Update go dependencies

vendor/google.golang.org/grpc/internal/channelz/funcs.go

@@ -27,16 +27,22 @@ import (
 	"sort"
 	"sync"
 	"sync/atomic"
+	"time"
 
 	"google.golang.org/grpc/grpclog"
 )
 
+const (
+	defaultMaxTraceEntry int32 = 30
+)
+
 var (
 	db    dbWrapper
 	idGen idGenerator
 	// EntryPerPage defines the number of channelz entries to be shown on a web page.
-	EntryPerPage = 50
-	curState     int32
+	EntryPerPage  = 50
+	curState      int32
+	maxTraceEntry = defaultMaxTraceEntry
 )
 
 // TurnOn turns on channelz data collection.
@@ -52,6 +58,22 @@ func IsOn() bool {
 	return atomic.CompareAndSwapInt32(&curState, 1, 1)
 }
 
+// SetMaxTraceEntry sets maximum number of trace entry per entity (i.e. channel/subchannel).
+// Setting it to 0 will disable channel tracing.
+func SetMaxTraceEntry(i int32) {
+	atomic.StoreInt32(&maxTraceEntry, i)
+}
+
+// ResetMaxTraceEntryToDefault resets the maximum number of trace entry per entity to default.
+func ResetMaxTraceEntryToDefault() {
+	atomic.StoreInt32(&maxTraceEntry, defaultMaxTraceEntry)
+}
+
+func getMaxTraceEntry() int {
+	i := atomic.LoadInt32(&maxTraceEntry)
+	return int(i)
+}
+
 // dbWarpper wraps around a reference to internal channelz data storage, and
 // provide synchronized functionality to set and get the reference.
 type dbWrapper struct {
@@ -146,6 +168,7 @@ func RegisterChannel(c Channel, pid int64, ref string) int64 {
 		nestedChans: make(map[int64]string),
 		id:          id,
 		pid:         pid,
+		trace:       &channelTrace{createdTime: time.Now(), events: make([]*TraceEvent, 0, getMaxTraceEntry())},
 	}
 	if pid == 0 {
 		db.get().addChannel(id, cn, true, pid, ref)
@@ -170,6 +193,7 @@ func RegisterSubChannel(c Channel, pid int64, ref string) int64 {
 		sockets: make(map[int64]string),
 		id:      id,
 		pid:     pid,
+		trace:   &channelTrace{createdTime: time.Now(), events: make([]*TraceEvent, 0, getMaxTraceEntry())},
 	}
 	db.get().addSubChannel(id, sc, pid, ref)
 	return id
@@ -226,6 +250,24 @@ func RemoveEntry(id int64) {
 	db.get().removeEntry(id)
 }
 
+// TraceEventDesc is what the caller of AddTraceEvent should provide to describe the event to be added
+// to the channel trace.
+// The Parent field is optional. It is used for event that will be recorded in the entity's parent
+// trace also.
+type TraceEventDesc struct {
+	Desc     string
+	Severity Severity
+	Parent   *TraceEventDesc
+}
+
+// AddTraceEvent adds trace related to the entity with specified id, using the provided TraceEventDesc.
+func AddTraceEvent(id int64, desc *TraceEventDesc) {
+	if getMaxTraceEntry() == 0 {
+		return
+	}
+	db.get().traceEvent(id, desc)
+}
+
 // channelMap is the storage data structure for channelz.
 // Methods of channelMap can be divided in two two categories with respect to locking.
 // 1. Methods acquire the global lock.
@@ -251,6 +293,7 @@ func (c *channelMap) addServer(id int64, s *server) {
 func (c *channelMap) addChannel(id int64, cn *channel, isTopChannel bool, pid int64, ref string) {
 	c.mu.Lock()
 	cn.cm = c
+	cn.trace.cm = c
 	c.channels[id] = cn
 	if isTopChannel {
 		c.topLevelChannels[id] = struct{}{}
@@ -263,6 +306,7 @@ func (c *channelMap) addChannel(id int64, cn *channel, isTopChannel bool, pid in
 func (c *channelMap) addSubChannel(id int64, sc *subChannel, pid int64, ref string) {
 	c.mu.Lock()
 	sc.cm = c
+	sc.trace.cm = c
 	c.subChannels[id] = sc
 	c.findEntry(pid).addChild(id, sc)
 	c.mu.Unlock()
@@ -284,16 +328,25 @@ func (c *channelMap) addNormalSocket(id int64, ns *normalSocket, pid int64, ref
 	c.mu.Unlock()
 }
 
