// Do interprets r and performs an operation on s.store according to r.Method
// and other fields. If r.Method is "POST", "PUT", "DELETE", or a "GET" with
// Quorum == true, r will be sent through consensus before performing its
// respective operation. Do will block until an action is performed or there is
// an error.
func (s *EtcdServer) Do(ctx context.Context, r pb.Request) (Response, error) {
r.ID = s.reqIDGen.Next()
if r.Method == "GET" && r.Quorum {
r.Method = "QGET"
}
switch r.Method {
case "POST", "PUT", "DELETE", "QGET":
var raftReq pb.InternalRaftRequest
raftReq.V2 = &r
data, err := raftReq.Marshal()
if err != nil {
return Response{}, err
}
ch := s.w.Register(r.ID)
// TODO: benchmark the cost of time.Now()
// might be sampling?
start := time.Now()
s.r.Propose(ctx, data)
proposePending.Inc()
defer proposePending.Dec()
select {
case x := <-ch:
proposeDurations.Observe(float64(time.Since(start).Nanoseconds() / int64(time.Millisecond)))
resp := x.(Response)
return resp, resp.err
case <-ctx.Done():
proposeFailed.Inc()
s.w.Trigger(r.ID, nil) // GC wait
return Response{}, parseCtxErr(ctx.Err())
case <-s.done:
return Response{}, ErrStopped
}
case "GET":
switch {
case r.Wait:
wc, err := s.store.Watch(r.Path, r.Recursive, r.Stream, r.Since)
if err != nil {
return Response{}, err
}
return Response{Watcher: wc}, nil
default:
ev, err := s.store.Get(r.Path, r.Recursive, r.Sorted)
if err != nil {
return Response{}, err
}
return Response{Event: ev}, nil
}
case "HEAD":
ev, err := s.store.Get(r.Path, r.Recursive, r.Sorted)
if err != nil {
return Response{}, err
}
return Response{Event: ev}, nil
default:
return Response{}, ErrUnknownMethod
}
}
// TestSync tests sync 1. is nonblocking 2. proposes SYNC request.
func TestSync(t *testing.T) {
n := newNodeRecorder()
srv := &EtcdServer{
r: raftNode{Node: n},
reqIDGen: idutil.NewGenerator(0, time.Time{}),
}
// check that sync is non-blocking
done := make(chan struct{})
go func() {
srv.sync(10 * time.Second)
done <- struct{}{}
}()
select {
case <-done:
case <-time.After(time.Second):
t.Fatal("sync should be non-blocking but did not return after 1s!")
}
action, _ := n.Wait(1)
if len(action) != 1 {
t.Fatalf("len(action) = %d, want 1", len(action))
}
if action[0].Name != "Propose" {
t.Fatalf("action = %s, want Propose", action[0].Name)
}
data := action[0].Params[0].([]byte)
var r pb.Request
if err := r.Unmarshal(data); err != nil {
t.Fatalf("unmarshal request error: %v", err)
}
if r.Method != "SYNC" {
t.Errorf("method = %s, want SYNC", r.Method)
}
}
func (s *EtcdServer) run() {
var syncC <-chan time.Time
for {
select {
case <-s.Ticker:
s.Node.Tick()
case rd := <-s.Node.Ready():
s.Save(rd.HardState, rd.Entries)
s.Send(rd.Messages)
// TODO(bmizerany): do this in the background, but take
// care to apply entries in a single goroutine, and not
// race them.
for _, e := range rd.CommittedEntries {
var r pb.Request
if err := r.Unmarshal(e.Data); err != nil {
panic("TODO: this is bad, what do we do about it?")
}
s.w.Trigger(r.Id, s.apply(r))
}
if rd.SoftState != nil {
if rd.RaftState == raft.StateLeader {
syncC = s.SyncTicker
} else {
syncC = nil
}
}
case <-syncC:
s.sync(defaultSyncTimeout)
case <-s.done:
return
}
}
}
func (a *v2apiStore) processRaftRequest(ctx context.Context, r *pb.Request) (Response, error) {
data, err := r.Marshal()
if err != nil {
return Response{}, err
}
ch := a.s.w.Register(r.ID)
// TODO: benchmark the cost of time.Now()
// might be sampling?
