clientconn.go raw

   1  /*
   2   *
   3   * Copyright 2014 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 grpc
  20  
  21  import (
  22  	"context"
  23  	"errors"
  24  	"fmt"
  25  	"math"
  26  	"net/url"
  27  	"slices"
  28  	"strings"
  29  	"sync"
  30  	"sync/atomic"
  31  	"time"
  32  
  33  	"google.golang.org/grpc/balancer"
  34  	"google.golang.org/grpc/balancer/base"
  35  	"google.golang.org/grpc/balancer/pickfirst"
  36  	"google.golang.org/grpc/codes"
  37  	"google.golang.org/grpc/connectivity"
  38  	"google.golang.org/grpc/credentials"
  39  	expstats "google.golang.org/grpc/experimental/stats"
  40  	"google.golang.org/grpc/internal"
  41  	"google.golang.org/grpc/internal/channelz"
  42  	"google.golang.org/grpc/internal/grpcsync"
  43  	"google.golang.org/grpc/internal/idle"
  44  	iresolver "google.golang.org/grpc/internal/resolver"
  45  	istats "google.golang.org/grpc/internal/stats"
  46  	"google.golang.org/grpc/internal/transport"
  47  	"google.golang.org/grpc/keepalive"
  48  	"google.golang.org/grpc/resolver"
  49  	"google.golang.org/grpc/serviceconfig"
  50  	"google.golang.org/grpc/stats"
  51  	"google.golang.org/grpc/status"
  52  
  53  	_ "google.golang.org/grpc/balancer/roundrobin"           // To register roundrobin.
  54  	_ "google.golang.org/grpc/internal/resolver/passthrough" // To register passthrough resolver.
  55  	_ "google.golang.org/grpc/internal/resolver/unix"        // To register unix resolver.
  56  	_ "google.golang.org/grpc/resolver/dns"                  // To register dns resolver.
  57  )
  58  
  59  const (
  60  	// minimum time to give a connection to complete
  61  	minConnectTimeout = 20 * time.Second
  62  )
  63  
  64  var (
  65  	// ErrClientConnClosing indicates that the operation is illegal because
  66  	// the ClientConn is closing.
  67  	//
  68  	// Deprecated: this error should not be relied upon by users; use the status
  69  	// code of Canceled instead.
  70  	ErrClientConnClosing = status.Error(codes.Canceled, "grpc: the client connection is closing")
  71  	// errConnDrain indicates that the connection starts to be drained and does not accept any new RPCs.
  72  	errConnDrain = errors.New("grpc: the connection is drained")
  73  	// errConnClosing indicates that the connection is closing.
  74  	errConnClosing = errors.New("grpc: the connection is closing")
  75  	// errConnIdling indicates the connection is being closed as the channel
  76  	// is moving to an idle mode due to inactivity.
  77  	errConnIdling = errors.New("grpc: the connection is closing due to channel idleness")
  78  	// invalidDefaultServiceConfigErrPrefix is used to prefix the json parsing error for the default
  79  	// service config.
  80  	invalidDefaultServiceConfigErrPrefix = "grpc: the provided default service config is invalid"
  81  	// PickFirstBalancerName is the name of the pick_first balancer.
  82  	PickFirstBalancerName = pickfirst.Name
  83  )
  84  
  85  // The following errors are returned from Dial and DialContext
  86  var (
  87  	// errNoTransportSecurity indicates that there is no transport security
  88  	// being set for ClientConn. Users should either set one or explicitly
  89  	// call WithInsecure DialOption to disable security.
  90  	errNoTransportSecurity = errors.New("grpc: no transport security set (use grpc.WithTransportCredentials(insecure.NewCredentials()) explicitly or set credentials)")
  91  	// errTransportCredsAndBundle indicates that creds bundle is used together
  92  	// with other individual Transport Credentials.
  93  	errTransportCredsAndBundle = errors.New("grpc: credentials.Bundle may not be used with individual TransportCredentials")
  94  	// errNoTransportCredsInBundle indicated that the configured creds bundle
  95  	// returned a transport credentials which was nil.
  96  	errNoTransportCredsInBundle = errors.New("grpc: credentials.Bundle must return non-nil transport credentials")
  97  	// errTransportCredentialsMissing indicates that users want to transmit
  98  	// security information (e.g., OAuth2 token) which requires secure
  99  	// connection on an insecure connection.
 100  	errTransportCredentialsMissing = errors.New("grpc: the credentials require transport level security (use grpc.WithTransportCredentials() to set)")
 101  )
 102  
 103  var (
 104  	disconnectionsMetric = expstats.RegisterInt64Count(expstats.MetricDescriptor{
 105  		Name:           "grpc.subchannel.disconnections",
 106  		Description:    "EXPERIMENTAL. Number of times the selected subchannel becomes disconnected.",
 107  		Unit:           "{disconnection}",
 108  		Labels:         []string{"grpc.target"},
 109  		OptionalLabels: []string{"grpc.lb.backend_service", "grpc.lb.locality", "grpc.disconnect_error"},
 110  		Default:        false,
 111  	})
 112  	connectionAttemptsSucceededMetric = expstats.RegisterInt64Count(expstats.MetricDescriptor{
 113  		Name:           "grpc.subchannel.connection_attempts_succeeded",
 114  		Description:    "EXPERIMENTAL. Number of successful connection attempts.",
 115  		Unit:           "{attempt}",
 116  		Labels:         []string{"grpc.target"},
 117  		OptionalLabels: []string{"grpc.lb.backend_service", "grpc.lb.locality"},
 118  		Default:        false,
 119  	})
 120  	connectionAttemptsFailedMetric = expstats.RegisterInt64Count(expstats.MetricDescriptor{
 121  		Name:           "grpc.subchannel.connection_attempts_failed",
 122  		Description:    "EXPERIMENTAL. Number of failed connection attempts.",
 123  		Unit:           "{attempt}",
 124  		Labels:         []string{"grpc.target"},
 125  		OptionalLabels: []string{"grpc.lb.backend_service", "grpc.lb.locality"},
 126  		Default:        false,
 127  	})
 128  	openConnectionsMetric = expstats.RegisterInt64UpDownCount(expstats.MetricDescriptor{
 129  		Name:           "grpc.subchannel.open_connections",
 130  		Description:    "EXPERIMENTAL. Number of open connections.",
 131  		Unit:           "{attempt}",
 132  		Labels:         []string{"grpc.target"},
 133  		OptionalLabels: []string{"grpc.lb.backend_service", "grpc.security_level", "grpc.lb.locality"},
 134  		Default:        false,
 135  	})
 136  )
 137  
 138  const (
 139  	defaultClientMaxReceiveMessageSize = 1024 * 1024 * 4
 140  	defaultClientMaxSendMessageSize    = math.MaxInt32
 141  	// http2IOBufSize specifies the buffer size for sending frames.
 142  	defaultWriteBufSize = 32 * 1024
 143  	defaultReadBufSize  = 32 * 1024
 144  )
 145  
 146  type defaultConfigSelector struct {
 147  	sc *ServiceConfig
 148  }
 149  
 150  func (dcs *defaultConfigSelector) SelectConfig(rpcInfo iresolver.RPCInfo) (*iresolver.RPCConfig, error) {
 151  	return &iresolver.RPCConfig{
 152  		Context:      rpcInfo.Context,
 153  		MethodConfig: getMethodConfig(dcs.sc, rpcInfo.Method),
 154  	}, nil
 155  }
 156  
 157  // NewClient creates a new gRPC "channel" for the target URI provided.  No I/O
 158  // is performed.  Use of the ClientConn for RPCs will automatically cause it to
 159  // connect.  The Connect method may be called to manually create a connection,
 160  // but for most users this should be unnecessary.
 161  //
 162  // The target name syntax is defined in
 163  // https://github.com/grpc/grpc/blob/master/doc/naming.md.  E.g. to use the dns
 164  // name resolver, a "dns:///" prefix may be applied to the target.  The default
 165  // name resolver will be used if no scheme is detected, or if the parsed scheme
 166  // is not a registered name resolver.  The default resolver is "dns" but can be
 167  // overridden using the resolver package's SetDefaultScheme.
 168  //
 169  // Examples:
 170  //
 171  //   - "foo.googleapis.com:8080"
 172  //   - "dns:///foo.googleapis.com:8080"
 173  //   - "dns:///foo.googleapis.com"
 174  //   - "dns:///10.0.0.213:8080"
 175  //   - "dns:///%5B2001:db8:85a3:8d3:1319:8a2e:370:7348%5D:443"
 176  //   - "dns://8.8.8.8/foo.googleapis.com:8080"
 177  //   - "dns://8.8.8.8/foo.googleapis.com"
 178  //   - "zookeeper://zk.example.com:9900/example_service"
 179  //
 180  // The DialOptions returned by WithBlock, WithTimeout,
 181  // WithReturnConnectionError, and FailOnNonTempDialError are ignored by this
 182  // function.
 183  func NewClient(target string, opts ...DialOption) (conn *ClientConn, err error) {
 184  	cc := &ClientConn{
 185  		target: target,
 186  		conns:  make(map[*addrConn]struct{}),
 187  		dopts:  defaultDialOptions(),
 188  	}
 189  
 190  	cc.retryThrottler.Store((*retryThrottler)(nil))
 191  	cc.safeConfigSelector.UpdateConfigSelector(&defaultConfigSelector{nil})
 192  	cc.ctx, cc.cancel = context.WithCancel(context.Background())
 193  
 194  	// Apply dial options.
 195  	disableGlobalOpts := false
 196  	for _, opt := range opts {
 197  		if _, ok := opt.(*disableGlobalDialOptions); ok {
 198  			disableGlobalOpts = true
 199  			break
 200  		}
 201  	}
 202  
 203  	if !disableGlobalOpts {
 204  		for _, opt := range globalDialOptions {
 205  			opt.apply(&cc.dopts)
 206  		}
 207  	}
 208  
 209  	for _, opt := range opts {
 210  		opt.apply(&cc.dopts)
 211  	}
 212  
 213  	// Determine the resolver to use.
 214  	if err := cc.initParsedTargetAndResolverBuilder(); err != nil {
 215  		return nil, err
 216  	}
 217  
 218  	for _, opt := range globalPerTargetDialOptions {
 219  		opt.DialOptionForTarget(cc.parsedTarget.URL).apply(&cc.dopts)
 220  	}
 221  
 222  	chainUnaryClientInterceptors(cc)
 223  	chainStreamClientInterceptors(cc)
 224  
 225  	if err := cc.validateTransportCredentials(); err != nil {
 226  		return nil, err
 227  	}
 228  
 229  	if cc.dopts.defaultServiceConfigRawJSON != nil {
 230  		scpr := parseServiceConfig(*cc.dopts.defaultServiceConfigRawJSON, cc.dopts.maxCallAttempts)
 231  		if scpr.Err != nil {
 232  			return nil, fmt.Errorf("%s: %v", invalidDefaultServiceConfigErrPrefix, scpr.Err)
 233  		}
 234  		cc.dopts.defaultServiceConfig, _ = scpr.Config.(*ServiceConfig)
 235  	}
 236  	cc.keepaliveParams = cc.dopts.copts.KeepaliveParams
 237  
 238  	if err = cc.initAuthority(); err != nil {
 239  		return nil, err
 240  	}
 241  
 242  	// Register ClientConn with channelz. Note that this is only done after
 243  	// channel creation cannot fail.