-// removeEntry triggers the removal of an entry, which may not indeed delete the
-// entry, if it has to wait on the deletion of its children, or may lead to a chain
-// of entry deletion. For example, deleting the last socket of a gracefully shutting
-// down server will lead to the server being also deleted.
+// removeEntry triggers the removal of an entry, which may not indeed delete the entry, if it has to
+// wait on the deletion of its children and until no other entity's channel trace references it.
+// It may lead to a chain of entry deletion. For example, deleting the last socket of a gracefully
+// shutting down server will lead to the server being also deleted.
 func (c *channelMap) removeEntry(id int64) {
 	c.mu.Lock()
 	c.findEntry(id).triggerDelete()
 	c.mu.Unlock()
 }
 
+// c.mu must be held by the caller
+func (c *channelMap) decrTraceRefCount(id int64) {
+	e := c.findEntry(id)
+	if v, ok := e.(tracedChannel); ok {
+		v.decrTraceRefCount()
+		e.deleteSelfIfReady()
+	}
+}
+
 // c.mu must be held by the caller.
 func (c *channelMap) findEntry(id int64) entry {
 	var v entry
@@ -347,6 +400,39 @@ func (c *channelMap) deleteEntry(id int64) {
 	}
 }
 
+func (c *channelMap) traceEvent(id int64, desc *TraceEventDesc) {
+	c.mu.Lock()
+	child := c.findEntry(id)
+	childTC, ok := child.(tracedChannel)
+	if !ok {
+		c.mu.Unlock()
+		return
+	}
+	childTC.getChannelTrace().append(&TraceEvent{Desc: desc.Desc, Severity: desc.Severity, Timestamp: time.Now()})
+	if desc.Parent != nil {
+		parent := c.findEntry(child.getParentID())
+		var chanType RefChannelType
+		switch child.(type) {
+		case *channel:
+			chanType = RefChannel
+		case *subChannel:
+			chanType = RefSubChannel
+		}
+		if parentTC, ok := parent.(tracedChannel); ok {
+			parentTC.getChannelTrace().append(&TraceEvent{
+				Desc:      desc.Parent.Desc,
+				Severity:  desc.Parent.Severity,
+				Timestamp: time.Now(),
+				RefID:     id,
+				RefName:   childTC.getRefName(),
+				RefType:   chanType,
+			})
+			childTC.incrTraceRefCount()
+		}
+	}
+	c.mu.Unlock()
+}
+
 type int64Slice []int64
 
 func (s int64Slice) Len() int           { return len(s) }
@@ -408,6 +494,7 @@ func (c *channelMap) GetTopChannels(id int64) ([]*ChannelMetric, bool) {
 		t[i].ChannelData = cn.c.ChannelzMetric()
 		t[i].ID = cn.id
 		t[i].RefName = cn.refName
+		t[i].Trace = cn.trace.dumpData()
 	}
 	return t, end
 }
@@ -470,7 +557,7 @@ func (c *channelMap) GetServerSockets(id int64, startID int64) ([]*SocketMetric,
 	for k := range svrskts {
 		ids = append(ids, k)
 	}
-	sort.Sort((int64Slice(ids)))
+	sort.Sort(int64Slice(ids))
 	idx := sort.Search(len(ids), func(i int) bool { return ids[i] >= id })
 	count := 0
 	var end bool
@@ -518,6 +605,7 @@ func (c *channelMap) GetChannel(id int64) *ChannelMetric {
 	cm.ChannelData = cn.c.ChannelzMetric()
 	cm.ID = cn.id
 	cm.RefName = cn.refName
+	cm.Trace = cn.trace.dumpData()
 	return cm
 }
 
@@ -536,6 +624,7 @@ func (c *channelMap) GetSubChannel(id int64) *SubChannelMetric {
 	cm.ChannelData = sc.c.ChannelzMetric()
 	cm.ID = sc.id
 	cm.RefName = sc.refName
+	cm.Trace = sc.trace.dumpData()
 	return cm
 }
 

vendor/google.golang.org/grpc/internal/channelz/types.go

@@ -20,6 +20,8 @@ package channelz
 
 import (
 	"net"
+	"sync"
+	"sync/atomic"
 	"time"
 
 	"google.golang.org/grpc/connectivity"
@@ -40,6 +42,8 @@ type entry interface {
 	// deleteSelfIfReady check whether triggerDelete() has been called before, and whether child
 	// list is now empty. If both conditions are met, then delete self from database.
 	deleteSelfIfReady()
+	// getParentID returns parent ID of the entry. 0 value parent ID means no parent.
+	getParentID() int64
 }
 
 // dummyEntry is a fake entry to handle entry not found case.
@@ -73,6 +77,10 @@ func (*dummyEntry) deleteSelfIfReady() {
 	// code should not reach here. deleteSelfIfReady is always called on an existing entry.
 }
 