start := time.Now()
a.s.r.Propose(ctx, data)
proposePending.Inc()
defer proposePending.Dec()
select {
case x := <-ch:
proposeDurations.Observe(float64(time.Since(start)) / float64(time.Second))
resp := x.(Response)
return resp, resp.err
case <-ctx.Done():
proposeFailed.Inc()
a.s.w.Trigger(r.ID, nil) // GC wait
return Response{}, a.s.parseProposeCtxErr(ctx.Err(), start)
case <-a.s.done:
}
return Response{}, ErrStopped
}
// Do interprets r and performs an operation on s.store according to r.Method
// and other fields. If r.Method is "POST", "PUT", "DELETE", or a "GET" with
// Quorum == true, r will be sent through consensus before performing its
// respective operation. Do will block until an action is performed or there is
// an error.
func (s *EtcdServer) Do(ctx context.Context, r pb.Request) (Response, error) {
if r.ID == 0 {
log.Panicf("request ID should never be 0")
}
if r.Method == "GET" && r.Quorum {
r.Method = "QGET"
}
switch r.Method {
case "POST", "PUT", "DELETE", "QGET":
data, err := r.Marshal()
if err != nil {
return Response{}, err
}
ch := s.w.Register(r.ID)
s.node.Propose(ctx, data)
select {
case x := <-ch:
resp := x.(Response)
return resp, resp.err
case <-ctx.Done():
s.w.Trigger(r.ID, nil) // GC wait
return Response{}, parseCtxErr(ctx.Err())
case <-s.done:
return Response{}, ErrStopped
}
case "GET":
switch {
case r.Wait:
wc, err := s.store.Watch(r.Path, r.Recursive, r.Stream, r.Since)
if err != nil {
return Response{}, err
}
return Response{Watcher: wc}, nil
default:
ev, err := s.store.Get(r.Path, r.Recursive, r.Sorted)
if err != nil {
return Response{}, err
}
return Response{Event: ev}, nil
}
case "HEAD":
ev, err := s.store.Get(r.Path, r.Recursive, r.Sorted)
if err != nil {
return Response{}, err
}
return Response{Event: ev}, nil
default:
return Response{}, ErrUnknownMethod
}
}
// TestSyncTrigger tests that the server proposes a SYNC request when its sync timer ticks
func TestSyncTrigger(t *testing.T) {
n := newReadyNode()
st := make(chan time.Time, 1)
srv := &EtcdServer{
cfg: &ServerConfig{TickMs: 1},
r: raftNode{
Node: n,
raftStorage: raft.NewMemoryStorage(),
transport: rafthttp.NewNopTransporter(),
storage: mockstorage.NewStorageRecorder(""),
},
store: mockstore.NewNop(),
SyncTicker: st,
reqIDGen: idutil.NewGenerator(0, time.Time{}),
}
// trigger the server to become a leader and accept sync requests
go func() {
srv.start()
n.readyc <- raft.Ready{
SoftState: &raft.SoftState{
RaftState: raft.StateLeader,
},
}
// trigger a sync request
st <- time.Time{}
}()
action, _ := n.Wait(1)
go srv.Stop()
if len(action) != 1 {
t.Fatalf("len(action) = %d, want 1", len(action))
}
if action[0].Name != "Propose" {
t.Fatalf("action = %s, want Propose", action[0].Name)
}
data := action[0].Params[0].([]byte)
var req pb.Request
if err := req.Unmarshal(data); err != nil {
t.Fatalf("error unmarshalling data: %v", err)
}
if req.Method != "SYNC" {
t.Fatalf("unexpected proposed request: %#v", req.Method)
}
// wait on stop message
<-n.Chan()
}
func main() {
version := flag.Int("version", 5, "4 or 5")
from := flag.String("data-dir", "", "")
flag.Parse()
if *from == "" {
log.Fatal("Must provide -data-dir flag")
}
var ents []raftpb.Entry
var err error
switch *version {
case 4:
ents, err = dump4(*from)
case 5:
ents, err = dump5(*from)
default:
err = errors.New("value of -version flag must be 4 or 5")
}
if err != nil {
log.Fatalf("Failed decoding log: %v", err)
}
for _, e := range ents {
msg := fmt.Sprintf("%2d %5d: ", e.Term, e.Index)
switch e.Type {
case raftpb.EntryNormal:
msg = fmt.Sprintf("%s norm", msg)
var r etcdserverpb.Request
if err := r.Unmarshal(e.Data); err != nil {
