1
0
Fork 0

Added support for Haystack tracing

This commit is contained in:
Alex Antonov 2019-05-08 17:14:04 -05:00 committed by Traefiker Bot
parent 681892148e
commit 9cf6827ccc
274 changed files with 38070 additions and 13436 deletions

View file

@ -0,0 +1,78 @@
/*
*
* Copyright 2017 gRPC authors.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*/
// Package backoff implement the backoff strategy for gRPC.
//
// This is kept in internal until the gRPC project decides whether or not to
// allow alternative backoff strategies.
package backoff
import (
"time"
"google.golang.org/grpc/internal/grpcrand"
)
// Strategy defines the methodology for backing off after a grpc connection
// failure.
//
type Strategy interface {
// Backoff returns the amount of time to wait before the next retry given
// the number of consecutive failures.
Backoff(retries int) time.Duration
}
const (
// baseDelay is the amount of time to wait before retrying after the first
// failure.
baseDelay = 1.0 * time.Second
// factor is applied to the backoff after each retry.
factor = 1.6
// jitter provides a range to randomize backoff delays.
jitter = 0.2
)
// Exponential implements exponential backoff algorithm as defined in
// https://github.com/grpc/grpc/blob/master/doc/connection-backoff.md.
type Exponential struct {
// MaxDelay is the upper bound of backoff delay.
MaxDelay time.Duration
}
// Backoff returns the amount of time to wait before the next retry given the
// number of retries.
func (bc Exponential) Backoff(retries int) time.Duration {
if retries == 0 {
return baseDelay
}
backoff, max := float64(baseDelay), float64(bc.MaxDelay)
for backoff < max && retries > 0 {
backoff *= factor
retries--
}
if backoff > max {
backoff = max
}
// Randomize backoff delays so that if a cluster of requests start at
// the same time, they won't operate in lockstep.
backoff *= 1 + jitter*(grpcrand.Float64()*2-1)
if backoff < 0 {
return 0
}
return time.Duration(backoff)
}

View file

@ -0,0 +1,46 @@
/*
* Copyright 2019 gRPC authors.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
// Package balancerload defines APIs to parse server loads in trailers. The
// parsed loads are sent to balancers in DoneInfo.
package balancerload
import (
"google.golang.org/grpc/metadata"
)
// Parser converts loads from metadata into a concrete type.
type Parser interface {
// Parse parses loads from metadata.
Parse(md metadata.MD) interface{}
}
var parser Parser
// SetParser sets the load parser.
//
// Not mutex-protected, should be called before any gRPC functions.
func SetParser(lr Parser) {
parser = lr
}
// Parse calls parser.Read().
func Parse(md metadata.MD) interface{} {
if parser == nil {
return nil
}
return parser.Parse(md)
}

View file

@ -0,0 +1,167 @@
/*
*
* Copyright 2018 gRPC authors.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*/
// Package binarylog implementation binary logging as defined in
// https://github.com/grpc/proposal/blob/master/A16-binary-logging.md.
package binarylog
import (
"fmt"
"os"
"google.golang.org/grpc/grpclog"
)
// Logger is the global binary logger. It can be used to get binary logger for
// each method.
type Logger interface {
getMethodLogger(methodName string) *MethodLogger
}
// binLogger is the global binary logger for the binary. One of this should be
// built at init time from the configuration (environment varialbe or flags).
//
// It is used to get a methodLogger for each individual method.
var binLogger Logger
// SetLogger sets the binarg logger.
//
// Only call this at init time.
func SetLogger(l Logger) {
binLogger = l
}
// GetMethodLogger returns the methodLogger for the given methodName.
//
// methodName should be in the format of "/service/method".
//
// Each methodLogger returned by this method is a new instance. This is to
// generate sequence id within the call.
func GetMethodLogger(methodName string) *MethodLogger {
if binLogger == nil {
return nil
}
return binLogger.getMethodLogger(methodName)
}
func init() {
const envStr = "GRPC_BINARY_LOG_FILTER"
configStr := os.Getenv(envStr)
binLogger = NewLoggerFromConfigString(configStr)
}
type methodLoggerConfig struct {
// Max length of header and message.
hdr, msg uint64
}
type logger struct {
all *methodLoggerConfig
services map[string]*methodLoggerConfig
methods map[string]*methodLoggerConfig
blacklist map[string]struct{}
}
// newEmptyLogger creates an empty logger. The map fields need to be filled in
// using the set* functions.
func newEmptyLogger() *logger {
return &logger{}
}
// Set method logger for "*".
func (l *logger) setDefaultMethodLogger(ml *methodLoggerConfig) error {
if l.all != nil {
return fmt.Errorf("conflicting global rules found")
}
l.all = ml
return nil
}
// Set method logger for "service/*".
//
// New methodLogger with same service overrides the old one.
func (l *logger) setServiceMethodLogger(service string, ml *methodLoggerConfig) error {
if _, ok := l.services[service]; ok {
return fmt.Errorf("conflicting rules for service %v found", service)
}
if l.services == nil {
l.services = make(map[string]*methodLoggerConfig)
}
l.services[service] = ml
return nil
}
// Set method logger for "service/method".
//
// New methodLogger with same method overrides the old one.
func (l *logger) setMethodMethodLogger(method string, ml *methodLoggerConfig) error {
if _, ok := l.blacklist[method]; ok {
return fmt.Errorf("conflicting rules for method %v found", method)
}
if _, ok := l.methods[method]; ok {
return fmt.Errorf("conflicting rules for method %v found", method)
}
if l.methods == nil {
l.methods = make(map[string]*methodLoggerConfig)
}
l.methods[method] = ml
return nil
}
// Set blacklist method for "-service/method".
func (l *logger) setBlacklist(method string) error {
if _, ok := l.blacklist[method]; ok {
return fmt.Errorf("conflicting rules for method %v found", method)
}
if _, ok := l.methods[method]; ok {
return fmt.Errorf("conflicting rules for method %v found", method)
}
if l.blacklist == nil {
l.blacklist = make(map[string]struct{})
}
l.blacklist[method] = struct{}{}
return nil
}
// getMethodLogger returns the methodLogger for the given methodName.
//
// methodName should be in the format of "/service/method".
//
// Each methodLogger returned by this method is a new instance. This is to
// generate sequence id within the call.
func (l *logger) getMethodLogger(methodName string) *MethodLogger {
s, m, err := parseMethodName(methodName)
if err != nil {
grpclog.Infof("binarylogging: failed to parse %q: %v", methodName, err)
return nil
}
if ml, ok := l.methods[s+"/"+m]; ok {
return newMethodLogger(ml.hdr, ml.msg)
}
if _, ok := l.blacklist[s+"/"+m]; ok {
return nil
}
if ml, ok := l.services[s]; ok {
return newMethodLogger(ml.hdr, ml.msg)
}
if l.all == nil {
return nil
}
return newMethodLogger(l.all.hdr, l.all.msg)
}

View file

@ -0,0 +1,42 @@
/*
*
* Copyright 2018 gRPC authors.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*/
// This file contains exported variables/functions that are exported for testing
// only.
//
// An ideal way for this would be to put those in a *_test.go but in binarylog
// package. But this doesn't work with staticcheck with go module. Error was:
// "MdToMetadataProto not declared by package binarylog". This could be caused
// by the way staticcheck looks for files for a certain package, which doesn't
// support *_test.go files.
//
// Move those to binary_test.go when staticcheck is fixed.
package binarylog
var (
// AllLogger is a logger that logs all headers/messages for all RPCs. It's
// for testing only.
AllLogger = NewLoggerFromConfigString("*")
// MdToMetadataProto converts metadata to a binary logging proto message.
// It's for testing only.
MdToMetadataProto = mdToMetadataProto
// AddrToProto converts an address to a binary logging proto message. It's
// for testing only.
AddrToProto = addrToProto
)

View file

@ -0,0 +1,210 @@
/*
*
* Copyright 2018 gRPC authors.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*/
package binarylog
import (
"errors"
"fmt"
"regexp"
"strconv"
"strings"
"google.golang.org/grpc/grpclog"
)
// NewLoggerFromConfigString reads the string and build a logger. It can be used
// to build a new logger and assign it to binarylog.Logger.
//
// Example filter config strings:
// - "" Nothing will be logged
// - "*" All headers and messages will be fully logged.
// - "*{h}" Only headers will be logged.
// - "*{m:256}" Only the first 256 bytes of each message will be logged.
// - "Foo/*" Logs every method in service Foo
// - "Foo/*,-Foo/Bar" Logs every method in service Foo except method /Foo/Bar
// - "Foo/*,Foo/Bar{m:256}" Logs the first 256 bytes of each message in method
// /Foo/Bar, logs all headers and messages in every other method in service
// Foo.
//
// If two configs exist for one certain method or service, the one specified
// later overrides the privous config.
func NewLoggerFromConfigString(s string) Logger {
if s == "" {
return nil
}
l := newEmptyLogger()
methods := strings.Split(s, ",")
for _, method := range methods {
if err := l.fillMethodLoggerWithConfigString(method); err != nil {
grpclog.Warningf("failed to parse binary log config: %v", err)
return nil
}
}
return l
}
// fillMethodLoggerWithConfigString parses config, creates methodLogger and adds
// it to the right map in the logger.
func (l *logger) fillMethodLoggerWithConfigString(config string) error {
// "" is invalid.
if config == "" {
return errors.New("empty string is not a valid method binary logging config")
}
// "-service/method", blacklist, no * or {} allowed.
if config[0] == '-' {
s, m, suffix, err := parseMethodConfigAndSuffix(config[1:])
if err != nil {
return fmt.Errorf("invalid config: %q, %v", config, err)
}
if m == "*" {
return fmt.Errorf("invalid config: %q, %v", config, "* not allowd in blacklist config")
}
if suffix != "" {
return fmt.Errorf("invalid config: %q, %v", config, "header/message limit not allowed in blacklist config")
}
if err := l.setBlacklist(s + "/" + m); err != nil {
return fmt.Errorf("invalid config: %v", err)
}
return nil
}
// "*{h:256;m:256}"
if config[0] == '*' {
hdr, msg, err := parseHeaderMessageLengthConfig(config[1:])
if err != nil {
return fmt.Errorf("invalid config: %q, %v", config, err)
}
if err := l.setDefaultMethodLogger(&methodLoggerConfig{hdr: hdr, msg: msg}); err != nil {
return fmt.Errorf("invalid config: %v", err)
}
return nil
}
s, m, suffix, err := parseMethodConfigAndSuffix(config)
if err != nil {
return fmt.Errorf("invalid config: %q, %v", config, err)
}
hdr, msg, err := parseHeaderMessageLengthConfig(suffix)
if err != nil {
return fmt.Errorf("invalid header/message length config: %q, %v", suffix, err)
}
if m == "*" {
if err := l.setServiceMethodLogger(s, &methodLoggerConfig{hdr: hdr, msg: msg}); err != nil {
return fmt.Errorf("invalid config: %v", err)
}
} else {
if err := l.setMethodMethodLogger(s+"/"+m, &methodLoggerConfig{hdr: hdr, msg: msg}); err != nil {
return fmt.Errorf("invalid config: %v", err)
}
}
return nil
}
const (
// TODO: this const is only used by env_config now. But could be useful for
// other config. Move to binarylog.go if necessary.
maxUInt = ^uint64(0)
// For "p.s/m" plus any suffix. Suffix will be parsed again. See test for
// expected output.
longMethodConfigRegexpStr = `^([\w./]+)/((?:\w+)|[*])(.+)?$`
// For suffix from above, "{h:123,m:123}". See test for expected output.
optionalLengthRegexpStr = `(?::(\d+))?` // Optional ":123".
headerConfigRegexpStr = `^{h` + optionalLengthRegexpStr + `}$`
messageConfigRegexpStr = `^{m` + optionalLengthRegexpStr + `}$`
headerMessageConfigRegexpStr = `^{h` + optionalLengthRegexpStr + `;m` + optionalLengthRegexpStr + `}$`
)
var (
longMethodConfigRegexp = regexp.MustCompile(longMethodConfigRegexpStr)
headerConfigRegexp = regexp.MustCompile(headerConfigRegexpStr)
messageConfigRegexp = regexp.MustCompile(messageConfigRegexpStr)
headerMessageConfigRegexp = regexp.MustCompile(headerMessageConfigRegexpStr)
)
// Turn "service/method{h;m}" into "service", "method", "{h;m}".
func parseMethodConfigAndSuffix(c string) (service, method, suffix string, _ error) {
// Regexp result:
//
// in: "p.s/m{h:123,m:123}",
// out: []string{"p.s/m{h:123,m:123}", "p.s", "m", "{h:123,m:123}"},
match := longMethodConfigRegexp.FindStringSubmatch(c)
if match == nil {
return "", "", "", fmt.Errorf("%q contains invalid substring", c)
}
service = match[1]
method = match[2]
suffix = match[3]
return
}
// Turn "{h:123;m:345}" into 123, 345.
//
// Return maxUInt if length is unspecified.
func parseHeaderMessageLengthConfig(c string) (hdrLenStr, msgLenStr uint64, err error) {
if c == "" {
return maxUInt, maxUInt, nil
}
// Header config only.
if match := headerConfigRegexp.FindStringSubmatch(c); match != nil {
if s := match[1]; s != "" {
hdrLenStr, err = strconv.ParseUint(s, 10, 64)
if err != nil {
return 0, 0, fmt.Errorf("failed to convert %q to uint", s)
}
return hdrLenStr, 0, nil
}
return maxUInt, 0, nil
}
// Message config only.
if match := messageConfigRegexp.FindStringSubmatch(c); match != nil {
if s := match[1]; s != "" {
msgLenStr, err = strconv.ParseUint(s, 10, 64)
if err != nil {
return 0, 0, fmt.Errorf("failed to convert %q to uint", s)
}
return 0, msgLenStr, nil
}
return 0, maxUInt, nil
}
// Header and message config both.
if match := headerMessageConfigRegexp.FindStringSubmatch(c); match != nil {
// Both hdr and msg are specified, but one or two of them might be empty.
hdrLenStr = maxUInt
msgLenStr = maxUInt
if s := match[1]; s != "" {
hdrLenStr, err = strconv.ParseUint(s, 10, 64)
if err != nil {
return 0, 0, fmt.Errorf("failed to convert %q to uint", s)
}
}
if s := match[2]; s != "" {
msgLenStr, err = strconv.ParseUint(s, 10, 64)
if err != nil {
return 0, 0, fmt.Errorf("failed to convert %q to uint", s)
}
}
return hdrLenStr, msgLenStr, nil
}
return 0, 0, fmt.Errorf("%q contains invalid substring", c)
}

