Matthias Andreas Benkard | 832a54e | 2019-01-29 09:27:38 +0100 | [diff] [blame] | 1 | /* |
| 2 | * |
| 3 | * Copyright 2018 gRPC authors. |
| 4 | * |
| 5 | * Licensed under the Apache License, Version 2.0 (the "License"); |
| 6 | * you may not use this file except in compliance with the License. |
| 7 | * You may obtain a copy of the License at |
| 8 | * |
| 9 | * http://www.apache.org/licenses/LICENSE-2.0 |
| 10 | * |
| 11 | * Unless required by applicable law or agreed to in writing, software |
| 12 | * distributed under the License is distributed on an "AS IS" BASIS, |
| 13 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 14 | * See the License for the specific language governing permissions and |
| 15 | * limitations under the License. |
| 16 | * |
| 17 | */ |
| 18 | |
| 19 | package channelz |
| 20 | |
| 21 | import ( |
| 22 | "net" |
| 23 | "time" |
| 24 | |
| 25 | "google.golang.org/grpc/connectivity" |
| 26 | "google.golang.org/grpc/grpclog" |
| 27 | ) |
| 28 | |
| 29 | // entry represents a node in the channelz database. |
| 30 | type entry interface { |
| 31 | // addChild adds a child e, whose channelz id is id to child list |
| 32 | addChild(id int64, e entry) |
| 33 | // deleteChild deletes a child with channelz id to be id from child list |
| 34 | deleteChild(id int64) |
| 35 | // triggerDelete tries to delete self from channelz database. However, if child |
| 36 | // list is not empty, then deletion from the database is on hold until the last |
| 37 | // child is deleted from database. |
| 38 | triggerDelete() |
| 39 | // deleteSelfIfReady check whether triggerDelete() has been called before, and whether child |
| 40 | // list is now empty. If both conditions are met, then delete self from database. |
| 41 | deleteSelfIfReady() |
| 42 | } |
| 43 | |
| 44 | // dummyEntry is a fake entry to handle entry not found case. |
| 45 | type dummyEntry struct { |
| 46 | idNotFound int64 |
| 47 | } |
| 48 | |
| 49 | func (d *dummyEntry) addChild(id int64, e entry) { |
| 50 | // Note: It is possible for a normal program to reach here under race condition. |
| 51 | // For example, there could be a race between ClientConn.Close() info being propagated |
| 52 | // to addrConn and http2Client. ClientConn.Close() cancel the context and result |
| 53 | // in http2Client to error. The error info is then caught by transport monitor |
| 54 | // and before addrConn.tearDown() is called in side ClientConn.Close(). Therefore, |
| 55 | // the addrConn will create a new transport. And when registering the new transport in |
| 56 | // channelz, its parent addrConn could have already been torn down and deleted |
| 57 | // from channelz tracking, and thus reach the code here. |
| 58 | grpclog.Infof("attempt to add child of type %T with id %d to a parent (id=%d) that doesn't currently exist", e, id, d.idNotFound) |
| 59 | } |
| 60 | |
| 61 | func (d *dummyEntry) deleteChild(id int64) { |
| 62 | // It is possible for a normal program to reach here under race condition. |
| 63 | // Refer to the example described in addChild(). |
| 64 | grpclog.Infof("attempt to delete child with id %d from a parent (id=%d) that doesn't currently exist", id, d.idNotFound) |
| 65 | } |
| 66 | |
| 67 | func (d *dummyEntry) triggerDelete() { |
| 68 | grpclog.Warningf("attempt to delete an entry (id=%d) that doesn't currently exist", d.idNotFound) |
| 69 | } |
| 70 | |
| 71 | func (*dummyEntry) deleteSelfIfReady() { |
| 72 | // code should not reach here. deleteSelfIfReady is always called on an existing entry. |