msg = fmt.Sprintf("%s ???", msg)
} else {
msg = fmt.Sprintf("%s %s %s %s", msg, r.Method, r.Path, r.Val)
}
case raftpb.EntryConfChange:
msg = fmt.Sprintf("%s conf", msg)
var r raftpb.ConfChange
if err := r.Unmarshal(e.Data); err != nil {
msg = fmt.Sprintf("%s ???", msg)
} else {
msg = fmt.Sprintf("%s %s %s %s", msg, r.Type, types.ID(r.NodeID), r.Context)
}
}
fmt.Println(msg)
}
}
func TestPublish(t *testing.T) {
n := newNodeRecorder()
ch := make(chan interface{}, 1)
// simulate that request has gone through consensus
ch <- Response{}
w := wait.NewWithResponse(ch)
srv := &EtcdServer{
readych: make(chan struct{}),
cfg: &ServerConfig{TickMs: 1},
id: 1,
r: raftNode{Node: n},
attributes: membership.Attributes{Name: "node1", ClientURLs: []string{"http://a", "http://b"}},
cluster: &membership.RaftCluster{},
w: w,
reqIDGen: idutil.NewGenerator(0, time.Time{}),
}
srv.publish(time.Hour)
action := n.Action()
if len(action) != 1 {
t.Fatalf("len(action) = %d, want 1", len(action))
}
if action[0].Name != "Propose" {
t.Fatalf("action = %s, want Propose", action[0].Name)
}
data := action[0].Params[0].([]byte)
var r pb.Request
if err := r.Unmarshal(data); err != nil {
t.Fatalf("unmarshal request error: %v", err)
}
if r.Method != "PUT" {
t.Errorf("method = %s, want PUT", r.Method)
}
wm := membership.Member{ID: 1, Attributes: membership.Attributes{Name: "node1", ClientURLs: []string{"http://a", "http://b"}}}
if wpath := membership.MemberAttributesStorePath(wm.ID); r.Path != wpath {
t.Errorf("path = %s, want %s", r.Path, wpath)
}
var gattr membership.Attributes
if err := json.Unmarshal([]byte(r.Val), &gattr); err != nil {
t.Fatalf("unmarshal val error: %v", err)
}
if !reflect.DeepEqual(gattr, wm.Attributes) {
t.Errorf("member = %v, want %v", gattr, wm.Attributes)
}
}
// Do interprets r and performs an operation on s.Store according to r.Method
// and other fields. If r.Method is "POST", "PUT", "DELETE", or a "GET with
// Quorum == true, r will be sent through consensus before performing its
// respective operation. Do will block until an action is performed or there is
// an error.
func (s *Server) Do(ctx context.Context, r pb.Request) (Response, error) {
if r.Id == 0 {
panic("r.Id cannot be 0")
}
if r.Method == "GET" && r.Quorum {
r.Method = "QGET"
}
switch r.Method {
case "POST", "PUT", "DELETE", "QGET":
data, err := r.Marshal()
if err != nil {
return Response{}, err
}
ch := s.w.Register(r.Id)
s.Node.Propose(ctx, data)
select {
case x := <-ch:
resp := x.(Response)
return resp, resp.err
case <-ctx.Done():
s.w.Trigger(r.Id, nil) // GC wait
return Response{}, ctx.Err()
case <-s.done:
return Response{}, ErrStopped
}
case "GET":
switch {
case r.Wait:
wc, err := s.Store.Watch(r.Path, r.Recursive, false, r.Since)
if err != nil {
return Response{}, err
}
return Response{Watcher: wc}, nil
default:
ev, err := s.Store.Get(r.Path, r.Recursive, r.Sorted)
if err != nil {
return Response{}, err
}
return Response{Event: ev}, nil
}
default:
return Response{}, ErrUnknownMethod
}
}
// sync proposes a SYNC request and is non-blocking.
// This makes no guarantee that the request will be proposed or performed.
// The request will be cancelled after the given timeout.
func (s *EtcdServer) sync(timeout time.Duration) {
ctx, cancel := context.WithTimeout(context.Background(), timeout)
req := pb.Request{
Method: "SYNC",
Id: GenID(),
Time: time.Now().UnixNano(),
}
data, err := req.Marshal()
if err != nil {
log.Printf("marshal request %#v error: %v", req, err)
return
}
// There is no promise that node has leader when do SYNC request,
// so it uses goroutine to propose.