View file

@ -0,0 +1,423 @@
/*
*
* Copyright 2018 gRPC authors.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*/
package binarylog
import (
"net"
"strings"
"sync/atomic"
"time"
"github.com/golang/protobuf/proto"
"github.com/golang/protobuf/ptypes"
pb "google.golang.org/grpc/binarylog/grpc_binarylog_v1"
"google.golang.org/grpc/grpclog"
"google.golang.org/grpc/metadata"
"google.golang.org/grpc/status"
)
type callIDGenerator struct {
id uint64
}
func (g *callIDGenerator) next() uint64 {
id := atomic.AddUint64(&g.id, 1)
return id
}
// reset is for testing only, and doesn't need to be thread safe.
func (g *callIDGenerator) reset() {
g.id = 0
}
var idGen callIDGenerator
// MethodLogger is the sub-logger for each method.
type MethodLogger struct {
headerMaxLen, messageMaxLen uint64
callID uint64
idWithinCallGen *callIDGenerator
sink Sink // TODO(blog): make this plugable.
}
func newMethodLogger(h, m uint64) *MethodLogger {
return &MethodLogger{
headerMaxLen: h,
messageMaxLen: m,
callID: idGen.next(),
idWithinCallGen: &callIDGenerator{},
sink: defaultSink, // TODO(blog): make it plugable.
}
}
// Log creates a proto binary log entry, and logs it to the sink.
func (ml *MethodLogger) Log(c LogEntryConfig) {
m := c.toProto()
timestamp, _ := ptypes.TimestampProto(time.Now())
m.Timestamp = timestamp
m.CallId = ml.callID
m.SequenceIdWithinCall = ml.idWithinCallGen.next()
switch pay := m.Payload.(type) {
case *pb.GrpcLogEntry_ClientHeader:
m.PayloadTruncated = ml.truncateMetadata(pay.ClientHeader.GetMetadata())
case *pb.GrpcLogEntry_ServerHeader:
m.PayloadTruncated = ml.truncateMetadata(pay.ServerHeader.GetMetadata())
case *pb.GrpcLogEntry_Message:
m.PayloadTruncated = ml.truncateMessage(pay.Message)
}
ml.sink.Write(m)
}
func (ml *MethodLogger) truncateMetadata(mdPb *pb.Metadata) (truncated bool) {
if ml.headerMaxLen == maxUInt {
return false
}
var (
bytesLimit = ml.headerMaxLen
index int
)
// At the end of the loop, index will be the first entry where the total
// size is greater than the limit:
//
// len(entry[:index]) <= ml.hdr && len(entry[:index+1]) > ml.hdr.
for ; index < len(mdPb.Entry); index++ {
entry := mdPb.Entry[index]
if entry.Key == "grpc-trace-bin" {
// "grpc-trace-bin" is a special key. It's kept in the log entry,
// but not counted towards the size limit.
continue
}
currentEntryLen := uint64(len(entry.Value))
if currentEntryLen > bytesLimit {
break
}
bytesLimit -= currentEntryLen
}
truncated = index < len(mdPb.Entry)
mdPb.Entry = mdPb.Entry[:index]
return truncated
}
func (ml *MethodLogger) truncateMessage(msgPb *pb.Message) (truncated bool) {
if ml.messageMaxLen == maxUInt {
return false
}
if ml.messageMaxLen >= uint64(len(msgPb.Data)) {
return false
}
msgPb.Data = msgPb.Data[:ml.messageMaxLen]
return true
}
// LogEntryConfig represents the configuration for binary log entry.
type LogEntryConfig interface {
toProto() *pb.GrpcLogEntry
}
// ClientHeader configs the binary log entry to be a ClientHeader entry.
type ClientHeader struct {
OnClientSide bool
Header metadata.MD
MethodName string
Authority string
Timeout time.Duration
// PeerAddr is required only when it's on server side.
PeerAddr net.Addr
}
func (c *ClientHeader) toProto() *pb.GrpcLogEntry {
// This function doesn't need to set all the fields (e.g. seq ID). The Log
// function will set the fields when necessary.
clientHeader := &pb.ClientHeader{
Metadata: mdToMetadataProto(c.Header),
MethodName: c.MethodName,
Authority: c.Authority,
}
if c.Timeout > 0 {
clientHeader.Timeout = ptypes.DurationProto(c.Timeout)
}
ret := &pb.GrpcLogEntry{
Type: pb.GrpcLogEntry_EVENT_TYPE_CLIENT_HEADER,
Payload: &pb.GrpcLogEntry_ClientHeader{
ClientHeader: clientHeader,
},
}
if c.OnClientSide {
ret.Logger = pb.GrpcLogEntry_LOGGER_CLIENT
} else {
ret.Logger = pb.GrpcLogEntry_LOGGER_SERVER
}
if c.PeerAddr != nil {
ret.Peer = addrToProto(c.PeerAddr)
}
return ret
}
// ServerHeader configs the binary log entry to be a ServerHeader entry.
type ServerHeader struct {
OnClientSide bool
Header metadata.MD
// PeerAddr is required only when it's on client side.
PeerAddr net.Addr
}
func (c *ServerHeader) toProto() *pb.GrpcLogEntry {
ret := &pb.GrpcLogEntry{
Type: pb.GrpcLogEntry_EVENT_TYPE_SERVER_HEADER,
Payload: &pb.GrpcLogEntry_ServerHeader{
ServerHeader: &pb.ServerHeader{
Metadata: mdToMetadataProto(c.Header),
},
},
}
if c.OnClientSide {
ret.Logger = pb.GrpcLogEntry_LOGGER_CLIENT
} else {
ret.Logger = pb.GrpcLogEntry_LOGGER_SERVER
}
if c.PeerAddr != nil {
ret.Peer = addrToProto(c.PeerAddr)
}
return ret
}
// ClientMessage configs the binary log entry to be a ClientMessage entry.
type ClientMessage struct {
OnClientSide bool
// Message can be a proto.Message or []byte. Other messages formats are not
// supported.
Message interface{}
}
func (c *ClientMessage) toProto() *pb.GrpcLogEntry {
var (
data []byte
err error
)
if m, ok := c.Message.(proto.Message); ok {
data, err = proto.Marshal(m)
if err != nil {
grpclog.Infof("binarylogging: failed to marshal proto message: %v", err)
}
} else if b, ok := c.Message.([]byte); ok {
data = b
} else {
grpclog.Infof("binarylogging: message to log is neither proto.message nor []byte")
}
ret := &pb.GrpcLogEntry{
Type: pb.GrpcLogEntry_EVENT_TYPE_CLIENT_MESSAGE,
Payload: &pb.GrpcLogEntry_Message{
Message: &pb.Message{
Length: uint32(len(data)),
Data: data,
},
},
}
if c.OnClientSide {
ret.Logger = pb.GrpcLogEntry_LOGGER_CLIENT
} else {
ret.Logger = pb.GrpcLogEntry_LOGGER_SERVER
}
return ret
}
// ServerMessage configs the binary log entry to be a ServerMessage entry.
type ServerMessage struct {
OnClientSide bool
// Message can be a proto.Message or []byte. Other messages formats are not
// supported.
Message interface{}
}
func (c *ServerMessage) toProto() *pb.GrpcLogEntry {
var (
data []byte
err error
)
if m, ok := c.Message.(proto.Message); ok {
data, err = proto.Marshal(m)
if err != nil {
grpclog.Infof("binarylogging: failed to marshal proto message: %v", err)
}
} else if b, ok := c.Message.([]byte); ok {
data = b
} else {
grpclog.Infof("binarylogging: message to log is neither proto.message nor []byte")
}
ret := &pb.GrpcLogEntry{
Type: pb.GrpcLogEntry_EVENT_TYPE_SERVER_MESSAGE,
Payload: &pb.GrpcLogEntry_Message{
Message: &pb.Message{
Length: uint32(len(data)),
Data: data,
},
},
}
if c.OnClientSide {
ret.Logger = pb.GrpcLogEntry_LOGGER_CLIENT
} else {
ret.Logger = pb.GrpcLogEntry_LOGGER_SERVER
}
return ret
}
// ClientHalfClose configs the binary log entry to be a ClientHalfClose entry.
type ClientHalfClose struct {
OnClientSide bool
}
func (c *ClientHalfClose) toProto() *pb.GrpcLogEntry {
ret := &pb.GrpcLogEntry{
Type: pb.GrpcLogEntry_EVENT_TYPE_CLIENT_HALF_CLOSE,
Payload: nil, // No payload here.
}
if c.OnClientSide {
ret.Logger = pb.GrpcLogEntry_LOGGER_CLIENT
} else {
ret.Logger = pb.GrpcLogEntry_LOGGER_SERVER
}
return ret
}
// ServerTrailer configs the binary log entry to be a ServerTrailer entry.
type ServerTrailer struct {
OnClientSide bool
Trailer metadata.MD
// Err is the status error.
Err error
// PeerAddr is required only when it's on client side and the RPC is trailer
// only.
PeerAddr net.Addr
}
func (c *ServerTrailer) toProto() *pb.GrpcLogEntry {
st, ok := status.FromError(c.Err)
if !ok {
grpclog.Info("binarylogging: error in trailer is not a status error")
}
var (
detailsBytes []byte
err error
)
stProto := st.Proto()
if stProto != nil && len(stProto.Details) != 0 {
detailsBytes, err = proto.Marshal(stProto)
if err != nil {
grpclog.Infof("binarylogging: failed to marshal status proto: %v", err)
}
}
ret := &pb.GrpcLogEntry{
Type: pb.GrpcLogEntry_EVENT_TYPE_SERVER_TRAILER,
Payload: &pb.GrpcLogEntry_Trailer{
Trailer: &pb.Trailer{
Metadata: mdToMetadataProto(c.Trailer),
StatusCode: uint32(st.Code()),
StatusMessage: st.Message(),
StatusDetails: detailsBytes,
},
},
}
if c.OnClientSide {
ret.Logger = pb.GrpcLogEntry_LOGGER_CLIENT
} else {
ret.Logger = pb.GrpcLogEntry_LOGGER_SERVER
}
if c.PeerAddr != nil {
ret.Peer = addrToProto(c.PeerAddr)
}
return ret
}
// Cancel configs the binary log entry to be a Cancel entry.
type Cancel struct {
OnClientSide bool
}
func (c *Cancel) toProto() *pb.GrpcLogEntry {
ret := &pb.GrpcLogEntry{
Type: pb.GrpcLogEntry_EVENT_TYPE_CANCEL,
Payload: nil,
}
if c.OnClientSide {
ret.Logger = pb.GrpcLogEntry_LOGGER_CLIENT
} else {
ret.Logger = pb.GrpcLogEntry_LOGGER_SERVER
}
return ret
}
// metadataKeyOmit returns whether the metadata entry with this key should be
// omitted.
func metadataKeyOmit(key string) bool {
switch key {
case "lb-token", ":path", ":authority", "content-encoding", "content-type", "user-agent", "te":
return true
case "grpc-trace-bin": // grpc-trace-bin is special because it's visiable to users.
return false
}
return strings.HasPrefix(key, "grpc-")
}
func mdToMetadataProto(md metadata.MD) *pb.Metadata {
ret := &pb.Metadata{}
for k, vv := range md {
if metadataKeyOmit(k) {
continue
}
for _, v := range vv {
ret.Entry = append(ret.Entry,
&pb.MetadataEntry{
Key: k,
Value: []byte(v),
},
)
}
}
return ret
}
func addrToProto(addr net.Addr) *pb.Address {
ret := &pb.Address{}
switch a := addr.(type) {
case *net.TCPAddr:
if a.IP.To4() != nil {
ret.Type = pb.Address_TYPE_IPV4
} else if a.IP.To16() != nil {
ret.Type = pb.Address_TYPE_IPV6
} else {
ret.Type = pb.Address_TYPE_UNKNOWN
// Do not set address and port fields.
break
}
ret.Address = a.IP.String()
ret.IpPort = uint32(a.Port)
case *net.UnixAddr:
ret.Type = pb.Address_TYPE_UNIX
ret.Address = a.String()
default:
ret.Type = pb.Address_TYPE_UNKNOWN
}
return ret
}