| 73 | } |
| 74 | |
| 75 | // ChannelMetric defines the info channelz provides for a specific Channel, which |
| 76 | // includes ChannelInternalMetric and channelz-specific data, such as channelz id, |
| 77 | // child list, etc. |
| 78 | type ChannelMetric struct { |
| 79 | // ID is the channelz id of this channel. |
| 80 | ID int64 |
| 81 | // RefName is the human readable reference string of this channel. |
| 82 | RefName string |
| 83 | // ChannelData contains channel internal metric reported by the channel through |
| 84 | // ChannelzMetric(). |
| 85 | ChannelData *ChannelInternalMetric |
| 86 | // NestedChans tracks the nested channel type children of this channel in the format of |
| 87 | // a map from nested channel channelz id to corresponding reference string. |
| 88 | NestedChans map[int64]string |
| 89 | // SubChans tracks the subchannel type children of this channel in the format of a |
| 90 | // map from subchannel channelz id to corresponding reference string. |
| 91 | SubChans map[int64]string |
| 92 | // Sockets tracks the socket type children of this channel in the format of a map |
| 93 | // from socket channelz id to corresponding reference string. |
| 94 | // Note current grpc implementation doesn't allow channel having sockets directly, |
| 95 | // therefore, this is field is unused. |
| 96 | Sockets map[int64]string |
| 97 | } |
| 98 | |
| 99 | // SubChannelMetric defines the info channelz provides for a specific SubChannel, |
| 100 | // which includes ChannelInternalMetric and channelz-specific data, such as |
| 101 | // channelz id, child list, etc. |
| 102 | type SubChannelMetric struct { |
| 103 | // ID is the channelz id of this subchannel. |
| 104 | ID int64 |
| 105 | // RefName is the human readable reference string of this subchannel. |
| 106 | RefName string |
| 107 | // ChannelData contains subchannel internal metric reported by the subchannel |
| 108 | // through ChannelzMetric(). |
| 109 | ChannelData *ChannelInternalMetric |
| 110 | // NestedChans tracks the nested channel type children of this subchannel in the format of |
| 111 | // a map from nested channel channelz id to corresponding reference string. |
| 112 | // Note current grpc implementation doesn't allow subchannel to have nested channels |
| 113 | // as children, therefore, this field is unused. |
| 114 | NestedChans map[int64]string |
| 115 | // SubChans tracks the subchannel type children of this subchannel in the format of a |
| 116 | // map from subchannel channelz id to corresponding reference string. |
| 117 | // Note current grpc implementation doesn't allow subchannel to have subchannels |
| 118 | // as children, therefore, this field is unused. |
| 119 | SubChans map[int64]string |
| 120 | // Sockets tracks the socket type children of this subchannel in the format of a map |
| 121 | // from socket channelz id to corresponding reference string. |
| 122 | Sockets map[int64]string |
| 123 | } |
| 124 | |
| 125 | // ChannelInternalMetric defines the struct that the implementor of Channel interface |
| 126 | // should return from ChannelzMetric(). |
| 127 | type ChannelInternalMetric struct { |
| 128 | // current connectivity state of the channel. |
| 129 | State connectivity.State |
| 130 | // The target this channel originally tried to connect to. May be absent |
| 131 | Target string |
| 132 | // The number of calls started on the channel. |
| 133 | CallsStarted int64 |
| 134 | // The number of calls that have completed with an OK status. |
| 135 | CallsSucceeded int64 |
