-
Notifications
You must be signed in to change notification settings - Fork 1.9k
/
rpc.go
499 lines (429 loc) · 13.9 KB
/
rpc.go
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
// Copyright (c) HashiCorp, Inc.
// SPDX-License-Identifier: BUSL-1.1
package client
import (
"errors"
"io"
"net"
"net/rpc"
"strings"
"time"
metrics "github.com/armon/go-metrics"
"github.com/hashicorp/go-msgpack/v2/codec"
"github.com/hashicorp/nomad/client/servers"
"github.com/hashicorp/nomad/helper"
inmem "github.com/hashicorp/nomad/helper/codec"
"github.com/hashicorp/nomad/helper/pool"
"github.com/hashicorp/nomad/nomad/structs"
)
// rpcEndpoints holds the RPC endpoints
type rpcEndpoints struct {
ClientStats *ClientStats
CSI *CSI
FileSystem *FileSystem
Allocations *Allocations
Agent *Agent
NodeMeta *NodeMeta
}
// ClientRPC is used to make a local, client only RPC call
func (c *Client) ClientRPC(method string, args interface{}, reply interface{}) error {
codec := &inmem.InmemCodec{
Method: method,
Args: args,
Reply: reply,
}
if err := c.rpcServer.ServeRequest(codec); err != nil {
return err
}
return codec.Err
}
// StreamingRpcHandler is used to make a local, client only streaming RPC
// call.
func (c *Client) StreamingRpcHandler(method string) (structs.StreamingRpcHandler, error) {
return c.streamingRpcs.GetHandler(method)
}
// RPC is used to forward an RPC call to a nomad server, or fail if no servers.
func (c *Client) RPC(method string, args any, reply any) error {
// Block if we have not yet registered the node, to enforce that we only
// send authenticated calls after the node has been registered
select {
case <-c.registeredCh:
case <-c.shutdownCh:
return nil
}
return c.rpc(method, args, reply)
}
// UnauthenticatedRPC special-cases the Node.Register RPC call, forwarding the
// call to a nomad server without blocking on the initial node registration.
func (c *Client) UnauthenticatedRPC(method string, args any, reply any) error {
return c.rpc(method, args, reply)
}
// rpc implements the forwarding of a RPC call to a nomad server, or fail if
// no servers.
func (c *Client) rpc(method string, args any, reply any) error {
conf := c.GetConfig()
// Invoke the RPCHandler if it exists
if conf.RPCHandler != nil {
return conf.RPCHandler.RPC(method, args, reply)
}
// We will try to automatically retry requests that fail due to things like server unavailability
// but instead of retrying forever, lets have a solid upper-bound
deadline := time.Now()
// A reasonable amount of time for leader election. Note when servers forward() our RPC requests
// to the leader they may also allow for an RPCHoldTimeout while waiting for leader election.
// That's OK, we won't double up because we are using it here not as a sleep but
// as a hint to give up
deadline = deadline.Add(conf.RPCHoldTimeout)
// If its a blocking query, allow the time specified by the request
if info, ok := args.(structs.RPCInfo); ok {
deadline = deadline.Add(info.TimeToBlock())
}
TRY:
var rpcErr error
server := c.servers.FindServer()
if server == nil {
rpcErr = noServersErr
} else {
// Make the request.
rpcErr = c.connPool.RPC(c.Region(), server.Addr, method, args, reply)
if rpcErr == nil {
c.fireRpcRetryWatcher()
return nil
}
// If shutting down, exit without logging the error
select {
case <-c.shutdownCh:
return nil
default:
}
// Move off to another server, and see if we can retry.