View file

@ -0,0 +1,162 @@
/*
*
* Copyright 2018 gRPC authors.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*/
package binarylog
import (
"bufio"
"encoding/binary"
"fmt"
"io"
"io/ioutil"
"sync"
"time"
"github.com/golang/protobuf/proto"
pb "google.golang.org/grpc/binarylog/grpc_binarylog_v1"
"google.golang.org/grpc/grpclog"
)
var (
defaultSink Sink = &noopSink{} // TODO(blog): change this default (file in /tmp).
)
// SetDefaultSink sets the sink where binary logs will be written to.
//
// Not thread safe. Only set during initialization.
func SetDefaultSink(s Sink) {
if defaultSink != nil {
defaultSink.Close()
}
defaultSink = s
}
// Sink writes log entry into the binary log sink.
type Sink interface {
// Write will be called to write the log entry into the sink.
//
// It should be thread-safe so it can be called in parallel.
Write(*pb.GrpcLogEntry) error
// Close will be called when the Sink is replaced by a new Sink.
Close() error
}
type noopSink struct{}
func (ns *noopSink) Write(*pb.GrpcLogEntry) error { return nil }
func (ns *noopSink) Close() error { return nil }
// newWriterSink creates a binary log sink with the given writer.
//
// Write() marshalls the proto message and writes it to the given writer. Each
// message is prefixed with a 4 byte big endian unsigned integer as the length.
//
// No buffer is done, Close() doesn't try to close the writer.
func newWriterSink(w io.Writer) *writerSink {
return &writerSink{out: w}
}
type writerSink struct {
out io.Writer
}
func (ws *writerSink) Write(e *pb.GrpcLogEntry) error {
b, err := proto.Marshal(e)
if err != nil {
grpclog.Infof("binary logging: failed to marshal proto message: %v", err)
}
hdr := make([]byte, 4)
binary.BigEndian.PutUint32(hdr, uint32(len(b)))
if _, err := ws.out.Write(hdr); err != nil {
return err
}
if _, err := ws.out.Write(b); err != nil {
return err
}
return nil
}
func (ws *writerSink) Close() error { return nil }
type bufWriteCloserSink struct {
mu sync.Mutex
closer io.Closer
out *writerSink // out is built on buf.
buf *bufio.Writer // buf is kept for flush.
writeStartOnce sync.Once
writeTicker *time.Ticker
}
func (fs *bufWriteCloserSink) Write(e *pb.GrpcLogEntry) error {
// Start the write loop when Write is called.
fs.writeStartOnce.Do(fs.startFlushGoroutine)
fs.mu.Lock()
if err := fs.out.Write(e); err != nil {
fs.mu.Unlock()
return err
}
fs.mu.Unlock()
return nil
}
const (
bufFlushDuration = 60 * time.Second
)
func (fs *bufWriteCloserSink) startFlushGoroutine() {
fs.writeTicker = time.NewTicker(bufFlushDuration)
go func() {
for range fs.writeTicker.C {
fs.mu.Lock()
fs.buf.Flush()
fs.mu.Unlock()
}
}()
}
func (fs *bufWriteCloserSink) Close() error {
if fs.writeTicker != nil {
fs.writeTicker.Stop()
}
fs.mu.Lock()
fs.buf.Flush()
fs.closer.Close()
fs.out.Close()
fs.mu.Unlock()
return nil
}
func newBufWriteCloserSink(o io.WriteCloser) Sink {
bufW := bufio.NewWriter(o)
return &bufWriteCloserSink{
closer: o,
out: newWriterSink(bufW),
buf: bufW,
}
}
// NewTempFileSink creates a temp file and returns a Sink that writes to this
// file.
func NewTempFileSink() (Sink, error) {
tempFile, err := ioutil.TempFile("/tmp", "grpcgo_binarylog_*.txt")
if err != nil {
return nil, fmt.Errorf("failed to create temp file: %v", err)
}
return newBufWriteCloserSink(tempFile), nil
}

View file

@ -0,0 +1,41 @@
/*
*
* Copyright 2018 gRPC authors.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*/
package binarylog
import (
"errors"
"strings"
)
// parseMethodName splits service and method from the input. It expects format
// "/service/method".
//
// TODO: move to internal/grpcutil.
func parseMethodName(methodName string) (service, method string, _ error) {
if !strings.HasPrefix(methodName, "/") {
return "", "", errors.New("invalid method name: should start with /")
}
methodName = methodName[1:]
pos := strings.LastIndex(methodName, "/")
if pos < 0 {
return "", "", errors.New("invalid method name: suffix /method is missing")
}
return methodName[:pos], methodName[pos+1:], nil
}

View file

@ -0,0 +1,699 @@
/*
*
* Copyright 2018 gRPC authors.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*/
// Package channelz defines APIs for enabling channelz service, entry
// registration/deletion, and accessing channelz data. It also defines channelz
// metric struct formats.
//
// All APIs in this package are experimental.
package channelz
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 = int64(50)
curState int32
maxTraceEntry = defaultMaxTraceEntry
)
// TurnOn turns on channelz data collection.
func TurnOn() {
if !IsOn() {
NewChannelzStorage()
atomic.StoreInt32(&curState, 1)
}
}
// IsOn returns whether channelz data collection is on.
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 {
mu sync.RWMutex
DB *channelMap
}
func (d *dbWrapper) set(db *channelMap) {
d.mu.Lock()
d.DB = db
d.mu.Unlock()
}
func (d *dbWrapper) get() *channelMap {
d.mu.RLock()
defer d.mu.RUnlock()
return d.DB
}
// NewChannelzStorage initializes channelz data storage and id generator.
//
// Note: This function is exported for testing purpose only. User should not call
// it in most cases.
func NewChannelzStorage() {
db.set(&channelMap{
topLevelChannels: make(map[int64]struct{}),
channels: make(map[int64]*channel),
listenSockets: make(map[int64]*listenSocket),
normalSockets: make(map[int64]*normalSocket),
servers: make(map[int64]*server),
subChannels: make(map[int64]*subChannel),
})
idGen.reset()
}
// GetTopChannels returns a slice of top channel's ChannelMetric, along with a
// boolean indicating whether there's more top channels to be queried for.
//
// The arg id specifies that only top channel with id at or above it will be included
// in the result. The returned slice is up to a length of the arg maxResults or
// EntryPerPage if maxResults is zero, and is sorted in ascending id order.
func GetTopChannels(id int64, maxResults int64) ([]*ChannelMetric, bool) {
return db.get().GetTopChannels(id, maxResults)
}
// GetServers returns a slice of server's ServerMetric, along with a
// boolean indicating whether there's more servers to be queried for.
//
// The arg id specifies that only server with id at or above it will be included
// in the result. The returned slice is up to a length of the arg maxResults or
// EntryPerPage if maxResults is zero, and is sorted in ascending id order.
func GetServers(id int64, maxResults int64) ([]*ServerMetric, bool) {
return db.get().GetServers(id, maxResults)
}
// GetServerSockets returns a slice of server's (identified by id) normal socket's
// SocketMetric, along with a boolean indicating whether there's more sockets to
// be queried for.
//
// The arg startID specifies that only sockets with id at or above it will be
// included in the result. The returned slice is up to a length of the arg maxResults
// or EntryPerPage if maxResults is zero, and is sorted in ascending id order.
func GetServerSockets(id int64, startID int64, maxResults int64) ([]*SocketMetric, bool) {
return db.get().GetServerSockets(id, startID, maxResults)
}
// GetChannel returns the ChannelMetric for the channel (identified by id).
func GetChannel(id int64) *ChannelMetric {
return db.get().GetChannel(id)
}
// GetSubChannel returns the SubChannelMetric for the subchannel (identified by id).
func GetSubChannel(id int64) *SubChannelMetric {
return db.get().GetSubChannel(id)
}
// GetSocket returns the SocketInternalMetric for the socket (identified by id).
func GetSocket(id int64) *SocketMetric {
return db.get().GetSocket(id)
}
// GetServer returns the ServerMetric for the server (identified by id).
func GetServer(id int64) *ServerMetric {
return db.get().GetServer(id)
}
// RegisterChannel registers the given channel c in channelz database with ref
// as its reference name, and add it to the child list of its parent (identified
// by pid). pid = 0 means no parent. It returns the unique channelz tracking id
// assigned to this channel.
func RegisterChannel(c Channel, pid int64, ref string) int64 {
id := idGen.genID()
cn := &channel{
refName: ref,
c: c,
subChans: make(map[int64]string),
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)
} else {
db.get().addChannel(id, cn, false, pid, ref)
}
return id
}
// RegisterSubChannel registers the given channel c in channelz database with ref
// as its reference name, and add it to the child list of its parent (identified
// by pid). It returns the unique channelz tracking id assigned to this subchannel.
func RegisterSubChannel(c Channel, pid int64, ref string) int64 {
if pid == 0 {
grpclog.Error("a SubChannel's parent id cannot be 0")
return 0
}
id := idGen.genID()
sc := &subChannel{
refName: ref,
c: c,
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
}
// RegisterServer registers the given server s in channelz database. It returns
// the unique channelz tracking id assigned to this server.
func RegisterServer(s Server, ref string) int64 {
id := idGen.genID()
svr := &server{
refName: ref,
s: s,
sockets: make(map[int64]string),
listenSockets: make(map[int64]string),
id: id,
}
db.get().addServer(id, svr)
return id
}
// RegisterListenSocket registers the given listen socket s in channelz database
// with ref as its reference name, and add it to the child list of its parent
// (identified by pid). It returns the unique channelz tracking id assigned to
// this listen socket.
func RegisterListenSocket(s Socket, pid int64, ref string) int64 {
if pid == 0 {
grpclog.Error("a ListenSocket's parent id cannot be 0")
return 0
}
id := idGen.genID()
ls := &listenSocket{refName: ref, s: s, id: id, pid: pid}
db.get().addListenSocket(id, ls, pid, ref)
return id
}
// RegisterNormalSocket registers the given normal socket s in channelz database
// with ref as its reference name, and add it to the child list of its parent
// (identified by pid). It returns the unique channelz tracking id assigned to
// this normal socket.
func RegisterNormalSocket(s Socket, pid int64, ref string) int64 {
if pid == 0 {
grpclog.Error("a NormalSocket's parent id cannot be 0")
return 0
}
id := idGen.genID()
ns := &normalSocket{refName: ref, s: s, id: id, pid: pid}
db.get().addNormalSocket(id, ns, pid, ref)
return id
}
// RemoveEntry removes an entry with unique channelz trakcing id to be id from
// channelz database.
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.
// 2. Methods that can only be called when global lock is held.
// A second type of method need always to be called inside a first type of method.
type channelMap struct {
mu sync.RWMutex
topLevelChannels map[int64]struct{}
servers map[int64]*server
channels map[int64]*channel
subChannels map[int64]*subChannel
listenSockets map[int64]*listenSocket
normalSockets map[int64]*normalSocket
}
func (c *channelMap) addServer(id int64, s *server) {
c.mu.Lock()
s.cm = c
c.servers[id] = s
c.mu.Unlock()
}
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{}{}
} else {
c.findEntry(pid).addChild(id, cn)
}
c.mu.Unlock()
}
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()
}
func (c *channelMap) addListenSocket(id int64, ls *listenSocket, pid int64, ref string) {
c.mu.Lock()
ls.cm = c
c.listenSockets[id] = ls
c.findEntry(pid).addChild(id, ls)
c.mu.Unlock()
}
func (c *channelMap) addNormalSocket(id int64, ns *normalSocket, pid int64, ref string) {
c.mu.Lock()
ns.cm = c
c.normalSockets[id] = ns
c.findEntry(pid).addChild(id, ns)
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 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
var ok bool
if v, ok = c.channels[id]; ok {
return v
}
if v, ok = c.subChannels[id]; ok {
return v
}
if v, ok = c.servers[id]; ok {
return v
}
if v, ok = c.listenSockets[id]; ok {
return v
}
if v, ok = c.normalSockets[id]; ok {
return v
}
return &dummyEntry{idNotFound: id}
}
// c.mu must be held by the caller
// deleteEntry simply deletes an entry from the channelMap. Before calling this
// method, caller must check this entry is ready to be deleted, i.e removeEntry()
// has been called on it, and no children still exist.
// Conditionals are ordered by the expected frequency of deletion of each entity
// type, in order to optimize performance.
func (c *channelMap) deleteEntry(id int64) {
var ok bool
if _, ok = c.normalSockets[id]; ok {
delete(c.normalSockets, id)
return
}
if _, ok = c.subChannels[id]; ok {
delete(c.subChannels, id)
return
}
if _, ok = c.channels[id]; ok {
delete(c.channels, id)
delete(c.topLevelChannels, id)
return
}
if _, ok = c.listenSockets[id]; ok {
delete(c.listenSockets, id)
return
}
if _, ok = c.servers[id]; ok {
delete(c.servers, id)
return
}
}
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) }
func (s int64Slice) Swap(i, j int) { s[i], s[j] = s[j], s[i] }
func (s int64Slice) Less(i, j int) bool { return s[i] < s[j] }
func copyMap(m map[int64]string) map[int64]string {
n := make(map[int64]string)
for k, v := range m {
n[k] = v
}
return n
}
func min(a, b int64) int64 {
if a < b {
return a
}
return b
}
func (c *channelMap) GetTopChannels(id int64, maxResults int64) ([]*ChannelMetric, bool) {
if maxResults <= 0 {
maxResults = EntryPerPage
}
c.mu.RLock()
l := int64(len(c.topLevelChannels))
ids := make([]int64, 0, l)
cns := make([]*channel, 0, min(l, maxResults))
for k := range c.topLevelChannels {
ids = append(ids, k)
}
sort.Sort(int64Slice(ids))
idx := sort.Search(len(ids), func(i int) bool { return ids[i] >= id })
count := int64(0)
var end bool
var t []*ChannelMetric
for i, v := range ids[idx:] {
if count == maxResults {
break
}
if cn, ok := c.channels[v]; ok {
cns = append(cns, cn)
t = append(t, &ChannelMetric{
NestedChans: copyMap(cn.nestedChans),
SubChans: copyMap(cn.subChans),
})
count++
}
if i == len(ids[idx:])-1 {
end = true
break
}
}
c.mu.RUnlock()
if count == 0 {
end = true
}
for i, cn := range cns {
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
}
func (c *channelMap) GetServers(id, maxResults int64) ([]*ServerMetric, bool) {
if maxResults <= 0 {
maxResults = EntryPerPage
}
c.mu.RLock()
l := int64(len(c.servers))
ids := make([]int64, 0, l)
ss := make([]*server, 0, min(l, maxResults))
for k := range c.servers {
ids = append(ids, k)
}
sort.Sort(int64Slice(ids))
idx := sort.Search(len(ids), func(i int) bool { return ids[i] >= id })
count := int64(0)
var end bool
var s []*ServerMetric
for i, v := range ids[idx:] {
if count == maxResults {
break
}
if svr, ok := c.servers[v]; ok {
ss = append(ss, svr)
s = append(s, &ServerMetric{
ListenSockets: copyMap(svr.listenSockets),
})
count++
}
if i == len(ids[idx:])-1 {
end = true
break
}
}
c.mu.RUnlock()
if count == 0 {
end = true
}
for i, svr := range ss {
s[i].ServerData = svr.s.ChannelzMetric()
s[i].ID = svr.id
s[i].RefName = svr.refName
}
return s, end
}
func (c *channelMap) GetServerSockets(id int64, startID int64, maxResults int64) ([]*SocketMetric, bool) {
if maxResults <= 0 {
maxResults = EntryPerPage
}
var svr *server
var ok bool
c.mu.RLock()
if svr, ok = c.servers[id]; !ok {
// server with id doesn't exist.
c.mu.RUnlock()
return nil, true
}
svrskts := svr.sockets
l := int64(len(svrskts))
ids := make([]int64, 0, l)
sks := make([]*normalSocket, 0, min(l, maxResults))
for k := range svrskts {
ids = append(ids, k)
}
sort.Sort(int64Slice(ids))
idx := sort.Search(len(ids), func(i int) bool { return ids[i] >= startID })
count := int64(0)
var end bool
for i, v := range ids[idx:] {
if count == maxResults {
break
}
if ns, ok := c.normalSockets[v]; ok {
sks = append(sks, ns)
count++
}
if i == len(ids[idx:])-1 {
end = true
break
}
}
c.mu.RUnlock()
if count == 0 {
end = true
}
var s []*SocketMetric
for _, ns := range sks {
sm := &SocketMetric{}
sm.SocketData = ns.s.ChannelzMetric()
sm.ID = ns.id
sm.RefName = ns.refName
s = append(s, sm)
}
return s, end
}
func (c *channelMap) GetChannel(id int64) *ChannelMetric {
cm := &ChannelMetric{}
var cn *channel
var ok bool
c.mu.RLock()
if cn, ok = c.channels[id]; !ok {
// channel with id doesn't exist.
c.mu.RUnlock()
return nil
}
cm.NestedChans = copyMap(cn.nestedChans)
cm.SubChans = copyMap(cn.subChans)
// cn.c can be set to &dummyChannel{} when deleteSelfFromMap is called. Save a copy of cn.c when
// holding the lock to prevent potential data race.
chanCopy := cn.c
c.mu.RUnlock()
cm.ChannelData = chanCopy.ChannelzMetric()
cm.ID = cn.id
cm.RefName = cn.refName
cm.Trace = cn.trace.dumpData()
return cm
}
func (c *channelMap) GetSubChannel(id int64) *SubChannelMetric {
cm := &SubChannelMetric{}
var sc *subChannel
var ok bool
c.mu.RLock()
if sc, ok = c.subChannels[id]; !ok {
// subchannel with id doesn't exist.
c.mu.RUnlock()
return nil
}
cm.Sockets = copyMap(sc.sockets)
// sc.c can be set to &dummyChannel{} when deleteSelfFromMap is called. Save a copy of sc.c when
// holding the lock to prevent potential data race.
chanCopy := sc.c
c.mu.RUnlock()
cm.ChannelData = chanCopy.ChannelzMetric()
cm.ID = sc.id
cm.RefName = sc.refName
cm.Trace = sc.trace.dumpData()
return cm
}
func (c *channelMap) GetSocket(id int64) *SocketMetric {
sm := &SocketMetric{}
c.mu.RLock()
if ls, ok := c.listenSockets[id]; ok {
c.mu.RUnlock()
sm.SocketData = ls.s.ChannelzMetric()
sm.ID = ls.id
sm.RefName = ls.refName
return sm
}
if ns, ok := c.normalSockets[id]; ok {
c.mu.RUnlock()
sm.SocketData = ns.s.ChannelzMetric()
sm.ID = ns.id
sm.RefName = ns.refName
return sm
}
c.mu.RUnlock()
return nil
}
func (c *channelMap) GetServer(id int64) *ServerMetric {
sm := &ServerMetric{}
var svr *server
var ok bool
c.mu.RLock()
if svr, ok = c.servers[id]; !ok {
c.mu.RUnlock()
return nil
}
sm.ListenSockets = copyMap(svr.listenSockets)
c.mu.RUnlock()
sm.ID = svr.id
sm.RefName = svr.refName
sm.ServerData = svr.s.ChannelzMetric()
return sm
}
type idGenerator struct {
id int64
}
func (i *idGenerator) reset() {
atomic.StoreInt64(&i.id, 0)
}
func (i *idGenerator) genID() int64 {
return atomic.AddInt64(&i.id, 1)
}