| 136 | // The number of calls that have a completed with a non-OK status. |
| 137 | CallsFailed int64 |
| 138 | // The last time a call was started on the channel. |
| 139 | LastCallStartedTimestamp time.Time |
| 140 | //TODO: trace |
| 141 | } |
| 142 | |
| 143 | // Channel is the interface that should be satisfied in order to be tracked by |
| 144 | // channelz as Channel or SubChannel. |
| 145 | type Channel interface { |
| 146 | ChannelzMetric() *ChannelInternalMetric |
| 147 | } |
| 148 | |
| 149 | type channel struct { |
| 150 | refName string |
| 151 | c Channel |
| 152 | closeCalled bool |
| 153 | nestedChans map[int64]string |
| 154 | subChans map[int64]string |
| 155 | id int64 |
| 156 | pid int64 |
| 157 | cm *channelMap |
| 158 | } |
| 159 | |
| 160 | func (c *channel) addChild(id int64, e entry) { |
| 161 | switch v := e.(type) { |
| 162 | case *subChannel: |
| 163 | c.subChans[id] = v.refName |
| 164 | case *channel: |
| 165 | c.nestedChans[id] = v.refName |
| 166 | default: |
| 167 | grpclog.Errorf("cannot add a child (id = %d) of type %T to a channel", id, e) |
| 168 | } |
| 169 | } |
| 170 | |
| 171 | func (c *channel) deleteChild(id int64) { |
| 172 | delete(c.subChans, id) |
| 173 | delete(c.nestedChans, id) |
| 174 | c.deleteSelfIfReady() |
| 175 | } |
| 176 | |
| 177 | func (c *channel) triggerDelete() { |
| 178 | c.closeCalled = true |
| 179 | c.deleteSelfIfReady() |
| 180 | } |
| 181 | |
| 182 | func (c *channel) deleteSelfIfReady() { |
| 183 | if !c.closeCalled || len(c.subChans)+len(c.nestedChans) != 0 { |
| 184 | return |
| 185 | } |
| 186 | c.cm.deleteEntry(c.id) |
| 187 | // not top channel |
| 188 | if c.pid != 0 { |
| 189 | c.cm.findEntry(c.pid).deleteChild(c.id) |
| 190 | } |
| 191 | } |
| 192 | |
| 193 | type subChannel struct { |
| 194 | refName string |
| 195 | c Channel |
| 196 | closeCalled bool |
| 197 | sockets map[int64]string |
| 198 | id int64 |
| 199 | pid int64 |
| 200 | cm *channelMap |
| 201 | } |
| 202 | |
| 203 | func (sc *subChannel) addChild(id int64, e entry) { |
| 204 | if v, ok := e.(*normalSocket); ok { |
| 205 | sc.sockets[id] = v.refName |
| 206 | } else { |
| 207 | grpclog.Errorf("cannot add a child (id = %d) of type %T to a subChannel", id, e) |
| 208 | } |
| 209 | } |
| 210 | |
| 211 | func (sc *subChannel) deleteChild(id int64) { |
| 212 | delete(sc.sockets, id) |
| 213 | sc.deleteSelfIfReady() |
| 214 | } |
| 215 | |
| 216 | func (sc *subChannel) triggerDelete() { |
| 217 | sc.closeCalled = true |
| 218 | sc.deleteSelfIfReady() |
| 219 | } |
| 220 | |
| 221 | func (sc *subChannel) deleteSelfIfReady() { |
| 222 | if !sc.closeCalled || len(sc.sockets) != 0 { |
| 223 | return |
| 224 | } |
| 225 | sc.cm.deleteEntry(sc.id) |
| 226 | sc.cm.findEntry(sc.pid).deleteChild(sc.id) |
| 227 | } |
| 228 | |
| 229 | // SocketMetric defines the info channelz provides for a specific Socket, which |
| 230 | // includes SocketInternalMetric and channelz-specific data, such as channelz id, etc. |
| 231 | type SocketMetric struct { |
| 232 | // ID is the channelz id of this socket. |
| 233 | ID int64 |
| 234 | // RefName is the human readable reference string of this socket. |
| 235 | RefName string |
| 236 | // SocketData contains socket internal metric reported by the socket through |
| 237 | // ChannelzMetric(). |
| 238 | SocketData *SocketInternalMetric |
| 239 | } |
| 240 | |
| 241 | // SocketInternalMetric defines the struct that the implementor of Socket interface |
| 242 | // should return from ChannelzMetric(). |
