex = nextIndex
e.PrevNode = n.Repr(false, false, s.clock)
eNode := e.Node
if n.IsDir() {
eNode.Dir = true
}
callback := func(path string) { // notify function
// notify the watchers with deleted set true
s.WatcherHub.notifyWatchers(e, path, true)
}
err = n.Remove(dir, recursive, callback)
if err != nil {
return nil, err
}
// update etcd index
s.CurrentIndex++
s.WatcherHub.notify(e)
return e, nil
}
// Join a server to the cluster
func (c *JoinCommand) Apply(raftServer *raft.Server) (interface{}, error) {
// check if the join command is from a previous machine, who lost all its previous log.
response, _ := etcdStore.RawGet(path.Join("_etcd/machines", c.Name))
b := make([]byte, 8)
binary.PutUvarint(b, raftServer.CommitIndex())
if response != nil {
return b, nil
}
// check machine number in the cluster
num := machineNum()
if num == maxClusterSize {
debug("Reject join request from ", c.Name)
return []byte{0}, etcdErr.NewError(103, "")
}
addNameToURL(c.Name, c.RaftVersion, c.RaftURL, c.EtcdURL)
// add peer in raft
err := raftServer.AddPeer(c.Name, "")
// add machine in etcd storage
key := path.Join("_etcd/machines", c.Name)
value := fmt.Sprintf("raft=%s&etcd=%s&raftVersion=%s", c.RaftURL, c.EtcdURL, c.RaftVersion)
etcdStore.Set(key, value, time.Unix(0, 0), raftServer.CommitIndex())
return b, err
}
// Join a server to the cluster
func (c *JoinCommand) Apply(server raft.Server) (interface{}, error) {
ps, _ := server.Context().(*PeerServer)
b := make([]byte, 8)
binary.PutUvarint(b, server.CommitIndex())
// Make sure we're not getting a cached value from the registry.
ps.registry.Invalidate(c.Name)
// Check if the join command is from a previous peer, who lost all its previous log.
if _, ok := ps.registry.ClientURL(c.Name); ok {
return b, nil
}
// Check peer number in the cluster
if ps.registry.Count() == ps.MaxClusterSize {
log.Debug("Reject join request from ", c.Name)
return []byte{0}, etcdErr.NewError(etcdErr.EcodeNoMorePeer, "", server.CommitIndex())
}
// Add to shared peer registry.
ps.registry.Register(c.Name, c.RaftURL, c.EtcdURL)
// Add peer in raft
err := server.AddPeer(c.Name, "")
// Add peer stats
if c.Name != ps.RaftServer().Name() {
ps.followersStats.Followers[c.Name] = &raftFollowerStats{}
ps.followersStats.Followers[c.Name].Latency.Minimum = 1 << 63
}
return b, err
}
// Read function gets the value of the node.
// If the receiver node is not a key-value pair, a "Not A File" error will be returned.
func (n *node) Read() (string, *etcdErr.Error) {
if n.IsDir() {
return "", etcdErr.NewError(etcdErr.EcodeNotFile, "", n.store.CurrentIndex)
}
return n.Value, nil
}
// scan function is enumerating events from the index in history and
// stops till the first point where the key has identified prefix
func (eh *EventHistory) scan(prefix string, index uint64) (*Event, *etcdErr.Error) {
eh.rwl.RLock()
defer eh.rwl.RUnlock()
// the index should locate after the event history's StartIndex
if index-eh.StartIndex < 0 {
return nil,
etcdErr.NewError(etcdErr.EcodeEventIndexCleared,
fmt.Sprintf("the requested history has been cleared [%v/%v]",
eh.StartIndex, index), 0)
}
// the index should locate before the size of the queue minus the duplicate count
if index > eh.LastIndex { // future index
return nil, nil
}
i := eh.Queue.Front
for {
e := eh.Queue.Events[i]
if strings.HasPrefix(e.Key, prefix) && index <= e.Index() { // make sure we bypass the smaller one
return e, nil
}
i = (i + 1) % eh.Queue.Capacity
if i > eh.Queue.back() {
return nil, nil
}
}
}
// getIndexHandler retrieves the current lock index.