View file

@ -0,0 +1,702 @@
/*
*
* Copyright 2018 gRPC authors.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*/
package channelz
import (
"net"
"sync"
"sync/atomic"
"time"
"google.golang.org/grpc/connectivity"
"google.golang.org/grpc/credentials"
"google.golang.org/grpc/grpclog"
)
// entry represents a node in the channelz database.
type entry interface {
// addChild adds a child e, whose channelz id is id to child list
addChild(id int64, e entry)
// deleteChild deletes a child with channelz id to be id from child list
deleteChild(id int64)
// triggerDelete tries to delete self from channelz database. However, if child
// list is not empty, then deletion from the database is on hold until the last
// child is deleted from database.
triggerDelete()
// 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.
type dummyEntry struct {
idNotFound int64
}
func (d *dummyEntry) addChild(id int64, e entry) {
// Note: It is possible for a normal program to reach here under race condition.
// For example, there could be a race between ClientConn.Close() info being propagated
// to addrConn and http2Client. ClientConn.Close() cancel the context and result
// in http2Client to error. The error info is then caught by transport monitor
// and before addrConn.tearDown() is called in side ClientConn.Close(). Therefore,
// the addrConn will create a new transport. And when registering the new transport in
// channelz, its parent addrConn could have already been torn down and deleted
// from channelz tracking, and thus reach the code here.
grpclog.Infof("attempt to add child of type %T with id %d to a parent (id=%d) that doesn't currently exist", e, id, d.idNotFound)
}
func (d *dummyEntry) deleteChild(id int64) {
// It is possible for a normal program to reach here under race condition.
// Refer to the example described in addChild().
grpclog.Infof("attempt to delete child with id %d from a parent (id=%d) that doesn't currently exist", id, d.idNotFound)
}
func (d *dummyEntry) triggerDelete() {
grpclog.Warningf("attempt to delete an entry (id=%d) that doesn't currently exist", d.idNotFound)
}
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.
type ChannelMetric struct {
// ID is the channelz id of this channel.
ID int64
// RefName is the human readable reference string of this channel.
RefName string
// ChannelData contains channel internal metric reported by the channel through
// ChannelzMetric().
ChannelData *ChannelInternalMetric
// NestedChans tracks the nested channel type children of this channel in the format of
// a map from nested channel channelz id to corresponding reference string.
NestedChans map[int64]string
// SubChans tracks the subchannel type children of this channel in the format of a
// map from subchannel channelz id to corresponding reference string.
SubChans map[int64]string
// Sockets tracks the socket type children of this channel in the format of a map
// from socket channelz id to corresponding reference string.
// 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,
// which includes ChannelInternalMetric and channelz-specific data, such as
// channelz id, child list, etc.
type SubChannelMetric struct {
// ID is the channelz id of this subchannel.
ID int64
// RefName is the human readable reference string of this subchannel.
RefName string
// ChannelData contains subchannel internal metric reported by the subchannel
// through ChannelzMetric().
ChannelData *ChannelInternalMetric
// NestedChans tracks the nested channel type children of this subchannel in the format of
// a map from nested channel channelz id to corresponding reference string.
// Note current grpc implementation doesn't allow subchannel to have nested channels
// as children, therefore, this field is unused.
NestedChans map[int64]string
// SubChans tracks the subchannel type children of this subchannel in the format of a
// map from subchannel channelz id to corresponding reference string.
// Note current grpc implementation doesn't allow subchannel to have subchannels
// as children, therefore, this field is unused.
SubChans map[int64]string
// 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
// should return from ChannelzMetric().
type ChannelInternalMetric struct {
// current connectivity state of the channel.
State connectivity.State
// The target this channel originally tried to connect to. May be absent
Target string
// The number of calls started on the channel.
CallsStarted int64
// The number of calls that have completed with an OK status.
CallsSucceeded int64
// The number of calls that have a completed with a non-OK status.
CallsFailed int64
// The last time a call was started on the channel.
LastCallStartedTimestamp time.Time
}
// 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
// channelz as Channel or SubChannel.
type Channel interface {
ChannelzMetric() *ChannelInternalMetric
}
type dummyChannel struct{}
func (d *dummyChannel) ChannelzMetric() *ChannelInternalMetric {
return &ChannelInternalMetric{}
}
type channel struct {
refName string
c Channel
closeCalled bool
nestedChans map[int64]string
subChans map[int64]string
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) {
switch v := e.(type) {
case *subChannel:
c.subChans[id] = v.refName
case *channel:
c.nestedChans[id] = v.refName
default:
grpclog.Errorf("cannot add a child (id = %d) of type %T to a channel", id, e)
}
}
func (c *channel) deleteChild(id int64) {
delete(c.subChans, id)
delete(c.nestedChans, id)
c.deleteSelfIfReady()
}
func (c *channel) triggerDelete() {
c.closeCalled = true
c.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 false
}
// 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
trace *channelTrace
traceRefCount int32
}
func (sc *subChannel) addChild(id int64, e entry) {
if v, ok := e.(*normalSocket); ok {
sc.sockets[id] = v.refName
} else {
grpclog.Errorf("cannot add a child (id = %d) of type %T to a subChannel", id, e)
}
}
func (sc *subChannel) deleteChild(id int64) {
delete(sc.sockets, id)
sc.deleteSelfIfReady()
}
func (sc *subChannel) triggerDelete() {
sc.closeCalled = true
sc.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.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
// includes SocketInternalMetric and channelz-specific data, such as channelz id, etc.
type SocketMetric struct {
// ID is the channelz id of this socket.
ID int64
// RefName is the human readable reference string of this socket.
RefName string
// SocketData contains socket internal metric reported by the socket through
// ChannelzMetric().
SocketData *SocketInternalMetric
}
// SocketInternalMetric defines the struct that the implementor of Socket interface
// should return from ChannelzMetric().
type SocketInternalMetric struct {
// The number of streams that have been started.
StreamsStarted int64
// The number of streams that have ended successfully:
// On client side, receiving frame with eos bit set.
// On server side, sending frame with eos bit set.
StreamsSucceeded int64
// The number of streams that have ended unsuccessfully:
// On client side, termination without receiving frame with eos bit set.
// On server side, termination without sending frame with eos bit set.
StreamsFailed int64
// The number of messages successfully sent on this socket.
MessagesSent int64
MessagesReceived int64
// The number of keep alives sent. This is typically implemented with HTTP/2
// ping messages.
KeepAlivesSent int64
// The last time a stream was created by this endpoint. Usually unset for
// servers.
LastLocalStreamCreatedTimestamp time.Time
// The last time a stream was created by the remote endpoint. Usually unset
// for clients.
LastRemoteStreamCreatedTimestamp time.Time
// The last time a message was sent by this endpoint.
LastMessageSentTimestamp time.Time
// The last time a message was received by this endpoint.
LastMessageReceivedTimestamp time.Time
// The amount of window, granted to the local endpoint by the remote endpoint.
// This may be slightly out of date due to network latency. This does NOT
// include stream level or TCP level flow control info.
LocalFlowControlWindow int64
// The amount of window, granted to the remote endpoint by the local endpoint.
// This may be slightly out of date due to network latency. This does NOT
// include stream level or TCP level flow control info.
RemoteFlowControlWindow int64
// The locally bound address.
LocalAddr net.Addr
// The remote bound address. May be absent.
RemoteAddr net.Addr
// Optional, represents the name of the remote endpoint, if different than
// the original target name.
RemoteName string
SocketOptions *SocketOptionData
Security credentials.ChannelzSecurityValue
}
// Socket is the interface that should be satisfied in order to be tracked by
// channelz as Socket.
type Socket interface {
ChannelzMetric() *SocketInternalMetric
}
type listenSocket struct {
refName string
s Socket
id int64
pid int64
cm *channelMap
}
func (ls *listenSocket) addChild(id int64, e entry) {
grpclog.Errorf("cannot add a child (id = %d) of type %T to a listen socket", id, e)
}
func (ls *listenSocket) deleteChild(id int64) {
grpclog.Errorf("cannot delete a child (id = %d) from a listen socket", id)
}
func (ls *listenSocket) triggerDelete() {
ls.cm.deleteEntry(ls.id)
ls.cm.findEntry(ls.pid).deleteChild(ls.id)
}
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
id int64
pid int64
cm *channelMap
}
func (ns *normalSocket) addChild(id int64, e entry) {
grpclog.Errorf("cannot add a child (id = %d) of type %T to a normal socket", id, e)
}
func (ns *normalSocket) deleteChild(id int64) {
grpclog.Errorf("cannot delete a child (id = %d) from a normal socket", id)
}
func (ns *normalSocket) triggerDelete() {
ns.cm.deleteEntry(ns.id)
ns.cm.findEntry(ns.pid).deleteChild(ns.id)
}
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.
type ServerMetric struct {
// ID is the channelz id of this server.
ID int64
// RefName is the human readable reference string of this server.
RefName string
// ServerData contains server internal metric reported by the server through
// ChannelzMetric().
ServerData *ServerInternalMetric
// ListenSockets tracks the listener socket type children of this server in the
// format of a map from socket channelz id to corresponding reference string.
ListenSockets map[int64]string
}
// ServerInternalMetric defines the struct that the implementor of Server interface
// should return from ChannelzMetric().
type ServerInternalMetric struct {
// The number of incoming calls started on the server.
CallsStarted int64
// The number of incoming calls that have completed with an OK status.
CallsSucceeded int64
// The number of incoming calls that have a completed with a non-OK status.
CallsFailed int64
// The last time a call was started on the server.
LastCallStartedTimestamp time.Time
}
// Server is the interface to be satisfied in order to be tracked by channelz as
// Server.
type Server interface {
ChannelzMetric() *ServerInternalMetric
}
type server struct {
refName string
s Server
closeCalled bool
sockets map[int64]string
listenSockets map[int64]string
id int64
cm *channelMap
}
func (s *server) addChild(id int64, e entry) {
switch v := e.(type) {
case *normalSocket:
s.sockets[id] = v.refName
case *listenSocket:
s.listenSockets[id] = v.refName
default:
grpclog.Errorf("cannot add a child (id = %d) of type %T to a server", id, e)
}
}
func (s *server) deleteChild(id int64) {
delete(s.sockets, id)
delete(s.listenSockets, id)
s.deleteSelfIfReady()
}
func (s *server) triggerDelete() {
s.closeCalled = true
s.deleteSelfIfReady()
}
func (s *server) deleteSelfIfReady() {
if !s.closeCalled || len(s.sockets)+len(s.listenSockets) != 0 {
return
}
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
}