| 243 | type SocketInternalMetric struct { |
| 244 | // The number of streams that have been started. |
| 245 | StreamsStarted int64 |
| 246 | // The number of streams that have ended successfully: |
| 247 | // On client side, receiving frame with eos bit set. |
| 248 | // On server side, sending frame with eos bit set. |
| 249 | StreamsSucceeded int64 |
| 250 | // The number of streams that have ended unsuccessfully: |
| 251 | // On client side, termination without receiving frame with eos bit set. |
| 252 | // On server side, termination without sending frame with eos bit set. |
| 253 | StreamsFailed int64 |
| 254 | // The number of messages successfully sent on this socket. |
| 255 | MessagesSent int64 |
| 256 | MessagesReceived int64 |
| 257 | // The number of keep alives sent. This is typically implemented with HTTP/2 |
| 258 | // ping messages. |
| 259 | KeepAlivesSent int64 |
| 260 | // The last time a stream was created by this endpoint. Usually unset for |
| 261 | // servers. |
| 262 | LastLocalStreamCreatedTimestamp time.Time |
| 263 | // The last time a stream was created by the remote endpoint. Usually unset |
| 264 | // for clients. |
| 265 | LastRemoteStreamCreatedTimestamp time.Time |
| 266 | // The last time a message was sent by this endpoint. |
| 267 | LastMessageSentTimestamp time.Time |
| 268 | // The last time a message was received by this endpoint. |
| 269 | LastMessageReceivedTimestamp time.Time |
| 270 | // The amount of window, granted to the local endpoint by the remote endpoint. |
| 271 | // This may be slightly out of date due to network latency. This does NOT |
| 272 | // include stream level or TCP level flow control info. |
| 273 | LocalFlowControlWindow int64 |
| 274 | // The amount of window, granted to the remote endpoint by the local endpoint. |
| 275 | // This may be slightly out of date due to network latency. This does NOT |
| 276 | // include stream level or TCP level flow control info. |
| 277 | RemoteFlowControlWindow int64 |
| 278 | // The locally bound address. |
| 279 | LocalAddr net.Addr |
| 280 | // The remote bound address. May be absent. |
| 281 | RemoteAddr net.Addr |
| 282 | // Optional, represents the name of the remote endpoint, if different than |
| 283 | // the original target name. |
| 284 | RemoteName string |
| 285 | //TODO: socket options |
| 286 | //TODO: Security |
| 287 | } |
| 288 | |
| 289 | // Socket is the interface that should be satisfied in order to be tracked by |
| 290 | // channelz as Socket. |
| 291 | type Socket interface { |
| 292 | ChannelzMetric() *SocketInternalMetric |
| 293 | } |
| 294 | |
| 295 | type listenSocket struct { |
| 296 | refName string |
| 297 | s Socket |
| 298 | id int64 |
| 299 | pid int64 |
| 300 | cm *channelMap |
| 301 | } |
| 302 | |
| 303 | func (ls *listenSocket) addChild(id int64, e entry) { |
| 304 | grpclog.Errorf("cannot add a child (id = %d) of type %T to a listen socket", id, e) |
| 305 | } |
| 306 | |
| 307 | func (ls *listenSocket) deleteChild(id int64) { |
| 308 | grpclog.Errorf("cannot delete a child (id = %d) from a listen socket", id) |
| 309 | } |
| 310 | |
| 311 | func (ls *listenSocket) triggerDelete() { |
| 312 | ls.cm.deleteEntry(ls.id) |
| 313 | ls.cm.findEntry(ls.pid).deleteChild(ls.id) |
| 314 | } |
| 315 | |
| 316 | func (ls *listenSocket) deleteSelfIfReady() { |
| 317 | grpclog.Errorf("cannot call deleteSelfIfReady on a listen socket") |
| 318 | } |
| 319 | |
| 320 | type normalSocket struct { |
| 321 | refName string |
| 322 | s Socket |
| 323 | id int64 |
