khenaidoo | ab1f7bd | 2019-11-14 14:00:27 -0500 | [diff] [blame] | 1 | // Copyright 2015 The etcd Authors |
| 2 | // |
| 3 | // Licensed under the Apache License, Version 2.0 (the "License"); |
| 4 | // you may not use this file except in compliance with the License. |
| 5 | // You may obtain a copy of the License at |
| 6 | // |
| 7 | // http://www.apache.org/licenses/LICENSE-2.0 |
| 8 | // |
| 9 | // Unless required by applicable law or agreed to in writing, software |
| 10 | // distributed under the License is distributed on an "AS IS" BASIS, |
| 11 | // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 12 | // See the License for the specific language governing permissions and |
| 13 | // limitations under the License. |
| 14 | |
| 15 | package rafthttp |
| 16 | |
| 17 | import ( |
| 18 | "bytes" |
| 19 | "context" |
| 20 | "io" |
| 21 | "io/ioutil" |
| 22 | "net/http" |
| 23 | "time" |
| 24 | |
| 25 | "go.etcd.io/etcd/etcdserver/api/snap" |
| 26 | "go.etcd.io/etcd/pkg/httputil" |
| 27 | pioutil "go.etcd.io/etcd/pkg/ioutil" |
| 28 | "go.etcd.io/etcd/pkg/types" |
| 29 | "go.etcd.io/etcd/raft" |
| 30 | |
| 31 | "github.com/dustin/go-humanize" |
| 32 | "go.uber.org/zap" |
| 33 | ) |
| 34 | |
| 35 | var ( |
| 36 | // timeout for reading snapshot response body |
| 37 | snapResponseReadTimeout = 5 * time.Second |
| 38 | ) |
| 39 | |
| 40 | type snapshotSender struct { |
| 41 | from, to types.ID |
| 42 | cid types.ID |
| 43 | |
| 44 | tr *Transport |
| 45 | picker *urlPicker |
| 46 | status *peerStatus |
| 47 | r Raft |
| 48 | errorc chan error |
| 49 | |
| 50 | stopc chan struct{} |
| 51 | } |
| 52 | |
| 53 | func newSnapshotSender(tr *Transport, picker *urlPicker, to types.ID, status *peerStatus) *snapshotSender { |
| 54 | return &snapshotSender{ |
| 55 | from: tr.ID, |
| 56 | to: to, |
| 57 | cid: tr.ClusterID, |
| 58 | tr: tr, |
| 59 | picker: picker, |
| 60 | status: status, |
| 61 | r: tr.Raft, |
| 62 | errorc: tr.ErrorC, |
| 63 | stopc: make(chan struct{}), |
| 64 | } |
| 65 | } |
| 66 | |
| 67 | func (s *snapshotSender) stop() { close(s.stopc) } |
| 68 | |
| 69 | func (s *snapshotSender) send(merged snap.Message) { |
| 70 | start := time.Now() |
| 71 | |
| 72 | m := merged.Message |
| 73 | to := types.ID(m.To).String() |
| 74 | |
| 75 | body := createSnapBody(s.tr.Logger, merged) |
| 76 | defer body.Close() |
| 77 | |
| 78 | u := s.picker.pick() |
| 79 | req := createPostRequest(u, RaftSnapshotPrefix, body, "application/octet-stream", s.tr.URLs, s.from, s.cid) |
| 80 | |
| 81 | if s.tr.Logger != nil { |
| 82 | s.tr.Logger.Info( |
| 83 | "sending database snapshot", |
| 84 | zap.Uint64("snapshot-index", m.Snapshot.Metadata.Index), |
| 85 | zap.String("remote-peer-id", to), |
| 86 | zap.Int64("bytes", merged.TotalSize), |
| 87 | zap.String("size", humanize.Bytes(uint64(merged.TotalSize))), |
| 88 | ) |
| 89 | } else { |
| 90 | plog.Infof("start to send database snapshot [index: %d, to %s]...", m.Snapshot.Metadata.Index, types.ID(m.To)) |
| 91 | } |
| 92 | |
| 93 | snapshotSendInflights.WithLabelValues(to).Inc() |
| 94 | defer func() { |
| 95 | snapshotSendInflights.WithLabelValues(to).Dec() |
| 96 | }() |
| 97 | |
| 98 | err := s.post(req) |
| 99 | defer merged.CloseWithError(err) |
| 100 | if err != nil { |
| 101 | if s.tr.Logger != nil { |
| 102 | s.tr.Logger.Warn( |
| 103 | "failed to send database snapshot", |
| 104 | zap.Uint64("snapshot-index", m.Snapshot.Metadata.Index), |
| 105 | zap.String("remote-peer-id", to), |
| 106 | zap.Int64("bytes", merged.TotalSize), |
| 107 | zap.String("size", humanize.Bytes(uint64(merged.TotalSize))), |
| 108 | zap.Error(err), |
| 109 | ) |
| 110 | } else { |
| 111 | plog.Warningf("database snapshot [index: %d, to: %s] failed to be sent out (%v)", m.Snapshot.Metadata.Index, types.ID(m.To), err) |