View file

@ -0,0 +1,53 @@
// +build !appengine
/*
*
* Copyright 2018 gRPC authors.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*/
package channelz
import (
"syscall"
"golang.org/x/sys/unix"
)
// SocketOptionData defines the struct to hold socket option data, and related
// getter function to obtain info from fd.
type SocketOptionData struct {
Linger *unix.Linger
RecvTimeout *unix.Timeval
SendTimeout *unix.Timeval
TCPInfo *unix.TCPInfo
}
// Getsockopt defines the function to get socket options requested by channelz.
// It is to be passed to syscall.RawConn.Control().
func (s *SocketOptionData) Getsockopt(fd uintptr) {
if v, err := unix.GetsockoptLinger(int(fd), syscall.SOL_SOCKET, syscall.SO_LINGER); err == nil {
s.Linger = v
}
if v, err := unix.GetsockoptTimeval(int(fd), syscall.SOL_SOCKET, syscall.SO_RCVTIMEO); err == nil {
s.RecvTimeout = v
}
if v, err := unix.GetsockoptTimeval(int(fd), syscall.SOL_SOCKET, syscall.SO_SNDTIMEO); err == nil {
s.SendTimeout = v
}
if v, err := unix.GetsockoptTCPInfo(int(fd), syscall.SOL_TCP, syscall.TCP_INFO); err == nil {
s.TCPInfo = v
}
}

View file

@ -0,0 +1,44 @@
// +build !linux appengine
/*
*
* Copyright 2018 gRPC authors.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*/
package channelz
import (
"sync"
"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.
// Windows OS doesn't support Socket Option
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) {
once.Do(func() {
grpclog.Warningln("Channelz: socket options are not supported on non-linux os and appengine.")
})
}

View file

@ -0,0 +1,39 @@
// +build linux,!appengine
/*
*
* Copyright 2018 gRPC authors.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*/
package channelz
import (
"syscall"
)
// GetSocketOption gets the socket option info of the conn.
func GetSocketOption(socket interface{}) *SocketOptionData {
c, ok := socket.(syscall.Conn)
if !ok {
return nil
}
data := &SocketOptionData{}
if rawConn, err := c.SyscallConn(); err == nil {
rawConn.Control(data.Getsockopt)
return data
}
return nil
}

View file

@ -0,0 +1,26 @@
// +build !linux appengine
/*
*
* Copyright 2018 gRPC authors.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*/
package channelz
// GetSocketOption gets the socket option info of the conn.
func GetSocketOption(c interface{}) *SocketOptionData {
return nil
}

View file

@ -0,0 +1,64 @@
/*
*
* Copyright 2018 gRPC authors.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*/
// Package envconfig contains grpc settings configured by environment variables.
package envconfig
import (
"os"
"strings"
)
const (
prefix = "GRPC_GO_"
retryStr = prefix + "RETRY"
requireHandshakeStr = prefix + "REQUIRE_HANDSHAKE"
)
// RequireHandshakeSetting describes the settings for handshaking.
type RequireHandshakeSetting int
const (
// RequireHandshakeOn indicates to wait for handshake before considering a
// connection ready/successful.
RequireHandshakeOn RequireHandshakeSetting = iota
// RequireHandshakeOff indicates to not wait for handshake before
// considering a connection ready/successful.
RequireHandshakeOff
)
var (
// Retry is set if retry is explicitly enabled via "GRPC_GO_RETRY=on".
Retry = strings.EqualFold(os.Getenv(retryStr), "on")
// RequireHandshake is set based upon the GRPC_GO_REQUIRE_HANDSHAKE
// environment variable.
//
// Will be removed after the 1.18 release.
RequireHandshake = RequireHandshakeOn
)
func init() {
switch strings.ToLower(os.Getenv(requireHandshakeStr)) {
case "on":
fallthrough
default:
RequireHandshake = RequireHandshakeOn
case "off":
RequireHandshake = RequireHandshakeOff
}
}

View file

@ -0,0 +1,56 @@
/*
*
* Copyright 2018 gRPC authors.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*/
// Package grpcrand implements math/rand functions in a concurrent-safe way
// with a global random source, independent of math/rand's global source.
package grpcrand
import (
"math/rand"
"sync"
"time"
)
var (
r = rand.New(rand.NewSource(time.Now().UnixNano()))
mu sync.Mutex
)
// Int63n implements rand.Int63n on the grpcrand global source.
func Int63n(n int64) int64 {
mu.Lock()
res := r.Int63n(n)
mu.Unlock()
return res
}
// Intn implements rand.Intn on the grpcrand global source.
func Intn(n int) int {
mu.Lock()
res := r.Intn(n)
mu.Unlock()
return res
}
// Float64 implements rand.Float64 on the grpcrand global source.
func Float64() float64 {
mu.Lock()
res := r.Float64()
mu.Unlock()
return res
}

View file

@ -0,0 +1,61 @@
/*
*
* Copyright 2018 gRPC authors.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*/
// Package grpcsync implements additional synchronization primitives built upon
// the sync package.
package grpcsync
import (
"sync"
"sync/atomic"
)
// Event represents a one-time event that may occur in the future.
type Event struct {
fired int32
c chan struct{}
o sync.Once
}
// Fire causes e to complete. It is safe to call multiple times, and
// concurrently. It returns true iff this call to Fire caused the signaling
// channel returned by Done to close.
func (e *Event) Fire() bool {
ret := false
e.o.Do(func() {
atomic.StoreInt32(&e.fired, 1)
close(e.c)
ret = true
})
return ret
}
// Done returns a channel that will be closed when Fire is called.
func (e *Event) Done() <-chan struct{} {
return e.c
}
// HasFired returns true if Fire has been called.
func (e *Event) HasFired() bool {
return atomic.LoadInt32(&e.fired) == 1
}
// NewEvent returns a new, ready-to-use Event.
func NewEvent() *Event {
return &Event{c: make(chan struct{})}
}

View file

@ -15,13 +15,40 @@
*
*/
// Package internal contains gRPC-internal code for testing, to avoid polluting
// the godoc of the top-level grpc package.
// Package internal contains gRPC-internal code, to avoid polluting
// the godoc of the top-level grpc package. It must not import any grpc
// symbols to avoid circular dependencies.
package internal
// TestingUseHandlerImpl enables the http.Handler-based server implementation.
// It must be called before Serve and requires TLS credentials.
//
// The provided grpcServer must be of type *grpc.Server. It is untyped
// for circular dependency reasons.
var TestingUseHandlerImpl func(grpcServer interface{})
import (
"context"
"time"
)
var (
// WithResolverBuilder is exported by dialoptions.go
WithResolverBuilder interface{} // func (resolver.Builder) grpc.DialOption
// WithHealthCheckFunc is not exported by dialoptions.go
WithHealthCheckFunc interface{} // func (HealthChecker) DialOption
// HealthCheckFunc is used to provide client-side LB channel health checking
HealthCheckFunc HealthChecker
// BalancerUnregister is exported by package balancer to unregister a balancer.
BalancerUnregister func(name string)
// KeepaliveMinPingTime is the minimum ping interval. This must be 10s by
// default, but tests may wish to set it lower for convenience.
KeepaliveMinPingTime = 10 * time.Second
)
// HealthChecker defines the signature of the client-side LB channel health checking function.
type HealthChecker 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"
)

View file

@ -0,0 +1,114 @@
// +build !appengine
/*
*
* Copyright 2018 gRPC authors.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*/
// Package syscall provides functionalities that grpc uses to get low-level operating system
// stats/info.
package syscall
import (
"fmt"
"net"
"syscall"
"time"
"golang.org/x/sys/unix"
"google.golang.org/grpc/grpclog"
)
// GetCPUTime returns the how much CPU time has passed since the start of this process.
func GetCPUTime() int64 {
var ts unix.Timespec
if err := unix.ClockGettime(unix.CLOCK_PROCESS_CPUTIME_ID, &ts); err != nil {
grpclog.Fatal(err)
}
return ts.Nano()
}
// Rusage is an alias for syscall.Rusage under linux non-appengine environment.
type Rusage syscall.Rusage
// GetRusage returns the resource usage of current process.
func GetRusage() (rusage *Rusage) {
rusage = new(Rusage)
syscall.Getrusage(syscall.RUSAGE_SELF, (*syscall.Rusage)(rusage))
return
}
// CPUTimeDiff returns the differences of user CPU time and system CPU time used
// between two Rusage structs.
func CPUTimeDiff(first *Rusage, latest *Rusage) (float64, float64) {
f := (*syscall.Rusage)(first)
l := (*syscall.Rusage)(latest)
var (
utimeDiffs = l.Utime.Sec - f.Utime.Sec
utimeDiffus = l.Utime.Usec - f.Utime.Usec
stimeDiffs = l.Stime.Sec - f.Stime.Sec
stimeDiffus = l.Stime.Usec - f.Stime.Usec
)
uTimeElapsed := float64(utimeDiffs) + float64(utimeDiffus)*1.0e-6
sTimeElapsed := float64(stimeDiffs) + float64(stimeDiffus)*1.0e-6
return uTimeElapsed, sTimeElapsed
}
// SetTCPUserTimeout sets the TCP user timeout on a connection's socket
func SetTCPUserTimeout(conn net.Conn, timeout time.Duration) error {
tcpconn, ok := conn.(*net.TCPConn)
if !ok {
// not a TCP connection. exit early
return nil
}
rawConn, err := tcpconn.SyscallConn()
if err != nil {
return fmt.Errorf("error getting raw connection: %v", err)
}
err = rawConn.Control(func(fd uintptr) {
err = syscall.SetsockoptInt(int(fd), syscall.IPPROTO_TCP, unix.TCP_USER_TIMEOUT, int(timeout/time.Millisecond))
})
if err != nil {
return fmt.Errorf("error setting option on socket: %v", err)
}
return nil
}
// GetTCPUserTimeout gets the TCP user timeout on a connection's socket
func GetTCPUserTimeout(conn net.Conn) (opt int, err error) {
tcpconn, ok := conn.(*net.TCPConn)
if !ok {
err = fmt.Errorf("conn is not *net.TCPConn. got %T", conn)
return
}
rawConn, err := tcpconn.SyscallConn()
if err != nil {
err = fmt.Errorf("error getting raw connection: %v", err)
return
}
err = rawConn.Control(func(fd uintptr) {
opt, err = syscall.GetsockoptInt(int(fd), syscall.IPPROTO_TCP, unix.TCP_USER_TIMEOUT)
})
if err != nil {
err = fmt.Errorf("error getting option on socket: %v", err)
return
}
return
}

View file

@ -0,0 +1,73 @@
// +build !linux appengine
/*
*
* Copyright 2018 gRPC authors.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*/
package syscall
import (
"net"
"sync"
"time"
"google.golang.org/grpc/grpclog"
)
var once sync.Once
func log() {
once.Do(func() {
grpclog.Info("CPU time info is unavailable on non-linux or appengine environment.")
})
}
// GetCPUTime returns the how much CPU time has passed since the start of this process.
// It always returns 0 under non-linux or appengine environment.
func GetCPUTime() int64 {
log()
return 0
}
// Rusage is an empty struct under non-linux or appengine environment.
type Rusage struct{}
// GetRusage is a no-op function under non-linux or appengine environment.
func GetRusage() (rusage *Rusage) {
log()
return nil
}
// CPUTimeDiff returns the differences of user CPU time and system CPU time used
// between two Rusage structs. It a no-op function for non-linux or appengine environment.
func CPUTimeDiff(first *Rusage, latest *Rusage) (float64, float64) {
log()
return 0, 0
}
// SetTCPUserTimeout is a no-op function under non-linux or appengine environments
func SetTCPUserTimeout(conn net.Conn, timeout time.Duration) error {
log()
return nil
}
// GetTCPUserTimeout is a no-op function under non-linux or appengine environments
// a negative return value indicates the operation is not supported
func GetTCPUserTimeout(conn net.Conn) (int, error) {
log()
return -1, nil
}