 244  	cc.channelzRegistration(target)
 245  	channelz.Infof(logger, cc.channelz, "parsed dial target is: %#v", cc.parsedTarget)
 246  	channelz.Infof(logger, cc.channelz, "Channel authority set to %q", cc.authority)
 247  
 248  	cc.csMgr = newConnectivityStateManager(cc.ctx, cc.channelz)
 249  	cc.pickerWrapper = newPickerWrapper()
 250  
 251  	cc.metricsRecorderList = istats.NewMetricsRecorderList(cc.dopts.copts.StatsHandlers)
 252  	cc.statsHandler = istats.NewCombinedHandler(cc.dopts.copts.StatsHandlers...)
 253  
 254  	cc.initIdleStateLocked() // Safe to call without the lock, since nothing else has a reference to cc.
 255  	cc.idlenessMgr = idle.NewManager((*idler)(cc), cc.dopts.idleTimeout)
 256  
 257  	return cc, nil
 258  }
 259  
 260  // Dial calls DialContext(context.Background(), target, opts...).
 261  //
 262  // Deprecated: use NewClient instead.  Will be supported throughout 1.x.
 263  func Dial(target string, opts ...DialOption) (*ClientConn, error) {
 264  	return DialContext(context.Background(), target, opts...)
 265  }
 266  
 267  // DialContext calls NewClient and then exits idle mode.  If WithBlock(true) is
 268  // used, it calls Connect and WaitForStateChange until either the context
 269  // expires or the state of the ClientConn is Ready.
 270  //
 271  // One subtle difference between NewClient and Dial and DialContext is that the
 272  // former uses "dns" as the default name resolver, while the latter use
 273  // "passthrough" for backward compatibility.  This distinction should not matter
 274  // to most users, but could matter to legacy users that specify a custom dialer
 275  // and expect it to receive the target string directly.
 276  //
 277  // Deprecated: use NewClient instead.  Will be supported throughout 1.x.
 278  func DialContext(ctx context.Context, target string, opts ...DialOption) (conn *ClientConn, err error) {
 279  	// At the end of this method, we kick the channel out of idle, rather than
 280  	// waiting for the first rpc.
 281  	//
 282  	// WithLocalDNSResolution dial option in `grpc.Dial` ensures that it
 283  	// preserves behavior: when default scheme passthrough is used, skip
 284  	// hostname resolution, when "dns" is used for resolution, perform
 285  	// resolution on the client.
 286  	opts = append([]DialOption{withDefaultScheme("passthrough"), WithLocalDNSResolution()}, opts...)
 287  	cc, err := NewClient(target, opts...)
 288  	if err != nil {
 289  		return nil, err
 290  	}
 291  
 292  	// We start the channel off in idle mode, but kick it out of idle now,
 293  	// instead of waiting for the first RPC.  This is the legacy behavior of
 294  	// Dial.
 295  	defer func() {
 296  		if err != nil {
 297  			cc.Close()
 298  		}
 299  	}()
 300  
 301  	// This creates the name resolver, load balancer, etc.
 302  	if err := cc.exitIdleMode(); err != nil {
 303  		return nil, fmt.Errorf("failed to exit idle mode: %w", err)
 304  	}
 305  	cc.idlenessMgr.UnsafeSetNotIdle()
 306  
 307  	// Return now for non-blocking dials.
 308  	if !cc.dopts.block {
 309  		return cc, nil
 310  	}
 311  
 312  	if cc.dopts.timeout > 0 {
 313  		var cancel context.CancelFunc
 314  		ctx, cancel = context.WithTimeout(ctx, cc.dopts.timeout)
 315  		defer cancel()
 316  	}
 317  	defer func() {
 318  		select {
 319  		case <-ctx.Done():
 320  			switch {
 321  			case ctx.Err() == err:
 322  				conn = nil
 323  			case err == nil || !cc.dopts.returnLastError:
 324  				conn, err = nil, ctx.Err()
 325  			default:
 326  				conn, err = nil, fmt.Errorf("%v: %v", ctx.Err(), err)
 327  			}
 328  		default:
 329  		}
 330  	}()
 331  
 332  	// A blocking dial blocks until the clientConn is ready.
 333  	for {
 334  		s := cc.GetState()
 335  		if s == connectivity.Idle {
 336  			cc.Connect()
 337  		}
 338  		if s == connectivity.Ready {
 339  			return cc, nil
 340  		} else if cc.dopts.copts.FailOnNonTempDialError && s == connectivity.TransientFailure {
 341  			if err = cc.connectionError(); err != nil {
 342  				terr, ok := err.(interface {
 343  					Temporary() bool
 344  				})
 345  				if ok && !terr.Temporary() {
 346  					return nil, err
 347  				}
 348  			}
 349  		}
 350  		if !cc.WaitForStateChange(ctx, s) {
 351  			// ctx got timeout or canceled.
 352  			if err = cc.connectionError(); err != nil && cc.dopts.returnLastError {
 353  				return nil, err
 354  			}
 355  			return nil, ctx.Err()
 356  		}
 357  	}
 358  }
 359  
 360  // addTraceEvent is a helper method to add a trace event on the channel. If the
 361  // channel is a nested one, the same event is also added on the parent channel.
 362  func (cc *ClientConn) addTraceEvent(msg string) {
 363  	ted := &channelz.TraceEvent{
 364  		Desc:     fmt.Sprintf("Channel %s", msg),
 365  		Severity: channelz.CtInfo,
 366  	}
 367  	if cc.dopts.channelzParent != nil {
 368  		ted.Parent = &channelz.TraceEvent{
 369  			Desc:     fmt.Sprintf("Nested channel(id:%d) %s", cc.channelz.ID, msg),
 370  			Severity: channelz.CtInfo,
 371  		}
 372  	}
 373  	channelz.AddTraceEvent(logger, cc.channelz, 1, ted)
 374  }
 375  
 376  type idler ClientConn
 377  
 378  func (i *idler) EnterIdleMode() {
 379  	(*ClientConn)(i).enterIdleMode()
 380  }
 381  
 382  func (i *idler) ExitIdleMode() {
 383  	// Ignore the error returned from this method, because from the perspective
 384  	// of the caller (idleness manager), the channel would have always moved out
 385  	// of IDLE by the time this method returns.
 386  	(*ClientConn)(i).exitIdleMode()
 387  }
 388  
 389  // exitIdleMode moves the channel out of idle mode by recreating the name
 390  // resolver and load balancer.  This should never be called directly; use
 391  // cc.idlenessMgr.ExitIdleMode instead.
 392  func (cc *ClientConn) exitIdleMode() error {
 393  	cc.mu.Lock()
 394  	if cc.conns == nil {
 395  		cc.mu.Unlock()
 396  		return errConnClosing
 397  	}
 398  	cc.mu.Unlock()
 399  
 400  	// Set state to CONNECTING before building the name resolver
 401  	// so the channel does not remain in IDLE.
 402  	cc.csMgr.updateState(connectivity.Connecting)
 403  
 404  	// This needs to be called without cc.mu because this builds a new resolver
 405  	// which might update state or report error inline, which would then need to
 406  	// acquire cc.mu.
 407  	if err := cc.resolverWrapper.start(); err != nil {
 408  		// If resolver creation fails, treat it like an error reported by the
 409  		// resolver before any valid updates. Set channel's state to
 410  		// TransientFailure, and set an erroring picker with the resolver build
 411  		// error, which will returned as part of any subsequent RPCs.
 412  		logger.Warningf("Failed to start resolver: %v", err)
 413  		cc.csMgr.updateState(connectivity.TransientFailure)
 414  		cc.mu.Lock()
 415  		cc.updateResolverStateAndUnlock(resolver.State{}, err)
 416  		return fmt.Errorf("failed to start resolver: %w", err)
 417  	}
 418  
 419  	cc.addTraceEvent("exiting idle mode")
 420  	return nil
 421  }
 422  
 423  // initIdleStateLocked initializes common state to how it should be while idle.
 424  func (cc *ClientConn) initIdleStateLocked() {
 425  	cc.resolverWrapper = newCCResolverWrapper(cc)
 426  	cc.balancerWrapper = newCCBalancerWrapper(cc)
 427  	cc.firstResolveEvent = grpcsync.NewEvent()
 428  	// cc.conns == nil is a proxy for the ClientConn being closed. So, instead
 429  	// of setting it to nil here, we recreate the map. This also means that we
 430  	// don't have to do this when exiting idle mode.
 431  	cc.conns = make(map[*addrConn]struct{})
 432  }
 433  
 434  // enterIdleMode puts the channel in idle mode, and as part of it shuts down the
 435  // name resolver, load balancer, and any subchannels.  This should never be
 436  // called directly; use cc.idlenessMgr.EnterIdleMode instead.
 437  func (cc *ClientConn) enterIdleMode() {
 438  	cc.mu.Lock()
 439  
 440  	if cc.conns == nil {
 441  		cc.mu.Unlock()
 442  		return
 443  	}
 444  
 445  	conns := cc.conns
 446  
 447  	rWrapper := cc.resolverWrapper
 448  	rWrapper.close()
 449  	cc.pickerWrapper.reset()
 450  	bWrapper := cc.balancerWrapper
 451  	bWrapper.close()
 452  	cc.csMgr.updateState(connectivity.Idle)
 453  	cc.addTraceEvent("entering idle mode")
 454  
 455  	cc.initIdleStateLocked()
 456  
 457  	cc.mu.Unlock()
 458  
 459  	// Block until the name resolver and LB policy are closed.
 460  	<-rWrapper.serializer.Done()
 461  	<-bWrapper.serializer.Done()
 462  
 463  	// Close all subchannels after the LB policy is closed.
 464  	for ac := range conns {
 465  		ac.tearDown(errConnIdling)
 466  	}
 467  }
 468  
 469  // validateTransportCredentials performs a series of checks on the configured
 470  // transport credentials. It returns a non-nil error if any of these conditions
 471  // are met:
 472  //   - no transport creds and no creds bundle is configured
 473  //   - both transport creds and creds bundle are configured
 474  //   - creds bundle is configured, but it lacks a transport credentials
 475  //   - insecure transport creds configured alongside call creds that require
 476  //     transport level security
 477  //
 478  // If none of the above conditions are met, the configured credentials are
 479  // deemed valid and a nil error is returned.