+func (*dummyEntry) getParentID() int64 {
+	return 0
+}
+
 // ChannelMetric defines the info channelz provides for a specific Channel, which
 // includes ChannelInternalMetric and channelz-specific data, such as channelz id,
 // child list, etc.
@@ -95,6 +103,8 @@ type ChannelMetric struct {
 	// Note current grpc implementation doesn't allow channel having sockets directly,
 	// therefore, this is field is unused.
 	Sockets map[int64]string
+	// Trace contains the most recent traced events.
+	Trace *ChannelTrace
 }
 
 // SubChannelMetric defines the info channelz provides for a specific SubChannel,
@@ -121,6 +131,8 @@ type SubChannelMetric struct {
 	// Sockets tracks the socket type children of this subchannel in the format of a map
 	// from socket channelz id to corresponding reference string.
 	Sockets map[int64]string
+	// Trace contains the most recent traced events.
+	Trace *ChannelTrace
 }
 
 // ChannelInternalMetric defines the struct that the implementor of Channel interface
@@ -138,7 +150,35 @@ type ChannelInternalMetric struct {
 	CallsFailed int64
 	// The last time a call was started on the channel.
 	LastCallStartedTimestamp time.Time
-	//TODO: trace
+}
+
+// ChannelTrace stores traced events on a channel/subchannel and related info.
+type ChannelTrace struct {
+	// EventNum is the number of events that ever got traced (i.e. including those that have been deleted)
+	EventNum int64
+	// CreationTime is the creation time of the trace.
+	CreationTime time.Time
+	// Events stores the most recent trace events (up to $maxTraceEntry, newer event will overwrite the
+	// oldest one)
+	Events []*TraceEvent
+}
+
+// TraceEvent represent a single trace event
+type TraceEvent struct {
+	// Desc is a simple description of the trace event.
+	Desc string
+	// Severity states the severity of this trace event.
+	Severity Severity
+	// Timestamp is the event time.
+	Timestamp time.Time
+	// RefID is the id of the entity that gets referenced in the event. RefID is 0 if no other entity is
+	// involved in this event.
+	// e.g. SubChannel (id: 4[]) Created. --> RefID = 4, RefName = "" (inside [])
+	RefID int64
+	// RefName is the reference name for the entity that gets referenced in the event.
+	RefName string
+	// RefType indicates the referenced entity type, i.e Channel or SubChannel.
+	RefType RefChannelType
 }
 
 // Channel is the interface that should be satisfied in order to be tracked by
@@ -147,6 +187,12 @@ type Channel interface {
 	ChannelzMetric() *ChannelInternalMetric
 }
 
+type dummyChannel struct{}
+
+func (d *dummyChannel) ChannelzMetric() *ChannelInternalMetric {
+	return &ChannelInternalMetric{}
+}
+
 type channel struct {
 	refName     string
 	c           Channel
@@ -156,6 +202,10 @@ type channel struct {
 	id          int64
 	pid         int64
 	cm          *channelMap
+	trace       *channelTrace
+	// traceRefCount is the number of trace events that reference this channel.
+	// Non-zero traceRefCount means the trace of this channel cannot be deleted.
+	traceRefCount int32
 }
 
 func (c *channel) addChild(id int64, e entry) {
@@ -180,25 +230,96 @@ func (c *channel) triggerDelete() {
 	c.deleteSelfIfReady()
 }
 
-func (c *channel) deleteSelfIfReady() {
+func (c *channel) getParentID() int64 {
+	return c.pid
+}
+
+// deleteSelfFromTree tries to delete the channel from the channelz entry relation tree, which means
+// deleting the channel reference from its parent's child list.
+//
+// In order for a channel to be deleted from the tree, it must meet the criteria that, removal of the
+// corresponding grpc object has been invoked, and the channel does not have any children left.
+//
+// The returned boolean value indicates whether the channel has been successfully deleted from tree.
+func (c *channel) deleteSelfFromTree() (deleted bool) {
 	if !c.closeCalled || len(c.subChans)+len(c.nestedChans) != 0 {
-		return
+		return false
 	}
-	c.cm.deleteEntry(c.id)
 	// not top channel
 	if c.pid != 0 {
 		c.cm.findEntry(c.pid).deleteChild(c.id)
 	}
+	return true
+}
+
+// deleteSelfFromMap checks whether it is valid to delete the channel from the map, which means
+// deleting the channel from channelz's tracking entirely. Users can no longer use id to query the
+// channel, and its memory will be garbage collected.
+//
+// The trace reference count of the channel must be 0 in order to be deleted from the map. This is
+// specified in the channel tracing gRFC that as long as some other trace has reference to an entity,
+// the trace of the referenced entity must not be deleted. In order to release the resource allocated
+// by grpc, the reference to the grpc object is reset to a dummy object.
+//
+// deleteSelfFromMap must be called after deleteSelfFromTree returns true.
+//
+// It returns a bool to indicate whether the channel can be safely deleted from map.
+func (c *channel) deleteSelfFromMap() (delete bool) {
+	if c.getTraceRefCount() != 0 {
+		c.c = &dummyChannel{}
+		return false
+	}
+	return true
+}
+
+// deleteSelfIfReady tries to delete the channel itself from the channelz database.
+// The delete process includes two steps:
+// 1. delete the channel from the entry relation tree, i.e. delete the channel reference from its
+//    parent's child list.
+// 2. delete the channel from the map, i.e. delete the channel entirely from channelz. Lookup by id
+//    will return entry not found error.
+func (c *channel) deleteSelfIfReady() {
+	if !c.deleteSelfFromTree() {
+		return
+	}
+	if !c.deleteSelfFromMap() {
+		return
+	}
+	c.cm.deleteEntry(c.id)
+	c.trace.clear()
+}
+
+func (c *channel) getChannelTrace() *channelTrace {
+	return c.trace
+}
+
+func (c *channel) incrTraceRefCount() {
+	atomic.AddInt32(&c.traceRefCount, 1)
+}
+
+func (c *channel) decrTraceRefCount() {
+	atomic.AddInt32(&c.traceRefCount, -1)
+}
+
+func (c *channel) getTraceRefCount() int {
+	i := atomic.LoadInt32(&c.traceRefCount)
+	return int(i)
+}
+
+func (c *channel) getRefName() string {
+	return c.refName
 }
 