View file

@ -0,0 +1,141 @@
/*
*
* Copyright 2017 gRPC authors.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*/
package transport
import (
"sync"
"time"
)
const (
// bdpLimit is the maximum value the flow control windows will be increased
// to. TCP typically limits this to 4MB, but some systems go up to 16MB.
// Since this is only a limit, it is safe to make it optimistic.
bdpLimit = (1 << 20) * 16
// alpha is a constant factor used to keep a moving average
// of RTTs.
alpha = 0.9
// If the current bdp sample is greater than or equal to
// our beta * our estimated bdp and the current bandwidth
// sample is the maximum bandwidth observed so far, we
// increase our bbp estimate by a factor of gamma.
beta = 0.66
// To put our bdp to be smaller than or equal to twice the real BDP,
// we should multiply our current sample with 4/3, however to round things out
// we use 2 as the multiplication factor.
gamma = 2
)
// Adding arbitrary data to ping so that its ack can be identified.
// Easter-egg: what does the ping message say?
var bdpPing = &ping{data: [8]byte{2, 4, 16, 16, 9, 14, 7, 7}}
type bdpEstimator struct {
// sentAt is the time when the ping was sent.
sentAt time.Time
mu sync.Mutex
// bdp is the current bdp estimate.
bdp uint32
// sample is the number of bytes received in one measurement cycle.
sample uint32
// bwMax is the maximum bandwidth noted so far (bytes/sec).
bwMax float64
// bool to keep track of the beginning of a new measurement cycle.
isSent bool
// Callback to update the window sizes.
updateFlowControl func(n uint32)
// sampleCount is the number of samples taken so far.
sampleCount uint64
// round trip time (seconds)
rtt float64
}
// timesnap registers the time bdp ping was sent out so that
// network rtt can be calculated when its ack is received.
// It is called (by controller) when the bdpPing is
// being written on the wire.
func (b *bdpEstimator) timesnap(d [8]byte) {
if bdpPing.data != d {
return
}
b.sentAt = time.Now()
}
// add adds bytes to the current sample for calculating bdp.
// It returns true only if a ping must be sent. This can be used
// by the caller (handleData) to make decision about batching
// a window update with it.
func (b *bdpEstimator) add(n uint32) bool {
b.mu.Lock()
defer b.mu.Unlock()
if b.bdp == bdpLimit {
return false
}
if !b.isSent {
b.isSent = true
b.sample = n
b.sentAt = time.Time{}
b.sampleCount++
return true
}
b.sample += n
return false
}
// calculate is called when an ack for a bdp ping is received.
// Here we calculate the current bdp and bandwidth sample and
// decide if the flow control windows should go up.
func (b *bdpEstimator) calculate(d [8]byte) {
// Check if the ping acked for was the bdp ping.
if bdpPing.data != d {
return
}
b.mu.Lock()
rttSample := time.Since(b.sentAt).Seconds()
if b.sampleCount < 10 {
// Bootstrap rtt with an average of first 10 rtt samples.
b.rtt += (rttSample - b.rtt) / float64(b.sampleCount)
} else {
// Heed to the recent past more.
b.rtt += (rttSample - b.rtt) * float64(alpha)
}
b.isSent = false
// The number of bytes accumulated so far in the sample is smaller
// than or equal to 1.5 times the real BDP on a saturated connection.
bwCurrent := float64(b.sample) / (b.rtt * float64(1.5))
if bwCurrent > b.bwMax {
b.bwMax = bwCurrent
}
// If the current sample (which is smaller than or equal to the 1.5 times the real BDP) is
// greater than or equal to 2/3rd our perceived bdp AND this is the maximum bandwidth seen so far, we
// should update our perception of the network BDP.
if float64(b.sample) >= beta*float64(b.bdp) && bwCurrent == b.bwMax && b.bdp != bdpLimit {
sampleFloat := float64(b.sample)
b.bdp = uint32(gamma * sampleFloat)
if b.bdp > bdpLimit {
b.bdp = bdpLimit
}
bdp := b.bdp
b.mu.Unlock()
b.updateFlowControl(bdp)
return
}
b.mu.Unlock()
}