 480  func (cc *ClientConn) validateTransportCredentials() error {
 481  	if cc.dopts.copts.TransportCredentials == nil && cc.dopts.copts.CredsBundle == nil {
 482  		return errNoTransportSecurity
 483  	}
 484  	if cc.dopts.copts.TransportCredentials != nil && cc.dopts.copts.CredsBundle != nil {
 485  		return errTransportCredsAndBundle
 486  	}
 487  	if cc.dopts.copts.CredsBundle != nil && cc.dopts.copts.CredsBundle.TransportCredentials() == nil {
 488  		return errNoTransportCredsInBundle
 489  	}
 490  	transportCreds := cc.dopts.copts.TransportCredentials
 491  	if transportCreds == nil {
 492  		transportCreds = cc.dopts.copts.CredsBundle.TransportCredentials()
 493  	}
 494  	if transportCreds.Info().SecurityProtocol == "insecure" {
 495  		for _, cd := range cc.dopts.copts.PerRPCCredentials {
 496  			if cd.RequireTransportSecurity() {
 497  				return errTransportCredentialsMissing
 498  			}
 499  		}
 500  	}
 501  	return nil
 502  }
 503  
 504  // channelzRegistration registers the newly created ClientConn with channelz and
 505  // stores the returned identifier in `cc.channelz`.  A channelz trace event is
 506  // emitted for ClientConn creation. If the newly created ClientConn is a nested
 507  // one, i.e a valid parent ClientConn ID is specified via a dial option, the
 508  // trace event is also added to the parent.
 509  //
 510  // Doesn't grab cc.mu as this method is expected to be called only at Dial time.
 511  func (cc *ClientConn) channelzRegistration(target string) {
 512  	parentChannel, _ := cc.dopts.channelzParent.(*channelz.Channel)
 513  	cc.channelz = channelz.RegisterChannel(parentChannel, target)
 514  	cc.addTraceEvent(fmt.Sprintf("created for target %q", target))
 515  }
 516  
 517  // chainUnaryClientInterceptors chains all unary client interceptors into one.
 518  func chainUnaryClientInterceptors(cc *ClientConn) {
 519  	interceptors := cc.dopts.chainUnaryInts
 520  	// Prepend dopts.unaryInt to the chaining interceptors if it exists, since unaryInt will
 521  	// be executed before any other chained interceptors.
 522  	if cc.dopts.unaryInt != nil {
 523  		interceptors = append([]UnaryClientInterceptor{cc.dopts.unaryInt}, interceptors...)
 524  	}
 525  	var chainedInt UnaryClientInterceptor
 526  	if len(interceptors) == 0 {
 527  		chainedInt = nil
 528  	} else if len(interceptors) == 1 {
 529  		chainedInt = interceptors[0]
 530  	} else {
 531  		chainedInt = func(ctx context.Context, method string, req, reply any, cc *ClientConn, invoker UnaryInvoker, opts ...CallOption) error {
 532  			return interceptors[0](ctx, method, req, reply, cc, getChainUnaryInvoker(interceptors, 0, invoker), opts...)
 533  		}
 534  	}
 535  	cc.dopts.unaryInt = chainedInt
 536  }
 537  
 538  // getChainUnaryInvoker recursively generate the chained unary invoker.
 539  func getChainUnaryInvoker(interceptors []UnaryClientInterceptor, curr int, finalInvoker UnaryInvoker) UnaryInvoker {
 540  	if curr == len(interceptors)-1 {
 541  		return finalInvoker
 542  	}
 543  	return func(ctx context.Context, method string, req, reply any, cc *ClientConn, opts ...CallOption) error {
 544  		return interceptors[curr+1](ctx, method, req, reply, cc, getChainUnaryInvoker(interceptors, curr+1, finalInvoker), opts...)
 545  	}
 546  }
 547  
 548  // chainStreamClientInterceptors chains all stream client interceptors into one.
 549  func chainStreamClientInterceptors(cc *ClientConn) {
 550  	interceptors := cc.dopts.chainStreamInts
 551  	// Prepend dopts.streamInt to the chaining interceptors if it exists, since streamInt will
 552  	// be executed before any other chained interceptors.
 553  	if cc.dopts.streamInt != nil {
 554  		interceptors = append([]StreamClientInterceptor{cc.dopts.streamInt}, interceptors...)
 555  	}
 556  	var chainedInt StreamClientInterceptor
 557  	if len(interceptors) == 0 {
 558  		chainedInt = nil
 559  	} else if len(interceptors) == 1 {
 560  		chainedInt = interceptors[0]
 561  	} else {
 562  		chainedInt = func(ctx context.Context, desc *StreamDesc, cc *ClientConn, method string, streamer Streamer, opts ...CallOption) (ClientStream, error) {
 563  			return interceptors[0](ctx, desc, cc, method, getChainStreamer(interceptors, 0, streamer), opts...)
 564  		}
 565  	}
 566  	cc.dopts.streamInt = chainedInt
 567  }
 568  
 569  // getChainStreamer recursively generate the chained client stream constructor.
 570  func getChainStreamer(interceptors []StreamClientInterceptor, curr int, finalStreamer Streamer) Streamer {
 571  	if curr == len(interceptors)-1 {
 572  		return finalStreamer
 573  	}
 574  	return func(ctx context.Context, desc *StreamDesc, cc *ClientConn, method string, opts ...CallOption) (ClientStream, error) {
 575  		return interceptors[curr+1](ctx, desc, cc, method, getChainStreamer(interceptors, curr+1, finalStreamer), opts...)
 576  	}
 577  }
 578  
 579  // newConnectivityStateManager creates an connectivityStateManager with
 580  // the specified channel.
 581  func newConnectivityStateManager(ctx context.Context, channel *channelz.Channel) *connectivityStateManager {
 582  	return &connectivityStateManager{
 583  		channelz: channel,
 584  		pubSub:   grpcsync.NewPubSub(ctx),
 585  	}
 586  }
 587  
 588  // connectivityStateManager keeps the connectivity.State of ClientConn.
 589  // This struct will eventually be exported so the balancers can access it.
 590  //
 591  // TODO: If possible, get rid of the `connectivityStateManager` type, and
 592  // provide this functionality using the `PubSub`, to avoid keeping track of
 593  // the connectivity state at two places.
 594  type connectivityStateManager struct {
 595  	mu         sync.Mutex
 596  	state      connectivity.State
 597  	notifyChan chan struct{}
 598  	channelz   *channelz.Channel
 599  	pubSub     *grpcsync.PubSub
 600  }
 601  
 602  // updateState updates the connectivity.State of ClientConn.
 603  // If there's a change it notifies goroutines waiting on state change to
 604  // happen.
 605  func (csm *connectivityStateManager) updateState(state connectivity.State) {
 606  	csm.mu.Lock()
 607  	defer csm.mu.Unlock()
 608  	if csm.state == connectivity.Shutdown {
 609  		return
 610  	}
 611  	if csm.state == state {
 612  		return
 613  	}
 614  	csm.state = state
 615  	csm.channelz.ChannelMetrics.State.Store(&state)
 616  	csm.pubSub.Publish(state)
 617  
 618  	channelz.Infof(logger, csm.channelz, "Channel Connectivity change to %v", state)
 619  	if csm.notifyChan != nil {
 620  		// There are other goroutines waiting on this channel.
 621  		close(csm.notifyChan)
 622  		csm.notifyChan = nil
 623  	}
 624  }
 625  
 626  func (csm *connectivityStateManager) getState() connectivity.State {
 627  	csm.mu.Lock()
 628  	defer csm.mu.Unlock()
 629  	return csm.state
 630  }
 631  
 632  func (csm *connectivityStateManager) getNotifyChan() <-chan struct{} {
 633  	csm.mu.Lock()
 634  	defer csm.mu.Unlock()
 635  	if csm.notifyChan == nil {
 636  		csm.notifyChan = make(chan struct{})
 637  	}
 638  	return csm.notifyChan
 639  }
 640  
 641  // ClientConnInterface defines the functions clients need to perform unary and
 642  // streaming RPCs.  It is implemented by *ClientConn, and is only intended to
 643  // be referenced by generated code.
 644  type ClientConnInterface interface {
 645  	// Invoke performs a unary RPC and returns after the response is received
 646  	// into reply.
 647  	Invoke(ctx context.Context, method string, args any, reply any, opts ...CallOption) error
 648  	// NewStream begins a streaming RPC.
 649  	NewStream(ctx context.Context, desc *StreamDesc, method string, opts ...CallOption) (ClientStream, error)
 650  }
 651  
 652  // Assert *ClientConn implements ClientConnInterface.
 653  var _ ClientConnInterface = (*ClientConn)(nil)
 654  
 655  // ClientConn represents a virtual connection to a conceptual endpoint, to
 656  // perform RPCs.
 657  //
 658  // A ClientConn is free to have zero or more actual connections to the endpoint
 659  // based on configuration, load, etc. It is also free to determine which actual
 660  // endpoints to use and may change it every RPC, permitting client-side load
 661  // balancing.
 662  //
 663  // A ClientConn encapsulates a range of functionality including name
 664  // resolution, TCP connection establishment (with retries and backoff) and TLS
 665  // handshakes. It also handles errors on established connections by
 666  // re-resolving the name and reconnecting.
 667  type ClientConn struct {
 668  	ctx    context.Context    // Initialized using the background context at dial time.
 669  	cancel context.CancelFunc // Cancelled on close.
 670  
 671  	// The following are initialized at dial time, and are read-only after that.
 672  	target              string            // User's dial target.
 673  	parsedTarget        resolver.Target   // See initParsedTargetAndResolverBuilder().
 674  	authority           string            // See initAuthority().
 675  	dopts               dialOptions       // Default and user specified dial options.
 676  	channelz            *channelz.Channel // Channelz object.
 677  	resolverBuilder     resolver.Builder  // See initParsedTargetAndResolverBuilder().
 678  	idlenessMgr         *idle.Manager
 679  	metricsRecorderList *istats.MetricsRecorderList
 680  	statsHandler        stats.Handler
 681  
 682  	// The following provide their own synchronization, and therefore don't
 683  	// require cc.mu to be held to access them.
 684  	csMgr              *connectivityStateManager
 685  	pickerWrapper      *pickerWrapper
 686  	safeConfigSelector iresolver.SafeConfigSelector
 687  	retryThrottler     atomic.Value // Updated from service config.
 688  
 689  	// mu protects the following fields.
 690  	// TODO: split mu so the same mutex isn't used for everything.
 691  	mu              sync.RWMutex
 692  	resolverWrapper *ccResolverWrapper         // Always recreated whenever entering idle to simplify Close.
 693  	balancerWrapper *ccBalancerWrapper         // Always recreated whenever entering idle to simplify Close.
 694  	sc              *ServiceConfig             // Latest service config received from the resolver.
 695  	conns           map[*addrConn]struct{}     // Set to nil on close.