 type subChannel struct {
-	refName     string
-	c           Channel
-	closeCalled bool
-	sockets     map[int64]string
-	id          int64
-	pid         int64
-	cm          *channelMap
+	refName       string
+	c             Channel
+	closeCalled   bool
+	sockets       map[int64]string
+	id            int64
+	pid           int64
+	cm            *channelMap
+	trace         *channelTrace
+	traceRefCount int32
 }
 
 func (sc *subChannel) addChild(id int64, e entry) {
@@ -219,12 +340,82 @@ func (sc *subChannel) triggerDelete() {
 	sc.deleteSelfIfReady()
 }
 
-func (sc *subChannel) deleteSelfIfReady() {
+func (sc *subChannel) getParentID() int64 {
+	return sc.pid
+}
+
+// deleteSelfFromTree tries to delete the subchannel from the channelz entry relation tree, which
+// means deleting the subchannel reference from its parent's child list.
+//
+// In order for a subchannel to be deleted from the tree, it must meet the criteria that, removal of
+// the corresponding grpc object has been invoked, and the subchannel does not have any children left.
+//
+// The returned boolean value indicates whether the channel has been successfully deleted from tree.
+func (sc *subChannel) deleteSelfFromTree() (deleted bool) {
 	if !sc.closeCalled || len(sc.sockets) != 0 {
+		return false
+	}
+	sc.cm.findEntry(sc.pid).deleteChild(sc.id)
+	return true
+}
+
+// deleteSelfFromMap checks whether it is valid to delete the subchannel from the map, which means
+// deleting the subchannel from channelz's tracking entirely. Users can no longer use id to query
+// the subchannel, and its memory will be garbage collected.
+//
+// The trace reference count of the subchannel must be 0 in order to be deleted from the map. This is
+// specified in the channel tracing gRFC that as long as some other trace has reference to an entity,
+// the trace of the referenced entity must not be deleted. In order to release the resource allocated
+// by grpc, the reference to the grpc object is reset to a dummy object.
+//
+// deleteSelfFromMap must be called after deleteSelfFromTree returns true.
+//
+// It returns a bool to indicate whether the channel can be safely deleted from map.
+func (sc *subChannel) deleteSelfFromMap() (delete bool) {
+	if sc.getTraceRefCount() != 0 {
+		// free the grpc struct (i.e. addrConn)
+		sc.c = &dummyChannel{}
+		return false
+	}
+	return true
+}
+
+// deleteSelfIfReady tries to delete the subchannel itself from the channelz database.
+// The delete process includes two steps:
+// 1. delete the subchannel from the entry relation tree, i.e. delete the subchannel reference from
+//    its parent's child list.
+// 2. delete the subchannel from the map, i.e. delete the subchannel entirely from channelz. Lookup
+//    by id will return entry not found error.
+func (sc *subChannel) deleteSelfIfReady() {
+	if !sc.deleteSelfFromTree() {
+		return
+	}
+	if !sc.deleteSelfFromMap() {
 		return
 	}
 	sc.cm.deleteEntry(sc.id)
-	sc.cm.findEntry(sc.pid).deleteChild(sc.id)
+	sc.trace.clear()
+}
+
+func (sc *subChannel) getChannelTrace() *channelTrace {
+	return sc.trace
+}
+
+func (sc *subChannel) incrTraceRefCount() {
+	atomic.AddInt32(&sc.traceRefCount, 1)
+}
+
+func (sc *subChannel) decrTraceRefCount() {
+	atomic.AddInt32(&sc.traceRefCount, -1)
+}
+
+func (sc *subChannel) getTraceRefCount() int {
+	i := atomic.LoadInt32(&sc.traceRefCount)
+	return int(i)
+}
+
+func (sc *subChannel) getRefName() string {
+	return sc.refName
 }
 