go func() {
s.Node.Propose(ctx, data)
cancel()
}()
}
func TestUpdateVersion(t *testing.T) {
n := newNodeRecorder()
ch := make(chan interface{}, 1)
// simulate that request has gone through consensus
ch <- Response{}
w := wait.NewWithResponse(ch)
srv := &EtcdServer{
id: 1,
cfg: &ServerConfig{TickMs: 1},
r: raftNode{Node: n},
attributes: membership.Attributes{Name: "node1", ClientURLs: []string{"http://node1.com"}},
cluster: &membership.RaftCluster{},
w: w,
reqIDGen: idutil.NewGenerator(0, time.Time{}),
}
srv.updateClusterVersion("2.0.0")
action := n.Action()
if len(action) != 1 {
t.Fatalf("len(action) = %d, want 1", len(action))
}
if action[0].Name != "Propose" {
t.Fatalf("action = %s, want Propose", action[0].Name)
}
data := action[0].Params[0].([]byte)
var r pb.Request
if err := r.Unmarshal(data); err != nil {
t.Fatalf("unmarshal request error: %v", err)
}
if r.Method != "PUT" {
t.Errorf("method = %s, want PUT", r.Method)
}
if wpath := path.Join(StoreClusterPrefix, "version"); r.Path != wpath {
t.Errorf("path = %s, want %s", r.Path, wpath)
}
if r.Val != "2.0.0" {
t.Errorf("val = %s, want %s", r.Val, "2.0.0")
}
}
// Do interprets r and performs an operation on s.store according to r.Method
// and other fields. If r.Method is "POST", "PUT", "DELETE", or a "GET" with
// Quorum == true, r will be sent through consensus before performing its
// respective operation. Do will block until an action is performed or there is
// an error.
func (s *EtcdServer) Do(ctx context.Context, r pb.Request) (Response, error) {
r.ID = s.reqIDGen.Next()
if r.Method == "GET" && r.Quorum {
r.Method = "QGET"
}
v2api := (v2API)(&v2apiStore{s})
switch r.Method {
case "POST":
return v2api.Post(ctx, &r)
case "PUT":
return v2api.Put(ctx, &r)
case "DELETE":
return v2api.Delete(ctx, &r)
case "QGET":
return v2api.QGet(ctx, &r)
case "GET":
return v2api.Get(ctx, &r)
case "HEAD":
return v2api.Head(ctx, &r)
}
return Response{}, ErrUnknownMethod
}
func (a *v2apiStore) processRaftRequest(ctx context.Context, r *pb.Request) (Response, error) {
data, err := r.Marshal()
if err != nil {
return Response{}, err
}
ch := a.s.w.Register(r.ID)
start := time.Now()
a.s.r.Propose(ctx, data)
proposalsPending.Inc()
defer proposalsPending.Dec()
select {
case x := <-ch:
resp := x.(Response)
return resp, resp.err
case <-ctx.Done():
proposalsFailed.Inc()
a.s.w.Trigger(r.ID, nil) // GC wait
return Response{}, a.s.parseProposeCtxErr(ctx.Err(), start)
case <-a.s.stopping:
}
return Response{}, ErrStopped
}
func (s *Server) run() {
for {
select {
case <-s.Ticker:
s.Node.Tick()
case rd := <-s.Node.Ready():
s.Save(rd.State, rd.Entries)
s.Send(rd.Messages)
// TODO(bmizerany): do this in the background, but take
// care to apply entries in a single goroutine, and not
// race them.
for _, e := range rd.CommittedEntries {
var r pb.Request
if err := r.Unmarshal(e.Data); err != nil {
panic("TODO: this is bad, what do we do about it?")
}
s.w.Trigger(r.Id, s.apply(r))
}
case <-s.done:
return
}
}
}
func main() {
from := flag.String("data-dir", "", "")
snapfile := flag.String("start-snap", "", "The base name of snapshot file to start dumping")
index := flag.Uint64("start-index", 0, "The index to start dumping")
flag.Parse()
if *from == "" {
log.Fatal("Must provide -data-dir flag.")
}
if *snapfile != "" && *index != 0 {
log.Fatal("start-snap and start-index flags cannot be used together.")
}
var (
walsnap walpb.Snapshot
snapshot *raftpb.Snapshot
err error
)
isIndex := *index != 0
if isIndex {
fmt.Printf("Start dumping log entries from index %d.\n", *index)
walsnap.Index = *index
} else {
if *snapfile == "" {
ss := snap.New(snapDir(*from))
snapshot, err = ss.Load()
} else {
snapshot, err = snap.Read(path.Join(snapDir(*from), *snapfile))
}
switch err {
case nil:
walsnap.Index, walsnap.Term = snapshot.Metadata.Index, snapshot.Metadata.Term
nodes := genIDSlice(snapshot.Metadata.ConfState.Nodes)
fmt.Printf("Snapshot:\nterm=%d index=%d nodes=%s\n",
walsnap.Term, walsnap.Index, nodes)
case snap.ErrNoSnapshot:
fmt.Printf("Snapshot:\nempty\n")
default:
log.Fatalf("Failed loading snapshot: %v", err)
}
fmt.Println("Start dupmping log entries from snapshot.")