 696  	keepaliveParams keepalive.ClientParameters // May be updated upon receipt of a GoAway.
 697  	// firstResolveEvent is used to track whether the name resolver sent us at
 698  	// least one update. RPCs block on this event.  May be accessed without mu
 699  	// if we know we cannot be asked to enter idle mode while accessing it (e.g.
 700  	// when the idle manager has already been closed, or if we are already
 701  	// entering idle mode).
 702  	firstResolveEvent *grpcsync.Event
 703  
 704  	lceMu               sync.Mutex // protects lastConnectionError
 705  	lastConnectionError error
 706  }
 707  
 708  // WaitForStateChange waits until the connectivity.State of ClientConn changes from sourceState or
 709  // ctx expires. A true value is returned in former case and false in latter.
 710  func (cc *ClientConn) WaitForStateChange(ctx context.Context, sourceState connectivity.State) bool {
 711  	ch := cc.csMgr.getNotifyChan()
 712  	if cc.csMgr.getState() != sourceState {
 713  		return true
 714  	}
 715  	select {
 716  	case <-ctx.Done():
 717  		return false
 718  	case <-ch:
 719  		return true
 720  	}
 721  }
 722  
 723  // GetState returns the connectivity.State of ClientConn.
 724  func (cc *ClientConn) GetState() connectivity.State {
 725  	return cc.csMgr.getState()
 726  }
 727  
 728  // Connect causes all subchannels in the ClientConn to attempt to connect if
 729  // the channel is idle.  Does not wait for the connection attempts to begin
 730  // before returning.
 731  //
 732  // # Experimental
 733  //
 734  // Notice: This API is EXPERIMENTAL and may be changed or removed in a later
 735  // release.
 736  func (cc *ClientConn) Connect() {
 737  	cc.idlenessMgr.ExitIdleMode()
 738  
 739  	// If the ClientConn was not in idle mode, we need to call ExitIdle on the
 740  	// LB policy so that connections can be created.
 741  	cc.mu.Lock()
 742  	cc.balancerWrapper.exitIdle()
 743  	cc.mu.Unlock()
 744  }
 745  
 746  // waitForResolvedAddrs blocks until the resolver provides addresses or the
 747  // context expires, whichever happens first.
 748  //
 749  // Error is nil unless the context expires first; otherwise returns a status
 750  // error based on the context.
 751  //
 752  // The returned boolean indicates whether it did block or not. If the
 753  // resolution has already happened once before, it returns false without
 754  // blocking. Otherwise, it wait for the resolution and return true if
 755  // resolution has succeeded or return false along with error if resolution has
 756  // failed.
 757  func (cc *ClientConn) waitForResolvedAddrs(ctx context.Context) (bool, error) {
 758  	// This is on the RPC path, so we use a fast path to avoid the
 759  	// more-expensive "select" below after the resolver has returned once.
 760  	if cc.firstResolveEvent.HasFired() {
 761  		return false, nil
 762  	}
 763  	internal.NewStreamWaitingForResolver()
 764  	select {
 765  	case <-cc.firstResolveEvent.Done():
 766  		return true, nil
 767  	case <-ctx.Done():
 768  		return false, status.FromContextError(ctx.Err()).Err()
 769  	case <-cc.ctx.Done():
 770  		return false, ErrClientConnClosing
 771  	}
 772  }
 773  
 774  var emptyServiceConfig *ServiceConfig
 775  
 776  func init() {
 777  	cfg := parseServiceConfig("{}", defaultMaxCallAttempts)
 778  	if cfg.Err != nil {
 779  		panic(fmt.Sprintf("impossible error parsing empty service config: %v", cfg.Err))
 780  	}
 781  	emptyServiceConfig = cfg.Config.(*ServiceConfig)
 782  
 783  	internal.SubscribeToConnectivityStateChanges = func(cc *ClientConn, s grpcsync.Subscriber) func() {
 784  		return cc.csMgr.pubSub.Subscribe(s)
 785  	}
 786  	internal.EnterIdleModeForTesting = func(cc *ClientConn) {
 787  		cc.idlenessMgr.EnterIdleModeForTesting()
 788  	}
 789  	internal.ExitIdleModeForTesting = func(cc *ClientConn) {
 790  		cc.idlenessMgr.ExitIdleMode()
 791  	}
 792  }
 793  
 794  func (cc *ClientConn) maybeApplyDefaultServiceConfig() {
 795  	if cc.sc != nil {
 796  		cc.applyServiceConfigAndBalancer(cc.sc, nil)
 797  		return
 798  	}
 799  	if cc.dopts.defaultServiceConfig != nil {
 800  		cc.applyServiceConfigAndBalancer(cc.dopts.defaultServiceConfig, &defaultConfigSelector{cc.dopts.defaultServiceConfig})
 801  	} else {
 802  		cc.applyServiceConfigAndBalancer(emptyServiceConfig, &defaultConfigSelector{emptyServiceConfig})
 803  	}
 804  }
 805  
 806  func (cc *ClientConn) updateResolverStateAndUnlock(s resolver.State, err error) error {
 807  	defer cc.firstResolveEvent.Fire()
 808  	// Check if the ClientConn is already closed. Some fields (e.g.
 809  	// balancerWrapper) are set to nil when closing the ClientConn, and could
 810  	// cause nil pointer panic if we don't have this check.
 811  	if cc.conns == nil {
 812  		cc.mu.Unlock()
 813  		return nil
 814  	}
 815  
 816  	if err != nil {
 817  		// May need to apply the initial service config in case the resolver
 818  		// doesn't support service configs, or doesn't provide a service config
 819  		// with the new addresses.
 820  		cc.maybeApplyDefaultServiceConfig()
 821  
 822  		cc.balancerWrapper.resolverError(err)
 823  
 824  		// No addresses are valid with err set; return early.
 825  		cc.mu.Unlock()
 826  		return balancer.ErrBadResolverState
 827  	}
 828  
 829  	var ret error
 830  	if cc.dopts.disableServiceConfig {
 831  		channelz.Infof(logger, cc.channelz, "ignoring service config from resolver (%v) and applying the default because service config is disabled", s.ServiceConfig)
 832  		cc.maybeApplyDefaultServiceConfig()
 833  	} else if s.ServiceConfig == nil {
 834  		cc.maybeApplyDefaultServiceConfig()
 835  		// TODO: do we need to apply a failing LB policy if there is no
 836  		// default, per the error handling design?
 837  	} else {
 838  		if sc, ok := s.ServiceConfig.Config.(*ServiceConfig); s.ServiceConfig.Err == nil && ok {
 839  			configSelector := iresolver.GetConfigSelector(s)
 840  			if configSelector != nil {
 841  				if len(s.ServiceConfig.Config.(*ServiceConfig).Methods) != 0 {
 842  					channelz.Infof(logger, cc.channelz, "method configs in service config will be ignored due to presence of config selector")
 843  				}
 844  			} else {
 845  				configSelector = &defaultConfigSelector{sc}
 846  			}
 847  			cc.applyServiceConfigAndBalancer(sc, configSelector)
 848  		} else {
 849  			ret = balancer.ErrBadResolverState
 850  			if cc.sc == nil {
 851  				// Apply the failing LB only if we haven't received valid service config
 852  				// from the name resolver in the past.
 853  				cc.applyFailingLBLocked(s.ServiceConfig)
 854  				cc.mu.Unlock()
 855  				return ret
 856  			}
 857  		}
 858  	}
 859  
 860  	balCfg := cc.sc.lbConfig
 861  	bw := cc.balancerWrapper
 862  	cc.mu.Unlock()
 863  
 864  	uccsErr := bw.updateClientConnState(&balancer.ClientConnState{ResolverState: s, BalancerConfig: balCfg})
 865  	if ret == nil {
 866  		ret = uccsErr // prefer ErrBadResolver state since any other error is
 867  		// currently meaningless to the caller.
 868  	}
 869  	return ret
 870  }
 871  
 872  // applyFailingLBLocked is akin to configuring an LB policy on the channel which
 873  // always fails RPCs. Here, an actual LB policy is not configured, but an always
 874  // erroring picker is configured, which returns errors with information about
 875  // what was invalid in the received service config. A config selector with no
 876  // service config is configured, and the connectivity state of the channel is
 877  // set to TransientFailure.
 878  func (cc *ClientConn) applyFailingLBLocked(sc *serviceconfig.ParseResult) {
 879  	var err error
 880  	if sc.Err != nil {
 881  		err = status.Errorf(codes.Unavailable, "error parsing service config: %v", sc.Err)
 882  	} else {
 883  		err = status.Errorf(codes.Unavailable, "illegal service config type: %T", sc.Config)
 884  	}
 885  	cc.safeConfigSelector.UpdateConfigSelector(&defaultConfigSelector{nil})
 886  	cc.pickerWrapper.updatePicker(base.NewErrPicker(err))
 887  	cc.csMgr.updateState(connectivity.TransientFailure)
 888  }
 889  
 890  // Makes a copy of the input addresses slice. Addresses are passed during
 891  // subconn creation and address update operations.
 892  func copyAddresses(in []resolver.Address) []resolver.Address {
 893  	out := make([]resolver.Address, len(in))
 894  	copy(out, in)
 895  	return out
 896  }
 897  
 898  // newAddrConnLocked creates an addrConn for addrs and adds it to cc.conns.
 899  //
 900  // Caller needs to make sure len(addrs) > 0.
 901  func (cc *ClientConn) newAddrConnLocked(addrs []resolver.Address, opts balancer.NewSubConnOptions) (*addrConn, error) {
 902  	if cc.conns == nil {
 903  		return nil, ErrClientConnClosing
 904  	}
 905  
 906  	ac := &addrConn{
 907  		state:        connectivity.Idle,
 908  		cc:           cc,
 909  		addrs:        copyAddresses(addrs),
 910  		scopts:       opts,
 911  		dopts:        cc.dopts,
 912  		channelz:     channelz.RegisterSubChannel(cc.channelz, ""),
 913  		resetBackoff: make(chan struct{}),
 914  	}
 915  	ac.updateTelemetryLabelsLocked()
 916  	ac.ctx, ac.cancel = context.WithCancel(cc.ctx)
 917  	// Start with our address set to the first address; this may be updated if
 918  	// we connect to different addresses.
 919  	ac.channelz.ChannelMetrics.Target.Store(&addrs[0].Addr)
 920  
 921  	channelz.AddTraceEvent(logger, ac.channelz, 0, &channelz.TraceEvent{
 922  		Desc:     "Subchannel created",
 923  		Severity: channelz.CtInfo,
 924  		Parent: &channelz.TraceEvent{
 925  			Desc:     fmt.Sprintf("Subchannel(id:%d) created", ac.channelz.ID),
 926  			Severity: channelz.CtInfo,
 927  		},
 928  	})
 929  
 930  	// Track ac in cc. This needs to be done before any getTransport(...) is called.