c.rpcLogger.Error("error performing RPC to server", "error", rpcErr, "rpc", method, "server", server.Addr)
c.servers.NotifyFailedServer(server)
if !canRetry(args, rpcErr) {
c.rpcLogger.Error("error performing RPC to server which is not safe to automatically retry", "error", rpcErr, "rpc", method, "server", server.Addr)
return rpcErr
}
}
if time.Now().After(deadline) {
// Blocking queries are tricky. jitters and rpcholdtimes in multiple places can result in our server call taking longer than we wanted it to. For example:
// a block time of 5s may easily turn into the server blocking for 10s since it applies its own RPCHoldTime. If the server dies at t=7s we still want to retry
// so before we give up on blocking queries make one last attempt for an immediate answer
if info, ok := args.(structs.RPCInfo); ok && info.TimeToBlock() > 0 {
info.SetTimeToBlock(0)
return c.RPC(method, args, reply)
}
c.rpcLogger.Error("error performing RPC to server, deadline exceeded, cannot retry", "error", rpcErr, "rpc", method)
return rpcErr
}
// Wait to avoid thundering herd
timer, cancel := helper.NewSafeTimer(helper.RandomStagger(conf.RPCHoldTimeout / structs.JitterFraction))
defer cancel()
select {
case <-timer.C:
// If we are going to retry a blocking query we need to update the time to block so it finishes by our deadline.
if info, ok := args.(structs.RPCInfo); ok && info.TimeToBlock() > 0 {
newBlockTime := time.Until(deadline)
// We can get below 0 here on slow computers because we slept for jitter so at least try to get an immediate response
if newBlockTime < 0 {
newBlockTime = 0
}
info.SetTimeToBlock(newBlockTime)
return c.RPC(method, args, reply)
}
goto TRY
case <-c.shutdownCh:
}
return rpcErr
}
// canRetry returns true if the given situation is safe for a retry.
func canRetry(args interface{}, err error) bool {
// No leader errors are always safe to retry since no state could have
// been changed.
if structs.IsErrNoLeader(err) {
return true
}
// Reads are safe to retry for stream errors, such as if a server was
// being shut down.
info, ok := args.(structs.RPCInfo)
if ok && info.IsRead() && helper.IsErrEOF(err) {
return true
}
return false
}
// RemoteStreamingRpcHandler is used to make a streaming RPC call to a remote
// server.
func (c *Client) RemoteStreamingRpcHandler(method string) (structs.StreamingRpcHandler, error) {
server := c.servers.FindServer()
if server == nil {
return nil, noServersErr
}
conn, err := c.streamingRpcConn(server, method)
if err != nil {
// Move off to another server
c.rpcLogger.Error("error performing RPC to server", "error", err, "rpc", method, "server", server.Addr)
c.servers.NotifyFailedServer(server)
return nil, err
}
return bridgedStreamingRpcHandler(conn), nil
}
// bridgedStreamingRpcHandler creates a bridged streaming RPC handler by copying
// data between the two sides.
func bridgedStreamingRpcHandler(sideA io.ReadWriteCloser) structs.StreamingRpcHandler {
return func(sideB io.ReadWriteCloser) {
defer sideA.Close()
defer sideB.Close()
structs.Bridge(sideA, sideB)
}
}
// streamingRpcConn is used to retrieve a connection to a server to conduct a
// streaming RPC.