 // SocketMetric defines the info channelz provides for a specific Socket, which
@@ -318,6 +509,10 @@ func (ls *listenSocket) deleteSelfIfReady() {
 	grpclog.Errorf("cannot call deleteSelfIfReady on a listen socket")
 }
 
+func (ls *listenSocket) getParentID() int64 {
+	return ls.pid
+}
+
 type normalSocket struct {
 	refName string
 	s       Socket
@@ -343,6 +538,10 @@ func (ns *normalSocket) deleteSelfIfReady() {
 	grpclog.Errorf("cannot call deleteSelfIfReady on a normal socket")
 }
 
+func (ns *normalSocket) getParentID() int64 {
+	return ns.pid
+}
+
 // ServerMetric defines the info channelz provides for a specific Server, which
 // includes ServerInternalMetric and channelz-specific data, such as channelz id,
 // child list, etc.
@@ -370,7 +569,6 @@ type ServerInternalMetric struct {
 	CallsFailed int64
 	// The last time a call was started on the server.
 	LastCallStartedTimestamp time.Time
-	//TODO: trace
 }
 
 // Server is the interface to be satisfied in order to be tracked by channelz as
@@ -417,3 +615,88 @@ func (s *server) deleteSelfIfReady() {
 	}
 	s.cm.deleteEntry(s.id)
 }
+
+func (s *server) getParentID() int64 {
+	return 0
+}
+
+type tracedChannel interface {
+	getChannelTrace() *channelTrace
+	incrTraceRefCount()
+	decrTraceRefCount()
+	getRefName() string
+}
+
+type channelTrace struct {
+	cm          *channelMap
+	createdTime time.Time
+	eventCount  int64
+	mu          sync.Mutex
+	events      []*TraceEvent
+}
+
+func (c *channelTrace) append(e *TraceEvent) {
+	c.mu.Lock()
+	if len(c.events) == getMaxTraceEntry() {
+		del := c.events[0]
+		c.events = c.events[1:]
+		if del.RefID != 0 {
+			// start recursive cleanup in a goroutine to not block the call originated from grpc.
+			go func() {
+				// need to acquire c.cm.mu lock to call the unlocked attemptCleanup func.
+				c.cm.mu.Lock()
+				c.cm.decrTraceRefCount(del.RefID)
+				c.cm.mu.Unlock()
+			}()
+		}
+	}
+	e.Timestamp = time.Now()
+	c.events = append(c.events, e)
+	c.eventCount++
+	c.mu.Unlock()
+}
+
+func (c *channelTrace) clear() {
+	c.mu.Lock()
+	for _, e := range c.events {
+		if e.RefID != 0 {
+			// caller should have already held the c.cm.mu lock.
+			c.cm.decrTraceRefCount(e.RefID)
+		}
+	}
+	c.mu.Unlock()
+}
+
+// Severity is the severity level of a trace event.
+// The canonical enumeration of all valid values is here:
+// https://github.com/grpc/grpc-proto/blob/9b13d199cc0d4703c7ea26c9c330ba695866eb23/grpc/channelz/v1/channelz.proto#L126.
+type Severity int
+
+const (
+	// CtUNKNOWN indicates unknown severity of a trace event.
+	CtUNKNOWN Severity = iota
+	// CtINFO indicates info level severity of a trace event.
+	CtINFO
+	// CtWarning indicates warning level severity of a trace event.
+	CtWarning
+	// CtError indicates error level severity of a trace event.
+	CtError
+)
+
+// RefChannelType is the type of the entity being referenced in a trace event.
+type RefChannelType int
+
+const (
+	// RefChannel indicates the referenced entity is a Channel.
+	RefChannel RefChannelType = iota
+	// RefSubChannel indicates the referenced entity is a SubChannel.
+	RefSubChannel
+)
+
+func (c *channelTrace) dumpData() *ChannelTrace {
+	c.mu.Lock()
+	ct := &ChannelTrace{EventNum: c.eventCount, CreationTime: c.createdTime}
+	ct.Events = c.events[:len(c.events)]
+	c.mu.Unlock()
+	return ct
+}

vendor/google.golang.org/grpc/internal/channelz/types_nonlinux.go

@@ -20,11 +20,13 @@
 
 package channelz
 
-import "google.golang.org/grpc/grpclog"
+import (
+	"sync"
 