View file

@ -0,0 +1,852 @@
/*
*
* Copyright 2014 gRPC authors.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*/
package transport
import (
"bytes"
"fmt"
"runtime"
"sync"
"golang.org/x/net/http2"
"golang.org/x/net/http2/hpack"
)
var updateHeaderTblSize = func(e *hpack.Encoder, v uint32) {
e.SetMaxDynamicTableSizeLimit(v)
}
type itemNode struct {
it interface{}
next *itemNode
}
type itemList struct {
head *itemNode
tail *itemNode
}
func (il *itemList) enqueue(i interface{}) {
n := &itemNode{it: i}
if il.tail == nil {
il.head, il.tail = n, n
return
}
il.tail.next = n
il.tail = n
}
// peek returns the first item in the list without removing it from the
// list.
func (il *itemList) peek() interface{} {
return il.head.it
}
func (il *itemList) dequeue() interface{} {
if il.head == nil {
return nil
}
i := il.head.it
il.head = il.head.next
if il.head == nil {
il.tail = nil
}
return i
}
func (il *itemList) dequeueAll() *itemNode {
h := il.head
il.head, il.tail = nil, nil
return h
}
func (il *itemList) isEmpty() bool {
return il.head == nil
}
// The following defines various control items which could flow through
// the control buffer of transport. They represent different aspects of
// control tasks, e.g., flow control, settings, streaming resetting, etc.
// registerStream is used to register an incoming stream with loopy writer.
type registerStream struct {
streamID uint32
wq *writeQuota
}
// headerFrame is also used to register stream on the client-side.
type headerFrame struct {
streamID uint32
hf []hpack.HeaderField
endStream bool // Valid on server side.
initStream func(uint32) (bool, error) // Used only on the client side.
onWrite func()
wq *writeQuota // write quota for the stream created.
cleanup *cleanupStream // Valid on the server side.
onOrphaned func(error) // Valid on client-side
}
type cleanupStream struct {
streamID uint32
rst bool
rstCode http2.ErrCode
onWrite func()
}
type dataFrame struct {
streamID uint32
endStream bool
h []byte
d []byte
// onEachWrite is called every time
// a part of d is written out.
onEachWrite func()
}
type incomingWindowUpdate struct {
streamID uint32
increment uint32
}
type outgoingWindowUpdate struct {
streamID uint32
increment uint32
}
type incomingSettings struct {
ss []http2.Setting
}
type outgoingSettings struct {
ss []http2.Setting
}
type incomingGoAway struct {
}
type goAway struct {
code http2.ErrCode
debugData []byte
headsUp bool
closeConn bool
}
type ping struct {
ack bool
data [8]byte
}
type outFlowControlSizeRequest struct {
resp chan uint32
}
type outStreamState int
const (
active outStreamState = iota
empty
waitingOnStreamQuota
)
type outStream struct {
id uint32
state outStreamState
itl *itemList
bytesOutStanding int
wq *writeQuota
next *outStream
prev *outStream
}
func (s *outStream) deleteSelf() {
if s.prev != nil {
s.prev.next = s.next
}
if s.next != nil {
s.next.prev = s.prev
}
s.next, s.prev = nil, nil
}
type outStreamList struct {
// Following are sentinel objects that mark the
// beginning and end of the list. They do not
// contain any item lists. All valid objects are
// inserted in between them.
// This is needed so that an outStream object can
// deleteSelf() in O(1) time without knowing which
// list it belongs to.
head *outStream
tail *outStream
}
func newOutStreamList() *outStreamList {
head, tail := new(outStream), new(outStream)
head.next = tail
tail.prev = head
return &outStreamList{
head: head,
tail: tail,
}
}
func (l *outStreamList) enqueue(s *outStream) {
e := l.tail.prev
e.next = s
s.prev = e
s.next = l.tail
l.tail.prev = s
}
// remove from the beginning of the list.
func (l *outStreamList) dequeue() *outStream {
b := l.head.next
if b == l.tail {
return nil
}
b.deleteSelf()
return b
}
// controlBuffer is a way to pass information to loopy.
// Information is passed as specific struct types called control frames.
// A control frame not only represents data, messages or headers to be sent out
// but can also be used to instruct loopy to update its internal state.
// It shouldn't be confused with an HTTP2 frame, although some of the control frames
// like dataFrame and headerFrame do go out on wire as HTTP2 frames.
type controlBuffer struct {
ch chan struct{}
done <-chan struct{}
mu sync.Mutex
consumerWaiting bool
list *itemList
err error
}
func newControlBuffer(done <-chan struct{}) *controlBuffer {
return &controlBuffer{
ch: make(chan struct{}, 1),
list: &itemList{},
done: done,
}
}
func (c *controlBuffer) put(it interface{}) error {
_, err := c.executeAndPut(nil, it)
return err
}
func (c *controlBuffer) executeAndPut(f func(it interface{}) bool, it interface{}) (bool, error) {
var wakeUp bool
c.mu.Lock()
if c.err != nil {
c.mu.Unlock()
return false, c.err
}
if f != nil {
if !f(it) { // f wasn't successful
c.mu.Unlock()
return false, nil
}
}
if c.consumerWaiting {
wakeUp = true
c.consumerWaiting = false
}
c.list.enqueue(it)
c.mu.Unlock()
if wakeUp {
select {
case c.ch <- struct{}{}:
default:
}
}
return true, nil
}
// Note argument f should never be nil.
func (c *controlBuffer) execute(f func(it interface{}) bool, it interface{}) (bool, error) {
c.mu.Lock()
if c.err != nil {
c.mu.Unlock()
return false, c.err
}
if !f(it) { // f wasn't successful
c.mu.Unlock()
return false, nil
}
c.mu.Unlock()
return true, nil
}
func (c *controlBuffer) get(block bool) (interface{}, error) {
for {
c.mu.Lock()
if c.err != nil {
c.mu.Unlock()
return nil, c.err
}
if !c.list.isEmpty() {
h := c.list.dequeue()
c.mu.Unlock()
return h, nil
}
if !block {
c.mu.Unlock()
return nil, nil
}
c.consumerWaiting = true
c.mu.Unlock()
select {
case <-c.ch:
case <-c.done:
c.finish()
return nil, ErrConnClosing
}
}
}
func (c *controlBuffer) finish() {
c.mu.Lock()
if c.err != nil {
c.mu.Unlock()
return
}
c.err = ErrConnClosing
// There may be headers for streams in the control buffer.
// These streams need to be cleaned out since the transport
// is still not aware of these yet.
for head := c.list.dequeueAll(); head != nil; head = head.next {
hdr, ok := head.it.(*headerFrame)
if !ok {
continue
}
if hdr.onOrphaned != nil { // It will be nil on the server-side.
hdr.onOrphaned(ErrConnClosing)
}
}
c.mu.Unlock()
}
type side int
const (
clientSide side = iota
serverSide
)
// Loopy receives frames from the control buffer.
// Each frame is handled individually; most of the work done by loopy goes
// into handling data frames. Loopy maintains a queue of active streams, and each
// stream maintains a queue of data frames; as loopy receives data frames
// it gets added to the queue of the relevant stream.
// Loopy goes over this list of active streams by processing one node every iteration,
// thereby closely resemebling to a round-robin scheduling over all streams. While
// processing a stream, loopy writes out data bytes from this stream capped by the min
// of http2MaxFrameLen, connection-level flow control and stream-level flow control.
type loopyWriter struct {
side side
cbuf *controlBuffer
sendQuota uint32
oiws uint32 // outbound initial window size.
// estdStreams is map of all established streams that are not cleaned-up yet.
// On client-side, this is all streams whose headers were sent out.
// On server-side, this is all streams whose headers were received.
estdStreams map[uint32]*outStream // Established streams.
// activeStreams is a linked-list of all streams that have data to send and some
// stream-level flow control quota.
// Each of these streams internally have a list of data items(and perhaps trailers
// on the server-side) to be sent out.
activeStreams *outStreamList
framer *framer
hBuf *bytes.Buffer // The buffer for HPACK encoding.
hEnc *hpack.Encoder // HPACK encoder.
bdpEst *bdpEstimator
draining bool
// Side-specific handlers
ssGoAwayHandler func(*goAway) (bool, error)
}
func newLoopyWriter(s side, fr *framer, cbuf *controlBuffer, bdpEst *bdpEstimator) *loopyWriter {
var buf bytes.Buffer
l := &loopyWriter{
side: s,
cbuf: cbuf,
sendQuota: defaultWindowSize,
oiws: defaultWindowSize,
estdStreams: make(map[uint32]*outStream),
activeStreams: newOutStreamList(),
framer: fr,
hBuf: &buf,
hEnc: hpack.NewEncoder(&buf),
bdpEst: bdpEst,
}
return l
}
const minBatchSize = 1000
// run should be run in a separate goroutine.
// It reads control frames from controlBuf and processes them by:
// 1. Updating loopy's internal state, or/and
// 2. Writing out HTTP2 frames on the wire.
//
// Loopy keeps all active streams with data to send in a linked-list.
// All streams in the activeStreams linked-list must have both:
// 1. Data to send, and
// 2. Stream level flow control quota available.
//
// In each iteration of run loop, other than processing the incoming control
// frame, loopy calls processData, which processes one node from the activeStreams linked-list.
// This results in writing of HTTP2 frames into an underlying write buffer.
// When there's no more control frames to read from controlBuf, loopy flushes the write buffer.
// As an optimization, to increase the batch size for each flush, loopy yields the processor, once
// if the batch size is too low to give stream goroutines a chance to fill it up.
func (l *loopyWriter) run() (err error) {
defer func() {
if err == ErrConnClosing {
// Don't log ErrConnClosing as error since it happens
// 1. When the connection is closed by some other known issue.
// 2. User closed the connection.
// 3. A graceful close of connection.
infof("transport: loopyWriter.run returning. %v", err)
err = nil
}
}()
for {
it, err := l.cbuf.get(true)
if err != nil {
return err
}
if err = l.handle(it); err != nil {
return err
}
if _, err = l.processData(); err != nil {
return err
}
gosched := true
hasdata:
for {
it, err := l.cbuf.get(false)
if err != nil {
return err
}
if it != nil {
if err = l.handle(it); err != nil {
return err
}
if _, err = l.processData(); err != nil {
return err
}
continue hasdata
}
isEmpty, err := l.processData()
if err != nil {
return err
}
if !isEmpty {
continue hasdata
}
if gosched {
gosched = false
if l.framer.writer.offset < minBatchSize {
runtime.Gosched()
continue hasdata
}
}
l.framer.writer.Flush()
break hasdata
}
}
}
func (l *loopyWriter) outgoingWindowUpdateHandler(w *outgoingWindowUpdate) error {
return l.framer.fr.WriteWindowUpdate(w.streamID, w.increment)
}
func (l *loopyWriter) incomingWindowUpdateHandler(w *incomingWindowUpdate) error {
// Otherwise update the quota.
if w.streamID == 0 {
l.sendQuota += w.increment
return nil
}
// Find the stream and update it.
if str, ok := l.estdStreams[w.streamID]; ok {
str.bytesOutStanding -= int(w.increment)
if strQuota := int(l.oiws) - str.bytesOutStanding; strQuota > 0 && str.state == waitingOnStreamQuota {
str.state = active
l.activeStreams.enqueue(str)
return nil
}
}
return nil
}
func (l *loopyWriter) outgoingSettingsHandler(s *outgoingSettings) error {
return l.framer.fr.WriteSettings(s.ss...)
}
func (l *loopyWriter) incomingSettingsHandler(s *incomingSettings) error {
if err := l.applySettings(s.ss); err != nil {
return err
}
return l.framer.fr.WriteSettingsAck()
}
func (l *loopyWriter) registerStreamHandler(h *registerStream) error {
str := &outStream{
id: h.streamID,
state: empty,
itl: &itemList{},
wq: h.wq,
}
l.estdStreams[h.streamID] = str
return nil
}
func (l *loopyWriter) headerHandler(h *headerFrame) error {
if l.side == serverSide {
str, ok := l.estdStreams[h.streamID]
if !ok {
warningf("transport: loopy doesn't recognize the stream: %d", h.streamID)
return nil
}
// Case 1.A: Server is responding back with headers.
if !h.endStream {
return l.writeHeader(h.streamID, h.endStream, h.hf, h.onWrite)
}
// else: Case 1.B: Server wants to close stream.
if str.state != empty { // either active or waiting on stream quota.
// add it str's list of items.
str.itl.enqueue(h)
return nil
}
if err := l.writeHeader(h.streamID, h.endStream, h.hf, h.onWrite); err != nil {
return err
}
return l.cleanupStreamHandler(h.cleanup)
}
// Case 2: Client wants to originate stream.
str := &outStream{
id: h.streamID,
state: empty,
itl: &itemList{},
wq: h.wq,
}
str.itl.enqueue(h)
return l.originateStream(str)
}
func (l *loopyWriter) originateStream(str *outStream) error {
hdr := str.itl.dequeue().(*headerFrame)
sendPing, err := hdr.initStream(str.id)
if err != nil {
if err == ErrConnClosing {
return err
}
// Other errors(errStreamDrain) need not close transport.
return nil
}
if err = l.writeHeader(str.id, hdr.endStream, hdr.hf, hdr.onWrite); err != nil {
return err
}
l.estdStreams[str.id] = str
if sendPing {
return l.pingHandler(&ping{data: [8]byte{}})
}
return nil
}
func (l *loopyWriter) writeHeader(streamID uint32, endStream bool, hf []hpack.HeaderField, onWrite func()) error {
if onWrite != nil {
onWrite()
}
l.hBuf.Reset()
for _, f := range hf {
if err := l.hEnc.WriteField(f); err != nil {
warningf("transport: loopyWriter.writeHeader encountered error while encoding headers:", err)
}
}
var (
err error
endHeaders, first bool
)
first = true
for !endHeaders {
size := l.hBuf.Len()
if size > http2MaxFrameLen {
size = http2MaxFrameLen
} else {
endHeaders = true
}
if first {
first = false
err = l.framer.fr.WriteHeaders(http2.HeadersFrameParam{
StreamID: streamID,
BlockFragment: l.hBuf.Next(size),
EndStream: endStream,
EndHeaders: endHeaders,
})
} else {
err = l.framer.fr.WriteContinuation(
streamID,
endHeaders,
l.hBuf.Next(size),
)
}
if err != nil {
return err
}
}
return nil
}
func (l *loopyWriter) preprocessData(df *dataFrame) error {
str, ok := l.estdStreams[df.streamID]
if !ok {
return nil
}
// If we got data for a stream it means that
// stream was originated and the headers were sent out.
str.itl.enqueue(df)
if str.state == empty {
str.state = active
l.activeStreams.enqueue(str)
}
return nil
}
func (l *loopyWriter) pingHandler(p *ping) error {
if !p.ack {
l.bdpEst.timesnap(p.data)
}
return l.framer.fr.WritePing(p.ack, p.data)
}
func (l *loopyWriter) outFlowControlSizeRequestHandler(o *outFlowControlSizeRequest) error {
o.resp <- l.sendQuota
return nil
}
func (l *loopyWriter) cleanupStreamHandler(c *cleanupStream) error {
c.onWrite()
if str, ok := l.estdStreams[c.streamID]; ok {
// On the server side it could be a trailers-only response or
// a RST_STREAM before stream initialization thus the stream might
// not be established yet.
delete(l.estdStreams, c.streamID)
str.deleteSelf()
}
if c.rst { // If RST_STREAM needs to be sent.
if err := l.framer.fr.WriteRSTStream(c.streamID, c.rstCode); err != nil {
return err
}
}
if l.side == clientSide && l.draining && len(l.estdStreams) == 0 {
return ErrConnClosing
}
return nil
}
func (l *loopyWriter) incomingGoAwayHandler(*incomingGoAway) error {
if l.side == clientSide {
l.draining = true
if len(l.estdStreams) == 0 {
return ErrConnClosing
}
}
return nil
}
func (l *loopyWriter) goAwayHandler(g *goAway) error {
// Handling of outgoing GoAway is very specific to side.
if l.ssGoAwayHandler != nil {
draining, err := l.ssGoAwayHandler(g)
if err != nil {
return err
}
l.draining = draining
}
return nil
}
func (l *loopyWriter) handle(i interface{}) error {
switch i := i.(type) {
case *incomingWindowUpdate:
return l.incomingWindowUpdateHandler(i)
case *outgoingWindowUpdate:
return l.outgoingWindowUpdateHandler(i)
case *incomingSettings:
return l.incomingSettingsHandler(i)
case *outgoingSettings:
return l.outgoingSettingsHandler(i)
case *headerFrame:
return l.headerHandler(i)
case *registerStream:
return l.registerStreamHandler(i)
case *cleanupStream:
return l.cleanupStreamHandler(i)
case *incomingGoAway:
return l.incomingGoAwayHandler(i)
case *dataFrame:
return l.preprocessData(i)
case *ping:
return l.pingHandler(i)
case *goAway:
return l.goAwayHandler(i)
case *outFlowControlSizeRequest:
return l.outFlowControlSizeRequestHandler(i)
default:
return fmt.Errorf("transport: unknown control message type %T", i)
}
}
func (l *loopyWriter) applySettings(ss []http2.Setting) error {
for _, s := range ss {
switch s.ID {
case http2.SettingInitialWindowSize:
o := l.oiws
l.oiws = s.Val
if o < l.oiws {
// If the new limit is greater make all depleted streams active.
for _, stream := range l.estdStreams {
if stream.state == waitingOnStreamQuota {
stream.state = active
l.activeStreams.enqueue(stream)
}
}
}
case http2.SettingHeaderTableSize:
updateHeaderTblSize(l.hEnc, s.Val)
}
}
return nil
}
// processData removes the first stream from active streams, writes out at most 16KB
// of its data and then puts it at the end of activeStreams if there's still more data
// to be sent and stream has some stream-level flow control.
func (l *loopyWriter) processData() (bool, error) {
if l.sendQuota == 0 {
return true, nil
}
str := l.activeStreams.dequeue() // Remove the first stream.
if str == nil {
return true, nil
}
dataItem := str.itl.peek().(*dataFrame) // Peek at the first data item this stream.
// A data item is represented by a dataFrame, since it later translates into
// multiple HTTP2 data frames.
// Every dataFrame has two buffers; h that keeps grpc-message header and d that is acutal data.
// As an optimization to keep wire traffic low, data from d is copied to h to make as big as the
// maximum possilbe HTTP2 frame size.
if len(dataItem.h) == 0 && len(dataItem.d) == 0 { // Empty data frame
// Client sends out empty data frame with endStream = true
if err := l.framer.fr.WriteData(dataItem.streamID, dataItem.endStream, nil); err != nil {
return false, err
}
str.itl.dequeue() // remove the empty data item from stream
if str.itl.isEmpty() {
str.state = empty
} else if trailer, ok := str.itl.peek().(*headerFrame); ok { // the next item is trailers.
if err := l.writeHeader(trailer.streamID, trailer.endStream, trailer.hf, trailer.onWrite); err != nil {
return false, err
}
if err := l.cleanupStreamHandler(trailer.cleanup); err != nil {
return false, nil
}
} else {
l.activeStreams.enqueue(str)
}
return false, nil
}
var (
idx int
buf []byte
)
if len(dataItem.h) != 0 { // data header has not been written out yet.
buf = dataItem.h
} else {
idx = 1
buf = dataItem.d
}
size := http2MaxFrameLen
if len(buf) < size {
size = len(buf)
}
if strQuota := int(l.oiws) - str.bytesOutStanding; strQuota <= 0 { // stream-level flow control.
str.state = waitingOnStreamQuota
return false, nil
} else if strQuota < size {
size = strQuota
}
if l.sendQuota < uint32(size) { // connection-level flow control.
size = int(l.sendQuota)
}
// Now that outgoing flow controls are checked we can replenish str's write quota
str.wq.replenish(size)
var endStream bool
// If this is the last data message on this stream and all of it can be written in this iteration.
if dataItem.endStream && size == len(buf) {
// buf contains either data or it contains header but data is empty.
if idx == 1 || len(dataItem.d) == 0 {
endStream = true
}
}
if dataItem.onEachWrite != nil {
dataItem.onEachWrite()
}
if err := l.framer.fr.WriteData(dataItem.streamID, endStream, buf[:size]); err != nil {
return false, err
}
buf = buf[size:]
str.bytesOutStanding += size
l.sendQuota -= uint32(size)
if idx == 0 {
dataItem.h = buf
} else {
dataItem.d = buf
}
if len(dataItem.h) == 0 && len(dataItem.d) == 0 { // All the data from that message was written out.
str.itl.dequeue()
}
if str.itl.isEmpty() {
str.state = empty
} else if trailer, ok := str.itl.peek().(*headerFrame); ok { // The next item is trailers.
if err := l.writeHeader(trailer.streamID, trailer.endStream, trailer.hf, trailer.onWrite); err != nil {
return false, err
}
if err := l.cleanupStreamHandler(trailer.cleanup); err != nil {
return false, err
}
} else if int(l.oiws)-str.bytesOutStanding <= 0 { // Ran out of stream quota.
str.state = waitingOnStreamQuota
} else { // Otherwise add it back to the list of active streams.
l.activeStreams.enqueue(str)
}
return false, nil
}

View file

@ -0,0 +1,49 @@
/*
*
* Copyright 2018 gRPC authors.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*/
package transport
import (
"math"
"time"
)
const (
// The default value of flow control window size in HTTP2 spec.
defaultWindowSize = 65535
// The initial window size for flow control.
initialWindowSize = defaultWindowSize // for an RPC
infinity = time.Duration(math.MaxInt64)
defaultClientKeepaliveTime = infinity
defaultClientKeepaliveTimeout = 20 * time.Second
defaultMaxStreamsClient = 100
defaultMaxConnectionIdle = infinity
defaultMaxConnectionAge = infinity
defaultMaxConnectionAgeGrace = infinity
defaultServerKeepaliveTime = 2 * time.Hour
defaultServerKeepaliveTimeout = 20 * time.Second
defaultKeepalivePolicyMinTime = 5 * time.Minute
// max window limit set by HTTP2 Specs.
maxWindowSize = math.MaxInt32
// defaultWriteQuota is the default value for number of data
// bytes that each stream can schedule before some of it being
// flushed out.
defaultWriteQuota = 64 * 1024
defaultClientMaxHeaderListSize = uint32(16 << 20)
defaultServerMaxHeaderListSize = uint32(16 << 20)
)