}
w, err := wal.OpenForRead(walDir(*from), walsnap)
if err != nil {
log.Fatalf("Failed opening WAL: %v", err)
}
wmetadata, state, ents, err := w.ReadAll()
w.Close()
if err != nil && (!isIndex || err != wal.ErrSnapshotNotFound) {
log.Fatalf("Failed reading WAL: %v", err)
}
id, cid := parseWALMetadata(wmetadata)
vid := types.ID(state.Vote)
fmt.Printf("WAL metadata:\nnodeID=%s clusterID=%s term=%d commitIndex=%d vote=%s\n",
id, cid, state.Term, state.Commit, vid)
fmt.Printf("WAL entries:\n")
fmt.Printf("lastIndex=%d\n", ents[len(ents)-1].Index)
fmt.Printf("%4s\t%10s\ttype\tdata\n", "term", "index")
for _, e := range ents {
msg := fmt.Sprintf("%4d\t%10d", e.Term, e.Index)
switch e.Type {
case raftpb.EntryNormal:
msg = fmt.Sprintf("%s\tnorm", msg)
var rr etcdserverpb.InternalRaftRequest
if err := rr.Unmarshal(e.Data); err == nil {
msg = fmt.Sprintf("%s\t%s", msg, rr.String())
break
}
var r etcdserverpb.Request
if err := r.Unmarshal(e.Data); err == nil {
switch r.Method {
case "":
msg = fmt.Sprintf("%s\tnoop", msg)
case "SYNC":
msg = fmt.Sprintf("%s\tmethod=SYNC time=%q", msg, time.Unix(0, r.Time))
case "QGET", "DELETE":
msg = fmt.Sprintf("%s\tmethod=%s path=%s", msg, r.Method, excerpt(r.Path, 64, 64))
default:
msg = fmt.Sprintf("%s\tmethod=%s path=%s val=%s", msg, r.Method, excerpt(r.Path, 64, 64), excerpt(r.Val, 128, 0))
}
break
}
msg = fmt.Sprintf("%s\t???", msg)
case raftpb.EntryConfChange:
msg = fmt.Sprintf("%s\tconf", msg)
var r raftpb.ConfChange
if err := r.Unmarshal(e.Data); err != nil {
msg = fmt.Sprintf("%s\t???", msg)
} else {
msg = fmt.Sprintf("%s\tmethod=%s id=%s", msg, r.Type, types.ID(r.NodeID))
}
}
fmt.Println(msg)
}
}
// parseKeyRequest converts a received http.Request on keysPrefix to
// a server Request, performing validation of supplied fields as appropriate.
// If any validation fails, an empty Request and non-nil error is returned.
func parseKeyRequest(r *http.Request, clock clockwork.Clock) (etcdserverpb.Request, error) {
emptyReq := etcdserverpb.Request{}
err := r.ParseForm()
if err != nil {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeInvalidForm,
err.Error(),
)
}
if !strings.HasPrefix(r.URL.Path, keysPrefix) {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeInvalidForm,
"incorrect key prefix",
)
}
p := path.Join(etcdserver.StoreKeysPrefix, r.URL.Path[len(keysPrefix):])
var pIdx, wIdx uint64
if pIdx, err = getUint64(r.Form, "prevIndex"); err != nil {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeIndexNaN,
`invalid value for "prevIndex"`,
)
}
if wIdx, err = getUint64(r.Form, "waitIndex"); err != nil {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeIndexNaN,
`invalid value for "waitIndex"`,
)
}
var rec, sort, wait, dir, quorum, stream bool
if rec, err = getBool(r.Form, "recursive"); err != nil {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeInvalidField,
`invalid value for "recursive"`,
)
}
if sort, err = getBool(r.Form, "sorted"); err != nil {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeInvalidField,
`invalid value for "sorted"`,
)
}
if wait, err = getBool(r.Form, "wait"); err != nil {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeInvalidField,
`invalid value for "wait"`,
)
}
// TODO(jonboulle): define what parameters dir is/isn't compatible with?