-func init() {
-	grpclog.Infof("Channelz: socket options are not supported on non-linux os and appengine.")
-}
+	"google.golang.org/grpc/grpclog"
+)
+
+var once sync.Once
 
 // SocketOptionData defines the struct to hold socket option data, and related
 // getter function to obtain info from fd.
@@ -35,4 +37,8 @@ type SocketOptionData struct {
 // Getsockopt defines the function to get socket options requested by channelz.
 // It is to be passed to syscall.RawConn.Control().
 // Windows OS doesn't support Socket Option
-func (s *SocketOptionData) Getsockopt(fd uintptr) {}
+func (s *SocketOptionData) Getsockopt(fd uintptr) {
+	once.Do(func() {
+		grpclog.Warningln("Channelz: socket options are not supported on non-linux os and appengine.")
+	})
+}

vendor/google.golang.org/grpc/internal/internal.go

@@ -20,9 +20,24 @@
 // symbols to avoid circular dependencies.
 package internal
 
+import "golang.org/x/net/context"
+
 var (
 	// WithContextDialer is exported by clientconn.go
 	WithContextDialer interface{} // func(context.Context, string) (net.Conn, error) grpc.DialOption
 	// WithResolverBuilder is exported by clientconn.go
 	WithResolverBuilder interface{} // func (resolver.Builder) grpc.DialOption
+	// HealthCheckFunc is used to provide client-side LB channel health checking
+	HealthCheckFunc func(ctx context.Context, newStream func() (interface{}, error), reportHealth func(bool), serviceName string) error
+)
+
+const (
+	// CredsBundleModeFallback switches GoogleDefaultCreds to fallback mode.
+	CredsBundleModeFallback = "fallback"
+	// CredsBundleModeBalancer switches GoogleDefaultCreds to grpclb balancer
+	// mode.
+	CredsBundleModeBalancer = "balancer"
+	// CredsBundleModeBackendFromBalancer switches GoogleDefaultCreds to mode
+	// that supports backend returned by grpclb balancer.
+	CredsBundleModeBackendFromBalancer = "backend-from-balancer"
 )

vendor/google.golang.org/grpc/internal/transport/http2_client.go

@@ -73,7 +73,7 @@ type http2Client struct {
 
 	isSecure bool
 
-	creds []credentials.PerRPCCredentials
+	perRPCCreds []credentials.PerRPCCredentials
 
 	// Boolean to keep track of reading activity on transport.
 	// 1 is true and 0 is false.
@@ -112,6 +112,9 @@ type http2Client struct {
 	// Fields below are for channelz metric collection.
 	channelzID int64 // channelz unique identification number
 	czData     *channelzData
+
+	onGoAway func(GoAwayReason)
+	onClose  func()
 }
 
 func dial(ctx context.Context, fn func(context.Context, string) (net.Conn, error), addr string) (net.Conn, error) {
@@ -140,7 +143,7 @@ func isTemporary(err error) bool {
 // newHTTP2Client constructs a connected ClientTransport to addr based on HTTP2
 // and starts to receive messages on it. Non-nil error returns if construction
 // fails.
-func newHTTP2Client(connectCtx, ctx context.Context, addr TargetInfo, opts ConnectOptions, onSuccess func()) (_ *http2Client, err error) {
+func newHTTP2Client(connectCtx, ctx context.Context, addr TargetInfo, opts ConnectOptions, onSuccess func(), onGoAway func(GoAwayReason), onClose func()) (_ *http2Client, err error) {
 	scheme := "http"
 	ctx, cancel := context.WithCancel(ctx)
 	defer func() {
@@ -166,9 +169,20 @@ func newHTTP2Client(connectCtx, ctx context.Context, addr TargetInfo, opts Conne
 		isSecure bool
 		authInfo credentials.AuthInfo
 	)
-	if creds := opts.TransportCredentials; creds != nil {
+	transportCreds := opts.TransportCredentials
+	perRPCCreds := opts.PerRPCCredentials
+
+	if b := opts.CredsBundle; b != nil {
+		if t := b.TransportCredentials(); t != nil {
+			transportCreds = t
+		}
+		if t := b.PerRPCCredentials(); t != nil {
+			perRPCCreds = append(perRPCCreds, t)
+		}
+	}
+	if transportCreds != nil {
 		scheme = "https"
-		conn, authInfo, err = creds.ClientHandshake(connectCtx, addr.Authority, conn)
+		conn, authInfo, err = transportCreds.ClientHandshake(connectCtx, addr.Authority, conn)
 		if err != nil {
 			return nil, connectionErrorf(isTemporary(err), err, "transport: authentication handshake failed: %v", err)
 		}
@@ -213,7 +227,7 @@ func newHTTP2Client(connectCtx, ctx context.Context, addr TargetInfo, opts Conne
 		scheme:                scheme,
 		activeStreams:         make(map[uint32]*Stream),
 		isSecure:              isSecure,
-		creds:                 opts.PerRPCCredentials,
+		perRPCCreds:           perRPCCreds,
 		kp:                    kp,
 		statsHandler:          opts.StatsHandler,
 		initialWindowSize:     initialWindowSize,
@@ -223,6 +237,8 @@ func newHTTP2Client(connectCtx, ctx context.Context, addr TargetInfo, opts Conne
 		streamQuota:           defaultMaxStreamsClient,
 		streamsQuotaAvailable: make(chan struct{}, 1),
 		czData:                new(channelzData),
+		onGoAway:              onGoAway,
+		onClose:               onClose,
 	}
 	t.controlBuf = newControlBuffer(t.ctxDone)
 	if opts.InitialWindowSize >= defaultWindowSize {
@@ -259,6 +275,7 @@ func newHTTP2Client(connectCtx, ctx context.Context, addr TargetInfo, opts Conne
 	// a dedicated goroutine which reads HTTP2 frame from network. Then it
 	// dispatches the frame to the corresponding stream entity.
 	go t.reader()
+
 	// Send connection preface to server.
 	n, err := t.conn.Write(clientPreface)
 	if err != nil {
@@ -295,6 +312,7 @@ func newHTTP2Client(connectCtx, ctx context.Context, addr TargetInfo, opts Conne
 			return nil, connectionErrorf(true, err, "transport: failed to write window update: %v", err)
 		}
 	}
+
 	t.framer.writer.Flush()
 	go func() {
 		t.loopy = newLoopyWriter(clientSide, t.framer, t.controlBuf, t.bdpEst)
@@ -443,7 +461,7 @@ func (t *http2Client) createHeaderFields(ctx context.Context, callHdr *CallHdr)
 