// The "field" parameter specifies to read either the lock "index" or lock "value".
func (h *handler) getIndexHandler(w http.ResponseWriter, req *http.Request) error {
h.client.SyncCluster()
vars := mux.Vars(req)
keypath := path.Join(prefix, vars["key"])
field := req.FormValue("field")
if len(field) == 0 {
field = "value"
}
// Read all indices.
resp, err := h.client.Get(keypath, true, true)
if err != nil {
return err
}
nodes := lockNodes{resp.Node.Nodes}
// Write out the requested field.
if node := nodes.First(); node != nil {
switch field {
case "index":
w.Write([]byte(path.Base(node.Key)))
case "value":
w.Write([]byte(node.Value))
default:
return etcdErr.NewError(etcdErr.EcodeInvalidField, "Get", 0)
}
}
return nil
}
func TestWriteError(t *testing.T) {
// nil error should not panic
rec := httptest.NewRecorder()
r := new(http.Request)
writeError(rec, r, nil)
h := rec.Header()
if len(h) > 0 {
t.Fatalf("unexpected non-empty headers: %#v", h)
}
b := rec.Body.String()
if len(b) > 0 {
t.Fatalf("unexpected non-empty body: %q", b)
}
tests := []struct {
err error
wcode int
wi string
}{
{
etcdErr.NewError(etcdErr.EcodeKeyNotFound, "/foo/bar", 123),
http.StatusNotFound,
"123",
},
{
etcdErr.NewError(etcdErr.EcodeTestFailed, "/foo/bar", 456),
http.StatusPreconditionFailed,
"456",
},
{
err: errors.New("something went wrong"),
wcode: http.StatusInternalServerError,
},
}
for i, tt := range tests {
rw := httptest.NewRecorder()
writeError(rw, r, tt.err)
if code := rw.Code; code != tt.wcode {
t.Errorf("#%d: code=%d, want %d", i, code, tt.wcode)
}
if idx := rw.Header().Get("X-Etcd-Index"); idx != tt.wi {
t.Errorf("#%d: X-Etcd-Index=%q, want %q", i, idx, tt.wi)
}
}
}
// Add function adds a node to the receiver node.
// If the receiver is not a directory, a "Not A Directory" error will be returned.
// If there is a existing node with the same name under the directory, a "Already Exist"
// error will be returned
func (n *Node) Add(child *Node) *etcdErr.Error {
if !n.IsDir() {
return etcdErr.NewError(etcdErr.EcodeNotDir, "", n.store.Index())
}
_, name := path.Split(child.Path)
_, ok := n.Children[name]
if ok {
return etcdErr.NewError(etcdErr.EcodeNodeExist, "", n.store.Index())
}
n.Children[name] = child
return nil
}
func (s *store) CompareAndSwap(nodePath string, prevValue string, prevIndex uint64,
value string, expireTime time.Time) (*Event, error) {
nodePath = path.Clean(path.Join("/", nodePath))
s.worldLock.Lock()
defer s.worldLock.Unlock()
n, err := s.internalGet(nodePath)
if err != nil {
s.Stats.Inc(CompareAndSwapFail)
return nil, err
}
if n.IsDir() { // can only test and set file
s.Stats.Inc(CompareAndSwapFail)
return nil, etcdErr.NewError(etcdErr.EcodeNotFile, nodePath, s.CurrentIndex)
}
// If both of the prevValue and prevIndex are given, we will test both of them.
// Command will be executed, only if both of the tests are successful.
if (prevValue == "" || n.Value == prevValue) && (prevIndex == 0 || n.ModifiedIndex == prevIndex) {
// update etcd index
s.CurrentIndex++
e := newEvent(CompareAndSwap, nodePath, s.CurrentIndex)
e.PrevValue = n.Value
// if test succeed, write the value
n.Write(value, s.CurrentIndex)
n.UpdateTTL(expireTime)
e.Value = value
e.Expiration, e.TTL = n.ExpirationAndTTL()
s.WatcherHub.notify(e)
s.Stats.Inc(CompareAndSwapSuccess)
return e, nil
}
cause := fmt.Sprintf("[%v != %v] [%v != %v]", prevValue, n.Value, prevIndex, n.ModifiedIndex)
s.Stats.Inc(CompareAndSwapFail)
return nil, etcdErr.NewError(etcdErr.EcodeTestFailed, cause, s.CurrentIndex)
}
// Set Command Handler
func SetHttpHandler(w http.ResponseWriter, req *http.Request) error {
key := req.URL.Path[len("/v1/keys/"):]
if store.CheckKeyword(key) {
return etcdErr.NewError(400, "Set")
}
debugf("[recv] POST %v/v1/keys/%s [%s]", e.url, key, req.RemoteAddr)
req.ParseForm()
value := req.Form.Get("value")
if len(value) == 0 {
return etcdErr.NewError(200, "Set")
}
strDuration := req.Form.Get("ttl")
expireTime, err := durationToExpireTime(strDuration)
if err != nil {
return etcdErr.NewError(202, "Set")
}
if prevValueArr, ok := req.Form["prevValue"]; ok && len(prevValueArr) > 0 {
command := &TestAndSetCommand{
Key: key,
Value: value,
PrevValue: prevValueArr[0],
ExpireTime: expireTime,
}
return dispatch(command, w, req, true)
} else {
command := &SetCommand{
Key: key,
Value: value,
ExpireTime: expireTime,
}
return dispatch(command, w, req, true)
}
}