if dir, err = getBool(r.Form, "dir"); err != nil {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeInvalidField,
`invalid value for "dir"`,
)
}
if quorum, err = getBool(r.Form, "quorum"); err != nil {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeInvalidField,
`invalid value for "quorum"`,
)
}
if stream, err = getBool(r.Form, "stream"); err != nil {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeInvalidField,
`invalid value for "stream"`,
)
}
if wait && r.Method != "GET" {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeInvalidField,
`"wait" can only be used with GET requests`,
)
}
pV := r.FormValue("prevValue")
if _, ok := r.Form["prevValue"]; ok && pV == "" {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodePrevValueRequired,
`"prevValue" cannot be empty`,
)
}
// TTL is nullable, so leave it null if not specified
// or an empty string
var ttl *uint64
if len(r.FormValue("ttl")) > 0 {
i, err := getUint64(r.Form, "ttl")
if err != nil {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeTTLNaN,
`invalid value for "ttl"`,
)
}
ttl = &i
}
// prevExist is nullable, so leave it null if not specified
var pe *bool
if _, ok := r.Form["prevExist"]; ok {
bv, err := getBool(r.Form, "prevExist")
if err != nil {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeInvalidField,
"invalid value for prevExist",
)
}
pe = &bv
}
// refresh is nullable, so leave it null if not specified
var refresh *bool
if _, ok := r.Form["refresh"]; ok {
bv, err := getBool(r.Form, "refresh")
if err != nil {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeInvalidField,
"invalid value for refresh",
)
}
refresh = &bv
if refresh != nil && *refresh {
val := r.FormValue("value")
if _, ok := r.Form["value"]; ok && val != "" {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeRefreshValue,
`A value was provided on a refresh`,
)
}
if ttl == nil {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeRefreshTTLRequired,
`No TTL value set`,
)
}
}
}
rr := etcdserverpb.Request{
Method: r.Method,
Path: p,
Val: r.FormValue("value"),
Dir: dir,
PrevValue: pV,
PrevIndex: pIdx,
PrevExist: pe,
Wait: wait,
Since: wIdx,
Recursive: rec,
Sorted: sort,
Quorum: quorum,
Stream: stream,
}
if pe != nil {
rr.PrevExist = pe
}
if refresh != nil {
rr.Refresh = refresh
}
// Null TTL is equivalent to unset Expiration
if ttl != nil {
expr := time.Duration(*ttl) * time.Second
rr.Expiration = clock.Now().Add(expr).UnixNano()
}
return rr, nil
}
func main() {
from := flag.String("data-dir", "", "")
flag.Parse()
if *from == "" {
log.Fatal("Must provide -data-dir flag")
}
ss := snap.New(snapDir(*from))
snapshot, err := ss.Load()
var walsnap walpb.Snapshot
switch err {
case nil:
walsnap.Index, walsnap.Term = snapshot.Metadata.Index, snapshot.Metadata.Term
nodes := genIDSlice(snapshot.Metadata.ConfState.Nodes)
fmt.Printf("Snapshot:\nterm=%d index=%d nodes=%s\n",
walsnap.Term, walsnap.Index, nodes)
case snap.ErrNoSnapshot:
fmt.Printf("Snapshot:\nempty\n")
default:
log.Fatalf("Failed loading snapshot: %v", err)
}
w, err := wal.Open(walDir(*from), walsnap)
if err != nil {
log.Fatalf("Failed opening WAL: %v", err)
}
wmetadata, state, ents, err := w.ReadAll()
w.Close()
if err != nil {
log.Fatalf("Failed reading WAL: %v", err)
}
id, cid := parseWALMetadata(wmetadata)
vid := types.ID(state.Vote)
fmt.Printf("WAL metadata:\nnodeID=%s clusterID=%s term=%d commitIndex=%d vote=%s\n",
id, cid, state.Term, state.Commit, vid)
fmt.Printf("WAL entries:\n")
fmt.Printf("lastIndex=%d\n", ents[len(ents)-1].Index)
fmt.Printf("%4s\t%10s\ttype\tdata\n", "term", "index")
for _, e := range ents {
msg := fmt.Sprintf("%4d\t%10d", e.Term, e.Index)
switch e.Type {
case raftpb.EntryNormal:
msg = fmt.Sprintf("%s\tnorm", msg)
var r etcdserverpb.Request
if err := r.Unmarshal(e.Data); err != nil {
msg = fmt.Sprintf("%s\t???", msg)
break
}
switch r.Method {
case "":
msg = fmt.Sprintf("%s\tnoop", msg)
case "SYNC":
msg = fmt.Sprintf("%s\tmethod=SYNC time=%q", msg, time.Unix(0, r.Time))
case "QGET", "DELETE":
msg = fmt.Sprintf("%s\tmethod=%s path=%s", msg, r.Method, excerpt(r.Path, 64, 64))
default:
msg = fmt.Sprintf("%s\tmethod=%s path=%s val=%s", msg, r.Method, excerpt(r.Path, 64, 64), excerpt(r.Val, 128, 0))
}
case raftpb.EntryConfChange:
msg = fmt.Sprintf("%s\tconf", msg)
var r raftpb.ConfChange
if err := r.Unmarshal(e.Data); err != nil {
msg = fmt.Sprintf("%s\t???", msg)
} else {
msg = fmt.Sprintf("%s\tmethod=%s id=%s", msg, r.Type, types.ID(r.NodeID))
}
}
fmt.Println(msg)
}
}
// parseRequest converts a received http.Request to a server Request,
// performing validation of supplied fields as appropriate.