 931  	cc.conns[ac] = struct{}{}
 932  	return ac, nil
 933  }
 934  
 935  // removeAddrConn removes the addrConn in the subConn from clientConn.
 936  // It also tears down the ac with the given error.
 937  func (cc *ClientConn) removeAddrConn(ac *addrConn, err error) {
 938  	cc.mu.Lock()
 939  	if cc.conns == nil {
 940  		cc.mu.Unlock()
 941  		return
 942  	}
 943  	delete(cc.conns, ac)
 944  	cc.mu.Unlock()
 945  	ac.tearDown(err)
 946  }
 947  
 948  // Target returns the target string of the ClientConn.
 949  func (cc *ClientConn) Target() string {
 950  	return cc.target
 951  }
 952  
 953  // CanonicalTarget returns the canonical target string used when creating cc.
 954  //
 955  // This always has the form "<scheme>://[authority]/<endpoint>".  For example:
 956  //
 957  //   - "dns:///example.com:42"
 958  //   - "dns://8.8.8.8/example.com:42"
 959  //   - "unix:///path/to/socket"
 960  func (cc *ClientConn) CanonicalTarget() string {
 961  	return cc.parsedTarget.String()
 962  }
 963  
 964  func (cc *ClientConn) incrCallsStarted() {
 965  	cc.channelz.ChannelMetrics.CallsStarted.Add(1)
 966  	cc.channelz.ChannelMetrics.LastCallStartedTimestamp.Store(time.Now().UnixNano())
 967  }
 968  
 969  func (cc *ClientConn) incrCallsSucceeded() {
 970  	cc.channelz.ChannelMetrics.CallsSucceeded.Add(1)
 971  }
 972  
 973  func (cc *ClientConn) incrCallsFailed() {
 974  	cc.channelz.ChannelMetrics.CallsFailed.Add(1)
 975  }
 976  
 977  // connect starts creating a transport.
 978  // It does nothing if the ac is not IDLE.
 979  // TODO(bar) Move this to the addrConn section.
 980  func (ac *addrConn) connect() error {
 981  	ac.mu.Lock()
 982  	if ac.state == connectivity.Shutdown {
 983  		if logger.V(2) {
 984  			logger.Infof("connect called on shutdown addrConn; ignoring.")
 985  		}
 986  		ac.mu.Unlock()
 987  		return errConnClosing
 988  	}
 989  	if ac.state != connectivity.Idle {
 990  		if logger.V(2) {
 991  			logger.Infof("connect called on addrConn in non-idle state (%v); ignoring.", ac.state)
 992  		}
 993  		ac.mu.Unlock()
 994  		return nil
 995  	}
 996  
 997  	ac.resetTransportAndUnlock()
 998  	return nil
 999  }
1000  
1001  // equalAddressIgnoringBalAttributes returns true is a and b are considered equal.
1002  // This is different from the Equal method on the resolver.Address type which
1003  // considers all fields to determine equality. Here, we only consider fields
1004  // that are meaningful to the subConn.
1005  func equalAddressIgnoringBalAttributes(a, b *resolver.Address) bool {
1006  	return a.Addr == b.Addr && a.ServerName == b.ServerName &&
1007  		a.Attributes.Equal(b.Attributes) &&
1008  		a.Metadata == b.Metadata
1009  }
1010  
1011  func equalAddressesIgnoringBalAttributes(a, b []resolver.Address) bool {
1012  	return slices.EqualFunc(a, b, func(a, b resolver.Address) bool { return equalAddressIgnoringBalAttributes(&a, &b) })
1013  }
1014  
1015  // updateAddrs updates ac.addrs with the new addresses list and handles active
1016  // connections or connection attempts.
1017  func (ac *addrConn) updateAddrs(addrs []resolver.Address) {
1018  	addrs = copyAddresses(addrs)
1019  	limit := len(addrs)
1020  	if limit > 5 {
1021  		limit = 5
1022  	}
1023  	channelz.Infof(logger, ac.channelz, "addrConn: updateAddrs addrs (%d of %d): %v", limit, len(addrs), addrs[:limit])
1024  
1025  	ac.mu.Lock()
1026  	if equalAddressesIgnoringBalAttributes(ac.addrs, addrs) {
1027  		ac.mu.Unlock()
1028  		return
1029  	}
1030  
1031  	ac.addrs = addrs
1032  	ac.updateTelemetryLabelsLocked()
1033  	if ac.state == connectivity.Shutdown ||
1034  		ac.state == connectivity.TransientFailure ||
1035  		ac.state == connectivity.Idle {
1036  		// We were not connecting, so do nothing but update the addresses.
1037  		ac.mu.Unlock()
1038  		return
1039  	}
1040  
1041  	if ac.state == connectivity.Ready {
1042  		// Try to find the connected address.
1043  		for _, a := range addrs {
1044  			a.ServerName = ac.cc.getServerName(a)
1045  			if equalAddressIgnoringBalAttributes(&a, &ac.curAddr) {
1046  				// We are connected to a valid address, so do nothing but
1047  				// update the addresses.
1048  				ac.mu.Unlock()
1049  				return
1050  			}
1051  		}
1052  	}
1053  
1054  	// We are either connected to the wrong address or currently connecting.
1055  	// Stop the current iteration and restart.
1056  
1057  	ac.cancel()
1058  	ac.ctx, ac.cancel = context.WithCancel(ac.cc.ctx)
1059  
1060  	// We have to defer here because GracefulClose => onClose, which requires
1061  	// locking ac.mu.
1062  	if ac.transport != nil {
1063  		defer ac.transport.GracefulClose()
1064  		ac.transport = nil
1065  	}
1066  
1067  	if len(addrs) == 0 {
1068  		ac.updateConnectivityState(connectivity.Idle, nil)
1069  	}
1070  
1071  	// Since we were connecting/connected, we should start a new connection
1072  	// attempt.
1073  	go ac.resetTransportAndUnlock()
1074  }
1075  
1076  // getServerName determines the serverName to be used in the connection
1077  // handshake. The default value for the serverName is the authority on the
1078  // ClientConn, which either comes from the user's dial target or through an
1079  // authority override specified using the WithAuthority dial option. Name
1080  // resolvers can specify a per-address override for the serverName through the
1081  // resolver.Address.ServerName field which is used only if the WithAuthority
1082  // dial option was not used. The rationale is that per-address authority
1083  // overrides specified by the name resolver can represent a security risk, while
1084  // an override specified by the user is more dependable since they probably know
1085  // what they are doing.
1086  func (cc *ClientConn) getServerName(addr resolver.Address) string {
1087  	if cc.dopts.authority != "" {
1088  		return cc.dopts.authority
1089  	}
1090  	if addr.ServerName != "" {
1091  		return addr.ServerName
1092  	}
1093  	return cc.authority
1094  }
1095  
1096  func getMethodConfig(sc *ServiceConfig, method string) MethodConfig {
1097  	if sc == nil {
1098  		return MethodConfig{}
1099  	}
1100  	if m, ok := sc.Methods[method]; ok {
1101  		return m
1102  	}
1103  	i := strings.LastIndex(method, "/")
1104  	if m, ok := sc.Methods[method[:i+1]]; ok {
1105  		return m
1106  	}
1107  	return sc.Methods[""]
1108  }
1109  
1110  // GetMethodConfig gets the method config of the input method.
1111  // If there's an exact match for input method (i.e. /service/method), we return
1112  // the corresponding MethodConfig.
1113  // If there isn't an exact match for the input method, we look for the service's default
1114  // config under the service (i.e /service/) and then for the default for all services (empty string).
1115  //
1116  // If there is a default MethodConfig for the service, we return it.
1117  // Otherwise, we return an empty MethodConfig.
1118  func (cc *ClientConn) GetMethodConfig(method string) MethodConfig {
1119  	// TODO: Avoid the locking here.
1120  	cc.mu.RLock()
1121  	defer cc.mu.RUnlock()
1122  	return getMethodConfig(cc.sc, method)
1123  }
1124  
1125  func (cc *ClientConn) healthCheckConfig() *healthCheckConfig {
1126  	cc.mu.RLock()
1127  	defer cc.mu.RUnlock()
1128  	if cc.sc == nil {
1129  		return nil
1130  	}
1131  	return cc.sc.healthCheckConfig
1132  }
1133  
1134  func (cc *ClientConn) applyServiceConfigAndBalancer(sc *ServiceConfig, configSelector iresolver.ConfigSelector) {
1135  	if sc == nil {
1136  		// should never reach here.
1137  		return
1138  	}
1139  	cc.sc = sc
1140  	if configSelector != nil {
1141  		cc.safeConfigSelector.UpdateConfigSelector(configSelector)
1142  	}
1143  
1144  	if cc.sc.retryThrottling != nil {
1145  		newThrottler := &retryThrottler{
1146  			tokens: cc.sc.retryThrottling.MaxTokens,
1147  			max:    cc.sc.retryThrottling.MaxTokens,
1148  			thresh: cc.sc.retryThrottling.MaxTokens / 2,
1149  			ratio:  cc.sc.retryThrottling.TokenRatio,
1150  		}
1151  		cc.retryThrottler.Store(newThrottler)
1152  	} else {
1153  		cc.retryThrottler.Store((*retryThrottler)(nil))
1154  	}
1155  }
1156  
1157  func (cc *ClientConn) resolveNow(o resolver.ResolveNowOptions) {
1158  	cc.mu.RLock()
1159  	cc.resolverWrapper.resolveNow(o)
1160  	cc.mu.RUnlock()
1161  }
1162  
1163  func (cc *ClientConn) resolveNowLocked(o resolver.ResolveNowOptions) {
1164  	cc.resolverWrapper.resolveNow(o)
1165  }
1166  
1167  // ResetConnectBackoff wakes up all subchannels in transient failure and causes
1168  // them to attempt another connection immediately.  It also resets the backoff
1169  // times used for subsequent attempts regardless of the current state.
1170  //
1171  // In general, this function should not be used.  Typical service or network
1172  // outages result in a reasonable client reconnection strategy by default.
1173  // However, if a previously unavailable network becomes available, this may be
1174  // used to trigger an immediate reconnect.
1175  //
1176  // # Experimental
1177  //
1178  // Notice: This API is EXPERIMENTAL and may be changed or removed in a
1179  // later release.
1180  func (cc *ClientConn) ResetConnectBackoff() {
1181  	cc.mu.Lock()
1182  	conns := cc.conns
1183  	cc.mu.Unlock()
1184  	for ac := range conns {
1185  		ac.resetConnectBackoff()
1186  	}
1187  }
1188  
1189  // Close tears down the ClientConn and all underlying connections.
1190  func (cc *ClientConn) Close() error {
1191  	defer func() {
1192  		cc.cancel()
1193  		<-cc.csMgr.pubSub.Done()
1194  	}()
1195  
1196  	// Prevent calls to enter/exit idle immediately, and ensure we are not
1197  	// currently entering/exiting idle mode.