| 324 | pid int64 |
| 325 | cm *channelMap |
| 326 | } |
| 327 | |
| 328 | func (ns *normalSocket) addChild(id int64, e entry) { |
| 329 | grpclog.Errorf("cannot add a child (id = %d) of type %T to a normal socket", id, e) |
| 330 | } |
| 331 | |
| 332 | func (ns *normalSocket) deleteChild(id int64) { |
| 333 | grpclog.Errorf("cannot delete a child (id = %d) from a normal socket", id) |
| 334 | } |
| 335 | |
| 336 | func (ns *normalSocket) triggerDelete() { |
| 337 | ns.cm.deleteEntry(ns.id) |
| 338 | ns.cm.findEntry(ns.pid).deleteChild(ns.id) |
| 339 | } |
| 340 | |
| 341 | func (ns *normalSocket) deleteSelfIfReady() { |
| 342 | grpclog.Errorf("cannot call deleteSelfIfReady on a normal socket") |
| 343 | } |
| 344 | |
| 345 | // ServerMetric defines the info channelz provides for a specific Server, which |
| 346 | // includes ServerInternalMetric and channelz-specific data, such as channelz id, |
| 347 | // child list, etc. |
| 348 | type ServerMetric struct { |
| 349 | // ID is the channelz id of this server. |
| 350 | ID int64 |
| 351 | // RefName is the human readable reference string of this server. |
| 352 | RefName string |
| 353 | // ServerData contains server internal metric reported by the server through |
| 354 | // ChannelzMetric(). |
| 355 | ServerData *ServerInternalMetric |
| 356 | // ListenSockets tracks the listener socket type children of this server in the |
| 357 | // format of a map from socket channelz id to corresponding reference string. |
| 358 | ListenSockets map[int64]string |
| 359 | } |
| 360 | |
| 361 | // ServerInternalMetric defines the struct that the implementor of Server interface |
| 362 | // should return from ChannelzMetric(). |
| 363 | type ServerInternalMetric struct { |
| 364 | // The number of incoming calls started on the server. |
| 365 | CallsStarted int64 |
| 366 | // The number of incoming calls that have completed with an OK status. |
| 367 | CallsSucceeded int64 |
| 368 | // The number of incoming calls that have a completed with a non-OK status. |
| 369 | CallsFailed int64 |
| 370 | // The last time a call was started on the server. |
| 371 | LastCallStartedTimestamp time.Time |
| 372 | //TODO: trace |
| 373 | } |
| 374 | |
| 375 | // Server is the interface to be satisfied in order to be tracked by channelz as |
| 376 | // Server. |
| 377 | type Server interface { |
| 378 | ChannelzMetric() *ServerInternalMetric |
| 379 | } |
| 380 | |
| 381 | type server struct { |
| 382 | refName string |
| 383 | s Server |
| 384 | closeCalled bool |
| 385 | sockets map[int64]string |
| 386 | listenSockets map[int64]string |
| 387 | id int64 |
| 388 | cm *channelMap |
| 389 | } |
| 390 | |
| 391 | func (s *server) addChild(id int64, e entry) { |
| 392 | switch v := e.(type) { |
| 393 | case *normalSocket: |
| 394 | s.sockets[id] = v.refName |
| 395 | case *listenSocket: |
| 396 | s.listenSockets[id] = v.refName |
| 397 | default: |
| 398 | grpclog.Errorf("cannot add a child (id = %d) of type %T to a server", id, e) |
| 399 | } |
| 400 | } |
| 401 | |
| 402 | func (s *server) deleteChild(id int64) { |
| 403 | delete(s.sockets, id) |
| 404 | delete(s.listenSockets, id) |
| 405 | s.deleteSelfIfReady() |
| 406 | } |
| 407 | |
| 408 | func (s *server) triggerDelete() { |
| 409 | s.closeCalled = true |
| 410 | s.deleteSelfIfReady() |
| 411 | } |
| 412 | |
| 413 | func (s *server) deleteSelfIfReady() { |
| 414 | if !s.closeCalled || len(s.sockets)+len(s.listenSockets) != 0 { |
| 415 | return |
| 416 | } |
| 417 | s.cm.deleteEntry(s.id) |
| 418 | } |