// If any validation fails, an empty Request and non-nil error is returned.
func parseRequest(r *http.Request, id int64) (etcdserverpb.Request, error) {
emptyReq := etcdserverpb.Request{}
err := r.ParseForm()
if err != nil {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeInvalidForm,
err.Error(),
)
}
if !strings.HasPrefix(r.URL.Path, keysPrefix) {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeInvalidForm,
"incorrect key prefix",
)
}
p := r.URL.Path[len(keysPrefix):]
var pIdx, wIdx uint64
if pIdx, err = getUint64(r.Form, "prevIndex"); err != nil {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeIndexNaN,
`invalid value for "prevIndex"`,
)
}
if wIdx, err = getUint64(r.Form, "waitIndex"); err != nil {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeIndexNaN,
`invalid value for "waitIndex"`,
)
}
var rec, sort, wait, dir, stream bool
if rec, err = getBool(r.Form, "recursive"); err != nil {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeInvalidField,
`invalid value for "recursive"`,
)
}
if sort, err = getBool(r.Form, "sorted"); err != nil {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeInvalidField,
`invalid value for "sorted"`,
)
}
if wait, err = getBool(r.Form, "wait"); err != nil {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeInvalidField,
`invalid value for "wait"`,
)
}
// TODO(jonboulle): define what parameters dir is/isn't compatible with?
if dir, err = getBool(r.Form, "dir"); err != nil {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeInvalidField,
`invalid value for "dir"`,
)
}
if stream, err = getBool(r.Form, "stream"); err != nil {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeInvalidField,
`invalid value for "stream"`,
)
}
if wait && r.Method != "GET" {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeInvalidField,
`"wait" can only be used with GET requests`,
)
}
pV := r.FormValue("prevValue")
if _, ok := r.Form["prevValue"]; ok && pV == "" {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeInvalidField,
`"prevValue" cannot be empty`,
)
}
// TTL is nullable, so leave it null if not specified
// or an empty string
var ttl *uint64
if len(r.FormValue("ttl")) > 0 {
i, err := getUint64(r.Form, "ttl")
if err != nil {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeTTLNaN,
`invalid value for "ttl"`,
)
}
ttl = &i
}
// prevExist is nullable, so leave it null if not specified
var pe *bool
if _, ok := r.Form["prevExist"]; ok {
bv, err := getBool(r.Form, "prevExist")
if err != nil {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeInvalidField,
"invalid value for prevExist",
)
}
pe = &bv
}
rr := etcdserverpb.Request{
ID: id,
Method: r.Method,
Path: p,
Val: r.FormValue("value"),
Dir: dir,
PrevValue: pV,
PrevIndex: pIdx,
PrevExist: pe,
Recursive: rec,
Since: wIdx,
Sorted: sort,
Stream: stream,
Wait: wait,
}
if pe != nil {
rr.PrevExist = pe
}
// Null TTL is equivalent to unset Expiration
// TODO(jonboulle): use fake clock instead of time module
// https://github.com/coreos/etcd/issues/1021
if ttl != nil {
expr := time.Duration(*ttl) * time.Second
rr.Expiration = time.Now().Add(expr).UnixNano()
}
return rr, nil
}
// parseRequest converts a received http.Request to a server Request,
// performing validation of supplied fields as appropriate.