1198  	cc.idlenessMgr.Close()
1199  
1200  	cc.mu.Lock()
1201  	if cc.conns == nil {
1202  		cc.mu.Unlock()
1203  		return ErrClientConnClosing
1204  	}
1205  
1206  	conns := cc.conns
1207  	cc.conns = nil
1208  	cc.csMgr.updateState(connectivity.Shutdown)
1209  
1210  	// We can safely unlock and continue to access all fields now as
1211  	// cc.conns==nil, preventing any further operations on cc.
1212  	cc.mu.Unlock()
1213  
1214  	cc.resolverWrapper.close()
1215  	// The order of closing matters here since the balancer wrapper assumes the
1216  	// picker is closed before it is closed.
1217  	cc.pickerWrapper.close()
1218  	cc.balancerWrapper.close()
1219  
1220  	<-cc.resolverWrapper.serializer.Done()
1221  	<-cc.balancerWrapper.serializer.Done()
1222  	var wg sync.WaitGroup
1223  	for ac := range conns {
1224  		wg.Add(1)
1225  		go func(ac *addrConn) {
1226  			defer wg.Done()
1227  			ac.tearDown(ErrClientConnClosing)
1228  		}(ac)
1229  	}
1230  	wg.Wait()
1231  	cc.addTraceEvent("deleted")
1232  	// TraceEvent needs to be called before RemoveEntry, as TraceEvent may add
1233  	// trace reference to the entity being deleted, and thus prevent it from being
1234  	// deleted right away.
1235  	channelz.RemoveEntry(cc.channelz.ID)
1236  
1237  	return nil
1238  }
1239  
1240  // addrConn is a network connection to a given address.
1241  type addrConn struct {
1242  	ctx    context.Context
1243  	cancel context.CancelFunc
1244  
1245  	cc     *ClientConn
1246  	dopts  dialOptions
1247  	acbw   *acBalancerWrapper
1248  	scopts balancer.NewSubConnOptions
1249  
1250  	// transport is set when there's a viable transport (note: ac state may not be READY as LB channel
1251  	// health checking may require server to report healthy to set ac to READY), and is reset
1252  	// to nil when the current transport should no longer be used to create a stream (e.g. after GoAway
1253  	// is received, transport is closed, ac has been torn down).
1254  	transport transport.ClientTransport // The current transport.
1255  
1256  	// This mutex is used on the RPC path, so its usage should be minimized as
1257  	// much as possible.
1258  	// TODO: Find a lock-free way to retrieve the transport and state from the
1259  	// addrConn.
1260  	mu      sync.Mutex
1261  	curAddr resolver.Address   // The current address.
1262  	addrs   []resolver.Address // All addresses that the resolver resolved to.
1263  
1264  	// Use updateConnectivityState for updating addrConn's connectivity state.
1265  	state connectivity.State
1266  
1267  	backoffIdx   int // Needs to be stateful for resetConnectBackoff.
1268  	resetBackoff chan struct{}
1269  
1270  	channelz *channelz.SubChannel
1271  
1272  	localityLabel       string
1273  	backendServiceLabel string
1274  }
1275  
1276  // Note: this requires a lock on ac.mu.
1277  func (ac *addrConn) updateConnectivityState(s connectivity.State, lastErr error) {
1278  	if ac.state == s {
1279  		return
1280  	}
1281  
1282  	// If we are transitioning out of Ready, it means there is a disconnection.
1283  	// A SubConn can also transition from CONNECTING directly to IDLE when
1284  	// a transport is successfully created, but the connection fails
1285  	// before the SubConn can send the notification for READY. We treat
1286  	// this as a successful connection and transition to IDLE.
1287  	// TODO: https://github.com/grpc/grpc-go/issues/7862 - Remove the second
1288  	// part of the if condition below once the issue is fixed.
1289  	if ac.state == connectivity.Ready || (ac.state == connectivity.Connecting && s == connectivity.Idle) {
1290  		disconnectionsMetric.Record(ac.cc.metricsRecorderList, 1, ac.cc.target, ac.backendServiceLabel, ac.localityLabel, "unknown")
1291  		openConnectionsMetric.Record(ac.cc.metricsRecorderList, -1, ac.cc.target, ac.backendServiceLabel, ac.securityLevelLocked(), ac.localityLabel)
1292  	}
1293  	ac.state = s
1294  	ac.channelz.ChannelMetrics.State.Store(&s)
1295  	if lastErr == nil {
1296  		channelz.Infof(logger, ac.channelz, "Subchannel Connectivity change to %v", s)
1297  	} else {
1298  		channelz.Infof(logger, ac.channelz, "Subchannel Connectivity change to %v, last error: %s", s, lastErr)
1299  	}
1300  	ac.acbw.updateState(s, ac.curAddr, lastErr)
1301  }
1302  
1303  // adjustParams updates parameters used to create transports upon
1304  // receiving a GoAway.
1305  func (ac *addrConn) adjustParams(r transport.GoAwayReason) {
1306  	if r == transport.GoAwayTooManyPings {
1307  		v := 2 * ac.dopts.copts.KeepaliveParams.Time
1308  		ac.cc.mu.Lock()
1309  		if v > ac.cc.keepaliveParams.Time {
1310  			ac.cc.keepaliveParams.Time = v
1311  		}
1312  		ac.cc.mu.Unlock()
1313  	}
1314  }
1315  
1316  // resetTransportAndUnlock unconditionally connects the addrConn.
1317  //
1318  // ac.mu must be held by the caller, and this function will guarantee it is released.
1319  func (ac *addrConn) resetTransportAndUnlock() {
1320  	acCtx := ac.ctx
1321  	if acCtx.Err() != nil {
1322  		ac.mu.Unlock()
1323  		return
1324  	}
1325  
1326  	addrs := ac.addrs
1327  	backoffFor := ac.dopts.bs.Backoff(ac.backoffIdx)
1328  	// This will be the duration that dial gets to finish.
1329  	dialDuration := minConnectTimeout
1330  	if ac.dopts.minConnectTimeout != nil {
1331  		dialDuration = ac.dopts.minConnectTimeout()
1332  	}
1333  
1334  	if dialDuration < backoffFor {
1335  		// Give dial more time as we keep failing to connect.
1336  		dialDuration = backoffFor
1337  	}
1338  	// We can potentially spend all the time trying the first address, and
1339  	// if the server accepts the connection and then hangs, the following
1340  	// addresses will never be tried.
1341  	//
1342  	// The spec doesn't mention what should be done for multiple addresses.
1343  	// https://github.com/grpc/grpc/blob/master/doc/connection-backoff.md#proposed-backoff-algorithm
1344  	connectDeadline := time.Now().Add(dialDuration)
1345  
1346  	ac.updateConnectivityState(connectivity.Connecting, nil)
1347  	ac.mu.Unlock()
1348  
1349  	if err := ac.tryAllAddrs(acCtx, addrs, connectDeadline); err != nil {
1350  		if !errors.Is(err, context.Canceled) {
1351  			connectionAttemptsFailedMetric.Record(ac.cc.metricsRecorderList, 1, ac.cc.target, ac.backendServiceLabel, ac.localityLabel)
1352  		} else {
1353  			if logger.V(2) {
1354  				// This records cancelled connection attempts which can be later
1355  				// replaced by a metric.
1356  				logger.Infof("Context cancellation detected; not recording this as a failed connection attempt.")
1357  			}
1358  		}
1359  		// TODO: #7534 - Move re-resolution requests into the pick_first LB policy
1360  		// to ensure one resolution request per pass instead of per subconn failure.
1361  		ac.cc.resolveNow(resolver.ResolveNowOptions{})
1362  		ac.mu.Lock()
1363  		if acCtx.Err() != nil {
1364  			// addrConn was torn down.
1365  			ac.mu.Unlock()
1366  			return
1367  		}
1368  		// After exhausting all addresses, the addrConn enters
1369  		// TRANSIENT_FAILURE.
1370  		ac.updateConnectivityState(connectivity.TransientFailure, err)
1371  
1372  		// Backoff.
1373  		b := ac.resetBackoff
1374  		ac.mu.Unlock()
1375  
1376  		timer := time.NewTimer(backoffFor)
1377  		select {
1378  		case <-timer.C:
1379  			ac.mu.Lock()
1380  			ac.backoffIdx++
1381  			ac.mu.Unlock()
1382  		case <-b:
1383  			timer.Stop()
1384  		case <-acCtx.Done():
1385  			timer.Stop()
1386  			return
1387  		}
1388  
1389  		ac.mu.Lock()
1390  		if acCtx.Err() == nil {
1391  			ac.updateConnectivityState(connectivity.Idle, err)
1392  		}
1393  		ac.mu.Unlock()
1394  		return
1395  	}
1396  	// Success; reset backoff.
1397  	ac.mu.Lock()
1398  	connectionAttemptsSucceededMetric.Record(ac.cc.metricsRecorderList, 1, ac.cc.target, ac.backendServiceLabel, ac.localityLabel)
1399  	openConnectionsMetric.Record(ac.cc.metricsRecorderList, 1, ac.cc.target, ac.backendServiceLabel, ac.securityLevelLocked(), ac.localityLabel)
1400  	ac.backoffIdx = 0
1401  	ac.mu.Unlock()
1402  }
1403  
1404  // updateTelemetryLabelsLocked calculates and caches the telemetry labels based on the
1405  // first address in addrConn.
1406  func (ac *addrConn) updateTelemetryLabelsLocked() {
1407  	labelsFunc, ok := internal.AddressToTelemetryLabels.(func(resolver.Address) map[string]string)
1408  	if !ok || len(ac.addrs) == 0 {
1409  		// Reset defaults
1410  		ac.localityLabel = ""
1411  		ac.backendServiceLabel = ""
1412  		return
1413  	}
1414  	labels := labelsFunc(ac.addrs[0])
1415  	ac.localityLabel = labels["grpc.lb.locality"]
1416  	ac.backendServiceLabel = labels["grpc.lb.backend_service"]
1417  }
1418  
1419  type securityLevelKey struct{}
1420  
1421  func (ac *addrConn) securityLevelLocked() string {
1422  	var secLevel string
1423  	// During disconnection, ac.transport is nil. Fall back to the security level
1424  	// stored in the current address during connection.
1425  	if ac.transport == nil {
1426  		secLevel, _ = ac.curAddr.Attributes.Value(securityLevelKey{}).(string)
1427  		return secLevel
1428  	}
1429  	authInfo := ac.transport.Peer().AuthInfo
1430  	if ci, ok := authInfo.(interface {
1431  		GetCommonAuthInfo() credentials.CommonAuthInfo
1432  	}); ok {
1433  		secLevel = ci.GetCommonAuthInfo().SecurityLevel.String()
1434  		// Store the security level in the current address' attributes so
1435  		// that it remains available for disconnection metrics after the
1436  		// transport is closed.