| 112 | } |
| 113 | |
| 114 | // errMemberRemoved is a critical error since a removed member should |
| 115 | // always be stopped. So we use reportCriticalError to report it to errorc. |
| 116 | if err == errMemberRemoved { |
| 117 | reportCriticalError(err, s.errorc) |
| 118 | } |
| 119 | |
| 120 | s.picker.unreachable(u) |
| 121 | s.status.deactivate(failureType{source: sendSnap, action: "post"}, err.Error()) |
| 122 | s.r.ReportUnreachable(m.To) |
| 123 | // report SnapshotFailure to raft state machine. After raft state |
| 124 | // machine knows about it, it would pause a while and retry sending |
| 125 | // new snapshot message. |
| 126 | s.r.ReportSnapshot(m.To, raft.SnapshotFailure) |
| 127 | sentFailures.WithLabelValues(to).Inc() |
| 128 | snapshotSendFailures.WithLabelValues(to).Inc() |
| 129 | return |
| 130 | } |
| 131 | s.status.activate() |
| 132 | s.r.ReportSnapshot(m.To, raft.SnapshotFinish) |
| 133 | |
| 134 | if s.tr.Logger != nil { |
| 135 | s.tr.Logger.Info( |
| 136 | "sent database snapshot", |
| 137 | zap.Uint64("snapshot-index", m.Snapshot.Metadata.Index), |
| 138 | zap.String("remote-peer-id", to), |
| 139 | zap.Int64("bytes", merged.TotalSize), |
| 140 | zap.String("size", humanize.Bytes(uint64(merged.TotalSize))), |
| 141 | ) |
| 142 | } else { |
| 143 | plog.Infof("database snapshot [index: %d, to: %s] sent out successfully", m.Snapshot.Metadata.Index, types.ID(m.To)) |
| 144 | } |
| 145 | |
| 146 | sentBytes.WithLabelValues(to).Add(float64(merged.TotalSize)) |
| 147 | snapshotSend.WithLabelValues(to).Inc() |
| 148 | snapshotSendSeconds.WithLabelValues(to).Observe(time.Since(start).Seconds()) |
| 149 | } |
| 150 | |
| 151 | // post posts the given request. |
| 152 | // It returns nil when request is sent out and processed successfully. |
| 153 | func (s *snapshotSender) post(req *http.Request) (err error) { |
| 154 | ctx, cancel := context.WithCancel(context.Background()) |
| 155 | req = req.WithContext(ctx) |
| 156 | defer cancel() |
| 157 | |
| 158 | type responseAndError struct { |
| 159 | resp *http.Response |
| 160 | body []byte |
| 161 | err error |
| 162 | } |
| 163 | result := make(chan responseAndError, 1) |
| 164 | |
| 165 | go func() { |
| 166 | resp, err := s.tr.pipelineRt.RoundTrip(req) |
| 167 | if err != nil { |
| 168 | result <- responseAndError{resp, nil, err} |
| 169 | return |
| 170 | } |
| 171 | |
| 172 | // close the response body when timeouts. |
| 173 | // prevents from reading the body forever when the other side dies right after |
| 174 | // successfully receives the request body. |
| 175 | time.AfterFunc(snapResponseReadTimeout, func() { httputil.GracefulClose(resp) }) |
| 176 | body, err := ioutil.ReadAll(resp.Body) |
| 177 | result <- responseAndError{resp, body, err} |
| 178 | }() |
| 179 | |
| 180 | select { |
| 181 | case <-s.stopc: |
| 182 | return errStopped |
| 183 | case r := <-result: |
| 184 | if r.err != nil { |
| 185 | return r.err |
| 186 | } |
| 187 | return checkPostResponse(r.resp, r.body, req, s.to) |
| 188 | } |
| 189 | } |
| 190 | |
| 191 | func createSnapBody(lg *zap.Logger, merged snap.Message) io.ReadCloser { |
| 192 | buf := new(bytes.Buffer) |
| 193 | enc := &messageEncoder{w: buf} |
| 194 | // encode raft message |
| 195 | if err := enc.encode(&merged.Message); err != nil { |
| 196 | if lg != nil { |
| 197 | lg.Panic("failed to encode message", zap.Error(err)) |
| 198 | } else { |
| 199 | plog.Panicf("encode message error (%v)", err) |
| 200 | } |
| 201 | } |
| 202 | |
| 203 | return &pioutil.ReaderAndCloser{ |
| 204 | Reader: io.MultiReader(buf, merged.ReadCloser), |
| 205 | Closer: merged.ReadCloser, |
| 206 | } |
| 207 | } |