 func (t *http2Client) createAudience(callHdr *CallHdr) string {
 	// Create an audience string only if needed.
-	if len(t.creds) == 0 && callHdr.Creds == nil {
+	if len(t.perRPCCreds) == 0 && callHdr.Creds == nil {
 		return ""
 	}
 	// Construct URI required to get auth request metadata.
@@ -458,7 +476,7 @@ func (t *http2Client) createAudience(callHdr *CallHdr) string {
 
 func (t *http2Client) getTrAuthData(ctx context.Context, audience string) (map[string]string, error) {
 	authData := map[string]string{}
-	for _, c := range t.creds {
+	for _, c := range t.perRPCCreds {
 		data, err := c.GetRequestMetadata(ctx, audience)
 		if err != nil {
 			if _, ok := status.FromError(err); ok {
@@ -664,7 +682,9 @@ func (t *http2Client) CloseStream(s *Stream, err error) {
 func (t *http2Client) closeStream(s *Stream, err error, rst bool, rstCode http2.ErrCode, st *status.Status, mdata map[string][]string, eosReceived bool) {
 	// Set stream status to done.
 	if s.swapState(streamDone) == streamDone {
-		// If it was already done, return.
+		// If it was already done, return.  If multiple closeStream calls
+		// happen simultaneously, wait for the first to finish.
+		<-s.done
 		return
 	}
 	// status and trailers can be updated here without any synchronization because the stream goroutine will
@@ -678,8 +698,6 @@ func (t *http2Client) closeStream(s *Stream, err error, rst bool, rstCode http2.
 		// This will unblock reads eventually.
 		s.write(recvMsg{err: err})
 	}
-	// This will unblock write.
-	close(s.done)
 	// If headerChan isn't closed, then close it.
 	if atomic.SwapUint32(&s.headerDone, 1) == 0 {
 		s.noHeaders = true
@@ -715,11 +733,17 @@ func (t *http2Client) closeStream(s *Stream, err error, rst bool, rstCode http2.
 		return true
 	}
 	t.controlBuf.executeAndPut(addBackStreamQuota, cleanup)
+	// This will unblock write.
+	close(s.done)
 }
 
 // Close kicks off the shutdown process of the transport. This should be called
 // only once on a transport. Once it is called, the transport should not be
 // accessed any more.
+//
+// This method blocks until the addrConn that initiated this transport is
+// re-connected. This happens because t.onClose() begins reconnect logic at the
+// addrConn level and blocks until the addrConn is successfully connected.
 func (t *http2Client) Close() error {
 	t.mu.Lock()
 	// Make sure we only Close once.
@@ -747,6 +771,7 @@ func (t *http2Client) Close() error {
 		}
 		t.statsHandler.HandleConn(t.ctx, connEnd)
 	}
+	go t.onClose()
 	return err
 }
 