1437  		ac.curAddr.Attributes = ac.curAddr.Attributes.WithValue(securityLevelKey{}, secLevel)
1438  	}
1439  	return secLevel
1440  }
1441  
1442  // tryAllAddrs tries to create a connection to the addresses, and stop when at
1443  // the first successful one. It returns an error if no address was successfully
1444  // connected, or updates ac appropriately with the new transport.
1445  func (ac *addrConn) tryAllAddrs(ctx context.Context, addrs []resolver.Address, connectDeadline time.Time) error {
1446  	var firstConnErr error
1447  	for _, addr := range addrs {
1448  		ac.channelz.ChannelMetrics.Target.Store(&addr.Addr)
1449  		if ctx.Err() != nil {
1450  			return errConnClosing
1451  		}
1452  		ac.mu.Lock()
1453  
1454  		ac.cc.mu.RLock()
1455  		ac.dopts.copts.KeepaliveParams = ac.cc.keepaliveParams
1456  		ac.cc.mu.RUnlock()
1457  
1458  		copts := ac.dopts.copts
1459  		if ac.scopts.CredsBundle != nil {
1460  			copts.CredsBundle = ac.scopts.CredsBundle
1461  		}
1462  		ac.mu.Unlock()
1463  
1464  		channelz.Infof(logger, ac.channelz, "Subchannel picks a new address %q to connect", addr.Addr)
1465  
1466  		err := ac.createTransport(ctx, addr, copts, connectDeadline)
1467  		if err == nil {
1468  			return nil
1469  		}
1470  		if firstConnErr == nil {
1471  			firstConnErr = err
1472  		}
1473  		ac.cc.updateConnectionError(err)
1474  	}
1475  
1476  	// Couldn't connect to any address.
1477  	return firstConnErr
1478  }
1479  
1480  // createTransport creates a connection to addr. It returns an error if the
1481  // address was not successfully connected, or updates ac appropriately with the
1482  // new transport.
1483  func (ac *addrConn) createTransport(ctx context.Context, addr resolver.Address, copts transport.ConnectOptions, connectDeadline time.Time) error {
1484  	addr.ServerName = ac.cc.getServerName(addr)
1485  	hctx, hcancel := context.WithCancel(ctx)
1486  
1487  	onClose := func(r transport.GoAwayReason) {
1488  		ac.mu.Lock()
1489  		defer ac.mu.Unlock()
1490  		// adjust params based on GoAwayReason
1491  		ac.adjustParams(r)
1492  		if ctx.Err() != nil {
1493  			// Already shut down or connection attempt canceled.  tearDown() or
1494  			// updateAddrs() already cleared the transport and canceled hctx
1495  			// via ac.ctx, and we expected this connection to be closed, so do
1496  			// nothing here.
1497  			return
1498  		}
1499  		hcancel()
1500  		if ac.transport == nil {
1501  			// We're still connecting to this address, which could error.  Do
1502  			// not update the connectivity state or resolve; these will happen
1503  			// at the end of the tryAllAddrs connection loop in the event of an
1504  			// error.
1505  			return
1506  		}
1507  		ac.transport = nil
1508  		// Refresh the name resolver on any connection loss.
1509  		ac.cc.resolveNow(resolver.ResolveNowOptions{})
1510  		// Always go idle and wait for the LB policy to initiate a new
1511  		// connection attempt.
1512  		ac.updateConnectivityState(connectivity.Idle, nil)
1513  	}
1514  
1515  	connectCtx, cancel := context.WithDeadline(ctx, connectDeadline)
1516  	defer cancel()
1517  	copts.ChannelzParent = ac.channelz
1518  
1519  	newTr, err := transport.NewHTTP2Client(connectCtx, ac.cc.ctx, addr, copts, onClose)
1520  	if err != nil {
1521  		if logger.V(2) {
1522  			logger.Infof("Creating new client transport to %q: %v", addr, err)
1523  		}
1524  		// newTr is either nil, or closed.
1525  		hcancel()
1526  		channelz.Warningf(logger, ac.channelz, "grpc: addrConn.createTransport failed to connect to %s. Err: %v", addr, err)
1527  		return err
1528  	}
1529  
1530  	ac.mu.Lock()
1531  	defer ac.mu.Unlock()
1532  	if ctx.Err() != nil {
1533  		// This can happen if the subConn was removed while in `Connecting`
1534  		// state. tearDown() would have set the state to `Shutdown`, but
1535  		// would not have closed the transport since ac.transport would not
1536  		// have been set at that point.
1537  		//
1538  		// We run this in a goroutine because newTr.Close() calls onClose()
1539  		// inline, which requires locking ac.mu.
1540  		//
1541  		// The error we pass to Close() is immaterial since there are no open
1542  		// streams at this point, so no trailers with error details will be sent
1543  		// out. We just need to pass a non-nil error.
1544  		//
1545  		// This can also happen when updateAddrs is called during a connection
1546  		// attempt.
1547  		go newTr.Close(transport.ErrConnClosing)
1548  		return nil
1549  	}
1550  	if hctx.Err() != nil {
1551  		// onClose was already called for this connection, but the connection
1552  		// was successfully established first.  Consider it a success and set
1553  		// the new state to Idle.
1554  		ac.updateConnectivityState(connectivity.Idle, nil)
1555  		return nil
1556  	}
1557  	ac.curAddr = addr
1558  	ac.transport = newTr
1559  	ac.startHealthCheck(hctx) // Will set state to READY if appropriate.
1560  	return nil
1561  }
1562  
1563  // startHealthCheck starts the health checking stream (RPC) to watch the health
1564  // stats of this connection if health checking is requested and configured.
1565  //
1566  // LB channel health checking is enabled when all requirements below are met:
1567  // 1. it is not disabled by the user with the WithDisableHealthCheck DialOption
1568  // 2. internal.HealthCheckFunc is set by importing the grpc/health package
1569  // 3. a service config with non-empty healthCheckConfig field is provided
1570  // 4. the load balancer requests it
1571  //
1572  // It sets addrConn to READY if the health checking stream is not started.
1573  //
1574  // Caller must hold ac.mu.
1575  func (ac *addrConn) startHealthCheck(ctx context.Context) {
1576  	var healthcheckManagingState bool
1577  	defer func() {
1578  		if !healthcheckManagingState {
1579  			ac.updateConnectivityState(connectivity.Ready, nil)
1580  		}
1581  	}()
1582  
1583  	if ac.cc.dopts.disableHealthCheck {
1584  		return
1585  	}
1586  	healthCheckConfig := ac.cc.healthCheckConfig()
1587  	if healthCheckConfig == nil {
1588  		return
1589  	}
1590  	if !ac.scopts.HealthCheckEnabled {
1591  		return
1592  	}
1593  	healthCheckFunc := internal.HealthCheckFunc
1594  	if healthCheckFunc == nil {
1595  		// The health package is not imported to set health check function.
1596  		//
1597  		// TODO: add a link to the health check doc in the error message.
1598  		channelz.Error(logger, ac.channelz, "Health check is requested but health check function is not set.")
1599  		return
1600  	}
1601  
1602  	healthcheckManagingState = true
1603  
1604  	// Set up the health check helper functions.
1605  	currentTr := ac.transport
1606  	newStream := func(method string) (any, error) {
1607  		ac.mu.Lock()
1608  		if ac.transport != currentTr {
1609  			ac.mu.Unlock()
1610  			return nil, status.Error(codes.Canceled, "the provided transport is no longer valid to use")
1611  		}
1612  		ac.mu.Unlock()
1613  		return newNonRetryClientStream(ctx, &StreamDesc{ServerStreams: true}, method, currentTr, ac)
1614  	}
1615  	setConnectivityState := func(s connectivity.State, lastErr error) {
1616  		ac.mu.Lock()
1617  		defer ac.mu.Unlock()
1618  		if ac.transport != currentTr {
1619  			return
1620  		}
1621  		ac.updateConnectivityState(s, lastErr)
1622  	}
1623  	// Start the health checking stream.
1624  	go func() {
1625  		err := healthCheckFunc(ctx, newStream, setConnectivityState, healthCheckConfig.ServiceName)
1626  		if err != nil {
1627  			if status.Code(err) == codes.Unimplemented {
1628  				channelz.Error(logger, ac.channelz, "Subchannel health check is unimplemented at server side, thus health check is disabled")
1629  			} else {
1630  				channelz.Errorf(logger, ac.channelz, "Health checking failed: %v", err)
1631  			}
1632  		}
1633  	}()
1634  }
1635  
1636  func (ac *addrConn) resetConnectBackoff() {
1637  	ac.mu.Lock()
1638  	close(ac.resetBackoff)
1639  	ac.backoffIdx = 0
1640  	ac.resetBackoff = make(chan struct{})
1641  	ac.mu.Unlock()
1642  }
1643  
1644  // getReadyTransport returns the transport if ac's state is READY or nil if not.
1645  func (ac *addrConn) getReadyTransport() transport.ClientTransport {
1646  	ac.mu.Lock()
1647  	defer ac.mu.Unlock()
1648  	if ac.state == connectivity.Ready {
1649  		return ac.transport
1650  	}
1651  	return nil
1652  }
1653  
1654  // tearDown starts to tear down the addrConn.
1655  //
1656  // Note that tearDown doesn't remove ac from ac.cc.conns, so the addrConn struct
1657  // will leak. In most cases, call cc.removeAddrConn() instead.
1658  func (ac *addrConn) tearDown(err error) {
1659  	ac.mu.Lock()
1660  	if ac.state == connectivity.Shutdown {
1661  		ac.mu.Unlock()
1662  		return
1663  	}
1664  	curTr := ac.transport
1665  	ac.transport = nil
1666  	// We have to set the state to Shutdown before anything else to prevent races
1667  	// between setting the state and logic that waits on context cancellation / etc.
1668  	ac.updateConnectivityState(connectivity.Shutdown, nil)
1669  	ac.cancel()
1670  	ac.curAddr = resolver.Address{}
1671  
1672  	channelz.AddTraceEvent(logger, ac.channelz, 0, &channelz.TraceEvent{
1673  		Desc:     "Subchannel deleted",
1674  		Severity: channelz.CtInfo,
1675  		Parent: &channelz.TraceEvent{
1676  			Desc:     fmt.Sprintf("Subchannel(id:%d) deleted", ac.channelz.ID),
1677  			Severity: channelz.CtInfo,
1678  		},
1679  	})
1680  	// TraceEvent needs to be called before RemoveEntry, as TraceEvent may add
1681  	// trace reference to the entity being deleted, and thus prevent it from
1682  	// being deleted right away.
1683  	channelz.RemoveEntry(ac.channelz.ID)
1684  	ac.mu.Unlock()
1685  
1686  	// We have to release the lock before the call to GracefulClose/Close here
1687  	// because both of them call onClose(), which requires locking ac.mu.