func (c *Client) streamingRpcConn(server *servers.Server, method string) (net.Conn, error) {
// Dial the server
conn, err := net.DialTimeout("tcp", server.Addr.String(), 10*time.Second)
if err != nil {
return nil, err
}
// Cast to TCPConn
if tcp, ok := conn.(*net.TCPConn); ok {
tcp.SetKeepAlive(true)
tcp.SetNoDelay(true)
}
// Check if TLS is enabled
c.tlsWrapLock.RLock()
tlsWrap := c.tlsWrap
c.tlsWrapLock.RUnlock()
if tlsWrap != nil {
// Switch the connection into TLS mode
if _, err := conn.Write([]byte{byte(pool.RpcTLS)}); err != nil {
conn.Close()
return nil, err
}
// Wrap the connection in a TLS client
tlsConn, err := tlsWrap(c.Region(), conn)
if err != nil {
conn.Close()
return nil, err
}
conn = tlsConn
}
// Write the multiplex byte to set the mode
if _, err := conn.Write([]byte{byte(pool.RpcStreaming)}); err != nil {
conn.Close()
return nil, err
}
// Send the header
encoder := codec.NewEncoder(conn, structs.MsgpackHandle)
decoder := codec.NewDecoder(conn, structs.MsgpackHandle)
header := structs.StreamingRpcHeader{
Method: method,
}
if err := encoder.Encode(header); err != nil {
conn.Close()
return nil, err
}
// Wait for the acknowledgement
var ack structs.StreamingRpcAck
if err := decoder.Decode(&ack); err != nil {
conn.Close()
return nil, err
}
if ack.Error != "" {
conn.Close()
return nil, errors.New(ack.Error)
}
return conn, nil
}
// setupClientRpc is used to setup the Client's RPC endpoints
func (c *Client) setupClientRpc(rpcs map[string]interface{}) {
// Create the RPC Server
c.rpcServer = rpc.NewServer()
// Initialize the RPC handlers
if rpcs != nil {
// override RPCs
for name, rpc := range rpcs {
c.rpcServer.RegisterName(name, rpc)
}
} else {
c.endpoints.ClientStats = &ClientStats{c}
c.endpoints.CSI = &CSI{c}
c.endpoints.FileSystem = NewFileSystemEndpoint(c)
c.endpoints.Allocations = NewAllocationsEndpoint(c)
c.endpoints.Agent = NewAgentEndpoint(c)
c.endpoints.NodeMeta = newNodeMetaEndpoint(c)
c.setupClientRpcServer(c.rpcServer)
}
go c.rpcConnListener()
}
// setupClientRpcServer is used to populate a client RPC server with endpoints.
func (c *Client) setupClientRpcServer(server *rpc.Server) {
// Register the endpoints
server.Register(c.endpoints.ClientStats)
server.Register(c.endpoints.CSI)
server.Register(c.endpoints.FileSystem)
server.Register(c.endpoints.Allocations)
server.Register(c.endpoints.Agent)
server.Register(c.endpoints.NodeMeta)
}
// rpcConnListener is a long lived function that listens for new connections
// being made on the connection pool and starts an RPC listener for each
// connection.
func (c *Client) rpcConnListener() {
// Make a channel for new connections.
conns := make(chan *pool.Conn, 4)
c.connPool.SetConnListener(conns)
for {
select {
case <-c.shutdownCh:
return
case conn, ok := <-conns:
if !ok {
continue
}
go c.listenConn(conn)
}
}
}
// listenConn is used to listen for connections being made from the server on
// pre-existing connection. This should be called in a goroutine.
func (c *Client) listenConn(conn *pool.Conn) {
for {
stream, err := conn.AcceptStream()
if err != nil {
if conn.IsClosed() {
return
}
c.rpcLogger.Error("failed to accept RPC conn", "error", err)
continue
}
go c.handleConn(stream)
metrics.IncrCounter([]string{"client", "rpc", "accept_conn"}, 1)
}
}
// handleConn is used to determine if this is a RPC or Streaming RPC connection and
// invoke the correct handler
func (c *Client) handleConn(conn net.Conn) {
// Read a single byte
buf := make([]byte, 1)
if _, err := conn.Read(buf); err != nil {
if err != io.EOF {
c.rpcLogger.Error("error reading byte", "error", err)
}
conn.Close()
return
}
// Switch on the byte
switch pool.RPCType(buf[0]) {
case pool.RpcNomad:
c.handleNomadConn(conn)
case pool.RpcStreaming:
c.handleStreamingConn(conn)
default:
c.rpcLogger.Error("unrecognized RPC byte", "byte", buf[0])
conn.Close()
return
}
}
// handleNomadConn is used to handle a single Nomad RPC connection.