// If any validation fails, an empty Request and non-nil error is returned.
func parseRequest(r *http.Request, id int64) (etcdserverpb.Request, error) {
emptyReq := etcdserverpb.Request{}
err := r.ParseForm()
if err != nil {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeInvalidForm,
err.Error(),
)
}
if !strings.HasPrefix(r.URL.Path, keysPrefix) {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeInvalidForm,
"incorrect key prefix",
)
}
p := r.URL.Path[len(keysPrefix):]
var pIdx, wIdx, ttl uint64
if pIdx, err = getUint64(r.Form, "prevIndex"); err != nil {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeIndexNaN,
`invalid value for "prevIndex"`,
)
}
if wIdx, err = getUint64(r.Form, "waitIndex"); err != nil {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeIndexNaN,
`invalid value for "waitIndex"`,
)
}
if ttl, err = getUint64(r.Form, "ttl"); err != nil {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeTTLNaN,
`invalid value for "ttl"`,
)
}
var rec, sort, wait bool
if rec, err = getBool(r.Form, "recursive"); err != nil {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeInvalidField,
`invalid value for "recursive"`,
)
}
if sort, err = getBool(r.Form, "sorted"); err != nil {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeInvalidField,
`invalid value for "sorted"`,
)
}
if wait, err = getBool(r.Form, "wait"); err != nil {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeInvalidField,
`invalid value for "wait"`,
)
}
// prevExists is nullable, so leave it null if not specified
var pe *bool
if _, ok := r.Form["prevExists"]; ok {
bv, err := getBool(r.Form, "prevExists")
if err != nil {
return emptyReq, etcdErr.NewRequestError(
etcdErr.EcodeInvalidField,
"invalid value for prevExists",
)
}
pe = &bv
}
rr := etcdserverpb.Request{
Id: id,
Method: r.Method,
Path: p,
Val: r.FormValue("value"),
PrevValue: r.FormValue("prevValue"),
PrevIndex: pIdx,
PrevExists: pe,
Recursive: rec,
Since: wIdx,
Sorted: sort,
Wait: wait,
}
if pe != nil {
rr.PrevExists = pe
}
// TODO(jonboulle): use fake clock instead of time module
// https://github.com/coreos/etcd/issues/1021
if ttl > 0 {
expr := time.Duration(ttl) * time.Second
rr.Expiration = time.Now().Add(expr).UnixNano()
}
return rr, nil
}
func (s *EtcdServer) run() {
var syncC <-chan time.Time
// snapi indicates the index of the last submitted snapshot request
var snapi, appliedi int64
for {
select {
case <-s.ticker:
s.node.Tick()
case rd := <-s.node.Ready():
s.storage.Save(rd.HardState, rd.Entries)
s.storage.SaveSnap(rd.Snapshot)
s.send(rd.Messages)
// TODO(bmizerany): do this in the background, but take
// care to apply entries in a single goroutine, and not
// race them.
// TODO: apply configuration change into ClusterStore.
for _, e := range rd.CommittedEntries {
switch e.Type {
case raftpb.EntryNormal:
var r pb.Request
if err := r.Unmarshal(e.Data); err != nil {
panic("TODO: this is bad, what do we do about it?")
}
s.w.Trigger(r.ID, s.apply(r))
case raftpb.EntryConfChange:
var cc raftpb.ConfChange
if err := cc.Unmarshal(e.Data); err != nil {
panic("TODO: this is bad, what do we do about it?")
}
s.node.ApplyConfChange(cc)
s.w.Trigger(cc.ID, nil)
default:
panic("unexpected entry type")
}
atomic.StoreInt64(&s.raftIndex, e.Index)
atomic.StoreInt64(&s.raftTerm, e.Term)
appliedi = e.Index
}
if rd.Snapshot.Index > snapi {
snapi = rd.Snapshot.Index
}
// recover from snapshot if it is more updated than current applied
if rd.Snapshot.Index > appliedi {
if err := s.store.Recovery(rd.Snapshot.Data); err != nil {
panic("TODO: this is bad, what do we do about it?")
}
appliedi = rd.Snapshot.Index
}
if appliedi-snapi > s.snapCount {
s.snapshot()
snapi = appliedi
}
if rd.SoftState != nil {
if rd.RaftState == raft.StateLeader {
syncC = s.syncTicker
} else {
syncC = nil
}
if rd.SoftState.ShouldStop {
s.Stop()
return
}
}
case <-syncC:
s.sync(defaultSyncTimeout)
case <-s.done:
return
}
}
}