1688  	if curTr != nil {
1689  		if err == errConnDrain {
1690  			// Close the transport gracefully when the subConn is being shutdown.
1691  			//
1692  			// GracefulClose() may be executed multiple times if:
1693  			// - multiple GoAway frames are received from the server
1694  			// - there are concurrent name resolver or balancer triggered
1695  			//   address removal and GoAway
1696  			curTr.GracefulClose()
1697  		} else {
1698  			// Hard close the transport when the channel is entering idle or is
1699  			// being shutdown. In the case where the channel is being shutdown,
1700  			// closing of transports is also taken care of by cancellation of cc.ctx.
1701  			// But in the case where the channel is entering idle, we need to
1702  			// explicitly close the transports here. Instead of distinguishing
1703  			// between these two cases, it is simpler to close the transport
1704  			// unconditionally here.
1705  			curTr.Close(err)
1706  		}
1707  	}
1708  }
1709  
1710  type retryThrottler struct {
1711  	max    float64
1712  	thresh float64
1713  	ratio  float64
1714  
1715  	mu     sync.Mutex
1716  	tokens float64 // TODO(dfawley): replace with atomic and remove lock.
1717  }
1718  
1719  // throttle subtracts a retry token from the pool and returns whether a retry
1720  // should be throttled (disallowed) based upon the retry throttling policy in
1721  // the service config.
1722  func (rt *retryThrottler) throttle() bool {
1723  	if rt == nil {
1724  		return false
1725  	}
1726  	rt.mu.Lock()
1727  	defer rt.mu.Unlock()
1728  	rt.tokens--
1729  	if rt.tokens < 0 {
1730  		rt.tokens = 0
1731  	}
1732  	return rt.tokens <= rt.thresh
1733  }
1734  
1735  func (rt *retryThrottler) successfulRPC() {
1736  	if rt == nil {
1737  		return
1738  	}
1739  	rt.mu.Lock()
1740  	defer rt.mu.Unlock()
1741  	rt.tokens += rt.ratio
1742  	if rt.tokens > rt.max {
1743  		rt.tokens = rt.max
1744  	}
1745  }
1746  
1747  func (ac *addrConn) incrCallsStarted() {
1748  	ac.channelz.ChannelMetrics.CallsStarted.Add(1)
1749  	ac.channelz.ChannelMetrics.LastCallStartedTimestamp.Store(time.Now().UnixNano())
1750  }
1751  
1752  func (ac *addrConn) incrCallsSucceeded() {
1753  	ac.channelz.ChannelMetrics.CallsSucceeded.Add(1)
1754  }
1755  
1756  func (ac *addrConn) incrCallsFailed() {
1757  	ac.channelz.ChannelMetrics.CallsFailed.Add(1)
1758  }
1759  
1760  // ErrClientConnTimeout indicates that the ClientConn cannot establish the
1761  // underlying connections within the specified timeout.
1762  //
1763  // Deprecated: This error is never returned by grpc and should not be
1764  // referenced by users.
1765  var ErrClientConnTimeout = errors.New("grpc: timed out when dialing")
1766  
1767  // getResolver finds the scheme in the cc's resolvers or the global registry.
1768  // scheme should always be lowercase (typically by virtue of url.Parse()
1769  // performing proper RFC3986 behavior).
1770  func (cc *ClientConn) getResolver(scheme string) resolver.Builder {
1771  	for _, rb := range cc.dopts.resolvers {
1772  		if scheme == rb.Scheme() {
1773  			return rb
1774  		}
1775  	}
1776  	return resolver.Get(scheme)
1777  }
1778  
1779  func (cc *ClientConn) updateConnectionError(err error) {
1780  	cc.lceMu.Lock()
1781  	cc.lastConnectionError = err
1782  	cc.lceMu.Unlock()
1783  }
1784  
1785  func (cc *ClientConn) connectionError() error {
1786  	cc.lceMu.Lock()
1787  	defer cc.lceMu.Unlock()
1788  	return cc.lastConnectionError
1789  }
1790  
1791  // initParsedTargetAndResolverBuilder parses the user's dial target and stores
1792  // the parsed target in `cc.parsedTarget`.
1793  //
1794  // The resolver to use is determined based on the scheme in the parsed target
1795  // and the same is stored in `cc.resolverBuilder`.
1796  //
1797  // Doesn't grab cc.mu as this method is expected to be called only at Dial time.
1798  func (cc *ClientConn) initParsedTargetAndResolverBuilder() error {
1799  	logger.Infof("original dial target is: %q", cc.target)
1800  
1801  	var rb resolver.Builder
1802  	parsedTarget, err := parseTarget(cc.target)
1803  	if err == nil {
1804  		rb = cc.getResolver(parsedTarget.URL.Scheme)
1805  		if rb != nil {
1806  			cc.parsedTarget = parsedTarget
1807  			cc.resolverBuilder = rb
1808  			return nil
1809  		}
1810  	}
1811  
1812  	// We are here because the user's dial target did not contain a scheme or
1813  	// specified an unregistered scheme. We should fallback to the default
1814  	// scheme, except when a custom dialer is specified in which case, we should
1815  	// always use passthrough scheme. For either case, we need to respect any overridden
1816  	// global defaults set by the user.
1817  	defScheme := cc.dopts.defaultScheme
1818  	if internal.UserSetDefaultScheme {
1819  		defScheme = resolver.GetDefaultScheme()
1820  	}
1821  
1822  	canonicalTarget := defScheme + ":///" + cc.target
1823  
1824  	parsedTarget, err = parseTarget(canonicalTarget)
1825  	if err != nil {
1826  		return err
1827  	}
1828  	rb = cc.getResolver(parsedTarget.URL.Scheme)
1829  	if rb == nil {
1830  		return fmt.Errorf("could not get resolver for default scheme: %q", parsedTarget.URL.Scheme)
1831  	}
1832  	cc.parsedTarget = parsedTarget
1833  	cc.resolverBuilder = rb
1834  	return nil
1835  }
1836  
1837  // parseTarget uses RFC 3986 semantics to parse the given target into a
1838  // resolver.Target struct containing url. Query params are stripped from the
1839  // endpoint.
1840  func parseTarget(target string) (resolver.Target, error) {
1841  	u, err := url.Parse(target)
1842  	if err != nil {
1843  		return resolver.Target{}, err
1844  	}
1845  
1846  	return resolver.Target{URL: *u}, nil
1847  }
1848  
1849  // encodeAuthority escapes the authority string based on valid chars defined in
1850  // https://datatracker.ietf.org/doc/html/rfc3986#section-3.2.
1851  func encodeAuthority(authority string) string {
1852  	const upperhex = "0123456789ABCDEF"
1853  
1854  	// Return for characters that must be escaped as per
1855  	// Valid chars are mentioned here:
1856  	// https://datatracker.ietf.org/doc/html/rfc3986#section-3.2
1857  	shouldEscape := func(c byte) bool {
1858  		// Alphanum are always allowed.
1859  		if 'a' <= c && c <= 'z' || 'A' <= c && c <= 'Z' || '0' <= c && c <= '9' {
1860  			return false
1861  		}
1862  		switch c {
1863  		case '-', '_', '.', '~': // Unreserved characters
1864  			return false
1865  		case '!', '$', '&', '\'', '(', ')', '*', '+', ',', ';', '=': // Subdelim characters
1866  			return false
1867  		case ':', '[', ']', '@': // Authority related delimiters
1868  			return false
1869  		}
1870  		// Everything else must be escaped.
1871  		return true
1872  	}
1873  
1874  	hexCount := 0
1875  	for i := 0; i < len(authority); i++ {
1876  		c := authority[i]
1877  		if shouldEscape(c) {
1878  			hexCount++
1879  		}
1880  	}
1881  
1882  	if hexCount == 0 {
1883  		return authority
1884  	}
1885  
1886  	required := len(authority) + 2*hexCount
1887  	t := make([]byte, required)
1888  
1889  	j := 0
1890  	// This logic is a barebones version of escape in the go net/url library.
1891  	for i := 0; i < len(authority); i++ {
1892  		switch c := authority[i]; {
1893  		case shouldEscape(c):
1894  			t[j] = '%'
1895  			t[j+1] = upperhex[c>>4]
1896  			t[j+2] = upperhex[c&15]
1897  			j += 3
1898  		default:
1899  			t[j] = authority[i]
1900  			j++
1901  		}
1902  	}
1903  	return string(t)
1904  }
1905  
1906  // Determine channel authority. The order of precedence is as follows:
1907  // - user specified authority override using `WithAuthority` dial option
1908  // - creds' notion of server name for the authentication handshake
1909  // - endpoint from dial target of the form "scheme://[authority]/endpoint"
1910  //
1911  // Stores the determined authority in `cc.authority`.
1912  //
1913  // Returns a non-nil error if the authority returned by the transport
1914  // credentials do not match the authority configured through the dial option.
1915  //
1916  // Doesn't grab cc.mu as this method is expected to be called only at Dial time.
1917  func (cc *ClientConn) initAuthority() error {
1918  	dopts := cc.dopts
1919  	// Historically, we had two options for users to specify the serverName or
1920  	// authority for a channel. One was through the transport credentials
1921  	// (either in its constructor, or through the OverrideServerName() method).
1922  	// The other option (for cases where WithInsecure() dial option was used)
1923  	// was to use the WithAuthority() dial option.
1924  	//
1925  	// A few things have changed since:
1926  	// - `insecure` package with an implementation of the `TransportCredentials`
1927  	//   interface for the insecure case
1928  	// - WithAuthority() dial option support for secure credentials
1929  	authorityFromCreds := ""
1930  	if creds := dopts.copts.TransportCredentials; creds != nil && creds.Info().ServerName != "" {
1931  		authorityFromCreds = creds.Info().ServerName
1932  	}
1933  	authorityFromDialOption := dopts.authority
1934  	if (authorityFromCreds != "" && authorityFromDialOption != "") && authorityFromCreds != authorityFromDialOption {
1935  		return fmt.Errorf("ClientConn's authority from transport creds %q and dial option %q don't match", authorityFromCreds, authorityFromDialOption)
1936  	}
1937  
1938  	endpoint := cc.parsedTarget.Endpoint()
1939  	if authorityFromDialOption != "" {
1940  		cc.authority = authorityFromDialOption
1941  	} else if authorityFromCreds != "" {
1942  		cc.authority = authorityFromCreds
1943  	} else if auth, ok := cc.resolverBuilder.(resolver.AuthorityOverrider); ok {
1944  		cc.authority = auth.OverrideAuthority(cc.parsedTarget)
1945  	} else if strings.HasPrefix(endpoint, ":") {
1946  		cc.authority = "localhost" + encodeAuthority(endpoint)
1947  	} else {
1948  		cc.authority = encodeAuthority(endpoint)
1949  	}
1950  	return nil
1951  }
1952