func (c *Client) handleNomadConn(conn net.Conn) {
defer conn.Close()
rpcCodec := pool.NewServerCodec(conn)
for {
select {
case <-c.shutdownCh:
return
default:
}
if err := c.rpcServer.ServeRequest(rpcCodec); err != nil {
if err != io.EOF && !strings.Contains(err.Error(), "closed") {
c.rpcLogger.Error("error performing RPC", "error", err, "addr", conn.RemoteAddr())
metrics.IncrCounter([]string{"client", "rpc", "request_error"}, 1)
}
return
}
metrics.IncrCounter([]string{"client", "rpc", "request"}, 1)
}
}
// handleStreamingConn is used to handle a single Streaming Nomad RPC connection.
func (c *Client) handleStreamingConn(conn net.Conn) {
defer conn.Close()
// Decode the header
var header structs.StreamingRpcHeader
decoder := codec.NewDecoder(conn, structs.MsgpackHandle)
if err := decoder.Decode(&header); err != nil {
if err != io.EOF && !strings.Contains(err.Error(), "closed") {
c.rpcLogger.Error("error performing streaming RPC", "error", err, "addr", conn.RemoteAddr())
metrics.IncrCounter([]string{"client", "streaming_rpc", "request_error"}, 1)
}
return
}
ack := structs.StreamingRpcAck{}
handler, err := c.streamingRpcs.GetHandler(header.Method)
if err != nil {
c.rpcLogger.Error("streaming RPC error", "addr", conn.RemoteAddr(), "error", err)
metrics.IncrCounter([]string{"client", "streaming_rpc", "request_error"}, 1)
ack.Error = err.Error()
}
// Send the acknowledgement
encoder := codec.NewEncoder(conn, structs.MsgpackHandle)
if err := encoder.Encode(ack); err != nil {
conn.Close()
return
}
if ack.Error != "" {
return
}
// Invoke the handler
metrics.IncrCounter([]string{"client", "streaming_rpc", "request"}, 1)
handler(conn)
}
// resolveServer given a sever's address as a string, return it's resolved
// net.Addr or an error.
func resolveServer(s string) (net.Addr, error) {
const defaultClientPort = "4647" // default client RPC port
host, port, err := net.SplitHostPort(s)
if err != nil {
if strings.Contains(err.Error(), "missing port") {
// with IPv6 addresses the `host` variable will have brackets
// removed, so send the original value thru again with only the
// correct port suffix
return resolveServer(s + ":" + defaultClientPort)
} else {
return nil, err
}
} else if port == "" {
return resolveServer(s + defaultClientPort)
}
return net.ResolveTCPAddr("tcp", net.JoinHostPort(host, port))
}
// Ping is used to ping a particular server and returns whether it is healthy or
// a potential error.
func (c *Client) Ping(srv net.Addr) error {
pingRequest := &structs.GenericRequest{
QueryOptions: structs.QueryOptions{
AuthToken: c.secretNodeID(),
},
}
var reply struct{}
err := c.connPool.RPC(c.Region(), srv, "Status.Ping", pingRequest, &reply)
return err
}
// rpcRetryWatcher returns a channel that will be closed if an event happens
// such that we expect the next RPC to be successful.
func (c *Client) rpcRetryWatcher() <-chan struct{} {
c.rpcRetryLock.Lock()
defer c.rpcRetryLock.Unlock()
if c.rpcRetryCh == nil {
c.rpcRetryCh = make(chan struct{})
}
return c.rpcRetryCh
}
// fireRpcRetryWatcher causes any RPC retryloops to retry their RPCs because we
// believe the will be successful.
func (c *Client) fireRpcRetryWatcher() {
c.rpcRetryLock.Lock()
defer c.rpcRetryLock.Unlock()
if c.rpcRetryCh != nil {
close(c.rpcRetryCh)
c.rpcRetryCh = nil
}
}