@@ -1043,6 +1068,9 @@ func (t *http2Client) handleGoAway(f *http2.GoAwayFrame) {
 		close(t.goAway)
 		t.state = draining
 		t.controlBuf.put(&incomingGoAway{})
+
+		// This has to be a new goroutine because we're still using the current goroutine to read in the transport.
+		t.onGoAway(t.goAwayReason)
 	}
 	// All streams with IDs greater than the GoAwayId
 	// and smaller than the previous GoAway ID should be killed.
@@ -1145,7 +1173,9 @@ func (t *http2Client) operateHeaders(frame *http2.MetaHeadersFrame) {
 	if !endStream {
 		return
 	}
-	t.closeStream(s, io.EOF, false, http2.ErrCodeNo, state.status(), state.mdata, true)
+	// if client received END_STREAM from server while stream was still active, send RST_STREAM
+	rst := s.getState() == streamActive
+	t.closeStream(s, io.EOF, rst, http2.ErrCodeNo, state.status(), state.mdata, true)
 }
 
 // reader runs as a separate goroutine in charge of reading data from network
@@ -1159,15 +1189,16 @@ func (t *http2Client) reader() {
 	// Check the validity of server preface.
 	frame, err := t.framer.fr.ReadFrame()
 	if err != nil {
-		t.Close()
+		t.Close() // this kicks off resetTransport, so must be last before return
 		return
 	}
+	t.conn.SetReadDeadline(time.Time{}) // reset deadline once we get the settings frame (we didn't time out, yay!)
 	if t.keepaliveEnabled {
 		atomic.CompareAndSwapUint32(&t.activity, 0, 1)
 	}
 	sf, ok := frame.(*http2.SettingsFrame)
 	if !ok {
-		t.Close()
+		t.Close() // this kicks off resetTransport, so must be last before return
 		return
 	}
 	t.onSuccess()

vendor/google.golang.org/grpc/internal/transport/http_util.go

@@ -24,6 +24,7 @@ import (
 	"encoding/base64"
 	"fmt"
 	"io"
+	"math"
 	"net"
 	"net/http"
 	"strconv"
@@ -435,6 +436,10 @@ func decodeTimeout(s string) (time.Duration, error) {
 	if size < 2 {
 		return 0, fmt.Errorf("transport: timeout string is too short: %q", s)
 	}
+	if size > 9 {
+		// Spec allows for 8 digits plus the unit.
+		return 0, fmt.Errorf("transport: timeout string is too long: %q", s)
+	}
 	unit := timeoutUnit(s[size-1])
 	d, ok := timeoutUnitToDuration(unit)
 	if !ok {
@@ -444,6 +449,11 @@ func decodeTimeout(s string) (time.Duration, error) {
 	if err != nil {
 		return 0, err
 	}
+	const maxHours = math.MaxInt64 / int64(time.Hour)
+	if d == time.Hour && t > maxHours {
+		// This timeout would overflow math.MaxInt64; clamp it.
+		return time.Duration(math.MaxInt64), nil
+	}
 	return d * time.Duration(t), nil
 }
 

vendor/google.golang.org/grpc/internal/transport/transport.go

@@ -465,8 +465,12 @@ type ConnectOptions struct {
 	FailOnNonTempDialError bool
 	// PerRPCCredentials stores the PerRPCCredentials required to issue RPCs.
 	PerRPCCredentials []credentials.PerRPCCredentials
-	// TransportCredentials stores the Authenticator required to setup a client connection.
+	// TransportCredentials stores the Authenticator required to setup a client
+	// connection. Only one of TransportCredentials and CredsBundle is non-nil.
 	TransportCredentials credentials.TransportCredentials
+	// CredsBundle is the credentials bundle to be used. Only one of
+	// TransportCredentials and CredsBundle is non-nil.
+	CredsBundle credentials.Bundle
 	// KeepaliveParams stores the keepalive parameters.
 	KeepaliveParams keepalive.ClientParameters
 	// StatsHandler stores the handler for stats.
@@ -494,8 +498,8 @@ type TargetInfo struct {
 
 // NewClientTransport establishes the transport with the required ConnectOptions
 // and returns it to the caller.
-func NewClientTransport(connectCtx, ctx context.Context, target TargetInfo, opts ConnectOptions, onSuccess func()) (ClientTransport, error) {
-	return newHTTP2Client(connectCtx, ctx, target, opts, onSuccess)
+func NewClientTransport(connectCtx, ctx context.Context, target TargetInfo, opts ConnectOptions, onSuccess func(), onGoAway func(GoAwayReason), onClose func()) (ClientTransport, error) {
+	return newHTTP2Client(connectCtx, ctx, target, opts, onSuccess, onGoAway, onClose)
 }
 
 // Options provides additional hints and information for message