View file

@ -0,0 +1,218 @@
/*
*
* Copyright 2014 gRPC authors.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*/
package transport
import (
"fmt"
"math"
"sync"
"sync/atomic"
)
// writeQuota is a soft limit on the amount of data a stream can
// schedule before some of it is written out.
type writeQuota struct {
quota int32
// get waits on read from when quota goes less than or equal to zero.
// replenish writes on it when quota goes positive again.
ch chan struct{}
// done is triggered in error case.
done <-chan struct{}
// replenish is called by loopyWriter to give quota back to.
// It is implemented as a field so that it can be updated
// by tests.
replenish func(n int)
}
func newWriteQuota(sz int32, done <-chan struct{}) *writeQuota {
w := &writeQuota{
quota: sz,
ch: make(chan struct{}, 1),
done: done,
}
w.replenish = w.realReplenish
return w
}
func (w *writeQuota) get(sz int32) error {
for {
if atomic.LoadInt32(&w.quota) > 0 {
atomic.AddInt32(&w.quota, -sz)
return nil
}
select {
case <-w.ch:
continue
case <-w.done:
return errStreamDone
}
}
}
func (w *writeQuota) realReplenish(n int) {
sz := int32(n)
a := atomic.AddInt32(&w.quota, sz)
b := a - sz
if b <= 0 && a > 0 {
select {
case w.ch <- struct{}{}:
default:
}
}
}
type trInFlow struct {
limit uint32
unacked uint32
effectiveWindowSize uint32
}
func (f *trInFlow) newLimit(n uint32) uint32 {
d := n - f.limit
f.limit = n
f.updateEffectiveWindowSize()
return d
}
func (f *trInFlow) onData(n uint32) uint32 {
f.unacked += n
if f.unacked >= f.limit/4 {
w := f.unacked
f.unacked = 0
f.updateEffectiveWindowSize()
return w
}
f.updateEffectiveWindowSize()
return 0
}
func (f *trInFlow) reset() uint32 {
w := f.unacked
f.unacked = 0
f.updateEffectiveWindowSize()
return w
}
func (f *trInFlow) updateEffectiveWindowSize() {
atomic.StoreUint32(&f.effectiveWindowSize, f.limit-f.unacked)
}
func (f *trInFlow) getSize() uint32 {
return atomic.LoadUint32(&f.effectiveWindowSize)
}
// TODO(mmukhi): Simplify this code.
// inFlow deals with inbound flow control
type inFlow struct {
mu sync.Mutex
// The inbound flow control limit for pending data.
limit uint32
// pendingData is the overall data which have been received but not been
// consumed by applications.
pendingData uint32
// The amount of data the application has consumed but grpc has not sent
// window update for them. Used to reduce window update frequency.
pendingUpdate uint32
// delta is the extra window update given by receiver when an application
// is reading data bigger in size than the inFlow limit.
delta uint32
}
// newLimit updates the inflow window to a new value n.
// It assumes that n is always greater than the old limit.
func (f *inFlow) newLimit(n uint32) uint32 {
f.mu.Lock()
d := n - f.limit
f.limit = n
f.mu.Unlock()
return d
}
func (f *inFlow) maybeAdjust(n uint32) uint32 {
if n > uint32(math.MaxInt32) {
n = uint32(math.MaxInt32)
}
f.mu.Lock()
// estSenderQuota is the receiver's view of the maximum number of bytes the sender
// can send without a window update.
estSenderQuota := int32(f.limit - (f.pendingData + f.pendingUpdate))
// estUntransmittedData is the maximum number of bytes the sends might not have put
// on the wire yet. A value of 0 or less means that we have already received all or
// more bytes than the application is requesting to read.
estUntransmittedData := int32(n - f.pendingData) // Casting into int32 since it could be negative.
// This implies that unless we send a window update, the sender won't be able to send all the bytes
// for this message. Therefore we must send an update over the limit since there's an active read
// request from the application.
if estUntransmittedData > estSenderQuota {
// Sender's window shouldn't go more than 2^31 - 1 as specified in the HTTP spec.
if f.limit+n > maxWindowSize {
f.delta = maxWindowSize - f.limit
} else {
// Send a window update for the whole message and not just the difference between
// estUntransmittedData and estSenderQuota. This will be helpful in case the message
// is padded; We will fallback on the current available window(at least a 1/4th of the limit).
f.delta = n
}
f.mu.Unlock()
return f.delta
}
f.mu.Unlock()
return 0
}
// onData is invoked when some data frame is received. It updates pendingData.
func (f *inFlow) onData(n uint32) error {
f.mu.Lock()
f.pendingData += n
if f.pendingData+f.pendingUpdate > f.limit+f.delta {
limit := f.limit
rcvd := f.pendingData + f.pendingUpdate
f.mu.Unlock()
return fmt.Errorf("received %d-bytes data exceeding the limit %d bytes", rcvd, limit)
}
f.mu.Unlock()
return nil
}
// onRead is invoked when the application reads the data. It returns the window size
// to be sent to the peer.
func (f *inFlow) onRead(n uint32) uint32 {
f.mu.Lock()
if f.pendingData == 0 {
f.mu.Unlock()
return 0
}
f.pendingData -= n
if n > f.delta {
n -= f.delta
f.delta = 0
} else {
f.delta -= n
n = 0
}
f.pendingUpdate += n
if f.pendingUpdate >= f.limit/4 {
wu := f.pendingUpdate
f.pendingUpdate = 0
f.mu.Unlock()
return wu
}
f.mu.Unlock()
return 0
}

View file

@ -0,0 +1,430 @@
/*
*
* Copyright 2016 gRPC authors.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*/
// This file is the implementation of a gRPC server using HTTP/2 which
// uses the standard Go http2 Server implementation (via the
// http.Handler interface), rather than speaking low-level HTTP/2
// frames itself. It is the implementation of *grpc.Server.ServeHTTP.
package transport
import (
"context"
"errors"
"fmt"
"io"
"net"
"net/http"
"strings"
"sync"
"time"
"github.com/golang/protobuf/proto"
"golang.org/x/net/http2"
"google.golang.org/grpc/codes"
"google.golang.org/grpc/credentials"
"google.golang.org/grpc/metadata"
"google.golang.org/grpc/peer"
"google.golang.org/grpc/stats"
"google.golang.org/grpc/status"
)
// NewServerHandlerTransport returns a ServerTransport handling gRPC
// from inside an http.Handler. It requires that the http Server
// supports HTTP/2.
func NewServerHandlerTransport(w http.ResponseWriter, r *http.Request, stats stats.Handler) (ServerTransport, error) {
if r.ProtoMajor != 2 {
return nil, errors.New("gRPC requires HTTP/2")
}
if r.Method != "POST" {
return nil, errors.New("invalid gRPC request method")
}
contentType := r.Header.Get("Content-Type")
// TODO: do we assume contentType is lowercase? we did before
contentSubtype, validContentType := contentSubtype(contentType)
if !validContentType {
return nil, errors.New("invalid gRPC request content-type")
}
if _, ok := w.(http.Flusher); !ok {
return nil, errors.New("gRPC requires a ResponseWriter supporting http.Flusher")
}
st := &serverHandlerTransport{
rw: w,
req: r,
closedCh: make(chan struct{}),
writes: make(chan func()),
contentType: contentType,
contentSubtype: contentSubtype,
stats: stats,
}
if v := r.Header.Get("grpc-timeout"); v != "" {
to, err := decodeTimeout(v)
if err != nil {
return nil, status.Errorf(codes.Internal, "malformed time-out: %v", err)
}
st.timeoutSet = true
st.timeout = to
}
metakv := []string{"content-type", contentType}
if r.Host != "" {
metakv = append(metakv, ":authority", r.Host)
}
for k, vv := range r.Header {
k = strings.ToLower(k)
if isReservedHeader(k) && !isWhitelistedHeader(k) {
continue
}
for _, v := range vv {
v, err := decodeMetadataHeader(k, v)
if err != nil {
return nil, status.Errorf(codes.Internal, "malformed binary metadata: %v", err)
}
metakv = append(metakv, k, v)
}
}
st.headerMD = metadata.Pairs(metakv...)
return st, nil
}
// serverHandlerTransport is an implementation of ServerTransport
// which replies to exactly one gRPC request (exactly one HTTP request),
// using the net/http.Handler interface. This http.Handler is guaranteed
// at this point to be speaking over HTTP/2, so it's able to speak valid
// gRPC.
type serverHandlerTransport struct {
rw http.ResponseWriter
req *http.Request
timeoutSet bool
timeout time.Duration
didCommonHeaders bool
headerMD metadata.MD
closeOnce sync.Once
closedCh chan struct{} // closed on Close
// writes is a channel of code to run serialized in the
// ServeHTTP (HandleStreams) goroutine. The channel is closed
// when WriteStatus is called.
writes chan func()
// block concurrent WriteStatus calls
// e.g. grpc/(*serverStream).SendMsg/RecvMsg
writeStatusMu sync.Mutex
// we just mirror the request content-type
contentType string
// we store both contentType and contentSubtype so we don't keep recreating them
// TODO make sure this is consistent across handler_server and http2_server
contentSubtype string
stats stats.Handler
}
func (ht *serverHandlerTransport) Close() error {
ht.closeOnce.Do(ht.closeCloseChanOnce)
return nil
}
func (ht *serverHandlerTransport) closeCloseChanOnce() { close(ht.closedCh) }
func (ht *serverHandlerTransport) RemoteAddr() net.Addr { return strAddr(ht.req.RemoteAddr) }
// strAddr is a net.Addr backed by either a TCP "ip:port" string, or
// the empty string if unknown.
type strAddr string
func (a strAddr) Network() string {
if a != "" {
// Per the documentation on net/http.Request.RemoteAddr, if this is
// set, it's set to the IP:port of the peer (hence, TCP):
// https://golang.org/pkg/net/http/#Request
//
// If we want to support Unix sockets later, we can
// add our own grpc-specific convention within the
// grpc codebase to set RemoteAddr to a different
// format, or probably better: we can attach it to the
// context and use that from serverHandlerTransport.RemoteAddr.
return "tcp"
}
return ""
}
func (a strAddr) String() string { return string(a) }
// do runs fn in the ServeHTTP goroutine.
func (ht *serverHandlerTransport) do(fn func()) error {
select {
case <-ht.closedCh:
return ErrConnClosing
case ht.writes <- fn:
return nil
}
}
func (ht *serverHandlerTransport) WriteStatus(s *Stream, st *status.Status) error {
ht.writeStatusMu.Lock()
defer ht.writeStatusMu.Unlock()
err := ht.do(func() {
ht.writeCommonHeaders(s)
// And flush, in case no header or body has been sent yet.
// This forces a separation of headers and trailers if this is the
// first call (for example, in end2end tests's TestNoService).
ht.rw.(http.Flusher).Flush()
h := ht.rw.Header()
h.Set("Grpc-Status", fmt.Sprintf("%d", st.Code()))
if m := st.Message(); m != "" {
h.Set("Grpc-Message", encodeGrpcMessage(m))
}
if p := st.Proto(); p != nil && len(p.Details) > 0 {
stBytes, err := proto.Marshal(p)
if err != nil {
// TODO: return error instead, when callers are able to handle it.
panic(err)
}
h.Set("Grpc-Status-Details-Bin", encodeBinHeader(stBytes))
}
if md := s.Trailer(); len(md) > 0 {
for k, vv := range md {
// Clients don't tolerate reading restricted headers after some non restricted ones were sent.
if isReservedHeader(k) {
continue
}
for _, v := range vv {
// http2 ResponseWriter mechanism to send undeclared Trailers after
// the headers have possibly been written.
h.Add(http2.TrailerPrefix+k, encodeMetadataHeader(k, v))
}
}
}
})
if err == nil { // transport has not been closed
if ht.stats != nil {
ht.stats.HandleRPC(s.Context(), &stats.OutTrailer{})
}
}
ht.Close()
return err
}
// writeCommonHeaders sets common headers on the first write
// call (Write, WriteHeader, or WriteStatus).
func (ht *serverHandlerTransport) writeCommonHeaders(s *Stream) {
if ht.didCommonHeaders {
return
}
ht.didCommonHeaders = true
h := ht.rw.Header()
h["Date"] = nil // suppress Date to make tests happy; TODO: restore
h.Set("Content-Type", ht.contentType)
// Predeclare trailers we'll set later in WriteStatus (after the body).
// This is a SHOULD in the HTTP RFC, and the way you add (known)
// Trailers per the net/http.ResponseWriter contract.
// See https://golang.org/pkg/net/http/#ResponseWriter
// and https://golang.org/pkg/net/http/#example_ResponseWriter_trailers
h.Add("Trailer", "Grpc-Status")
h.Add("Trailer", "Grpc-Message")
h.Add("Trailer", "Grpc-Status-Details-Bin")
if s.sendCompress != "" {
h.Set("Grpc-Encoding", s.sendCompress)
}
}
func (ht *serverHandlerTransport) Write(s *Stream, hdr []byte, data []byte, opts *Options) error {
return ht.do(func() {
ht.writeCommonHeaders(s)
ht.rw.Write(hdr)
ht.rw.Write(data)
ht.rw.(http.Flusher).Flush()
})
}
func (ht *serverHandlerTransport) WriteHeader(s *Stream, md metadata.MD) error {
err := ht.do(func() {
ht.writeCommonHeaders(s)
h := ht.rw.Header()
for k, vv := range md {
// Clients don't tolerate reading restricted headers after some non restricted ones were sent.
if isReservedHeader(k) {
continue
}
for _, v := range vv {
v = encodeMetadataHeader(k, v)
h.Add(k, v)
}
}
ht.rw.WriteHeader(200)
ht.rw.(http.Flusher).Flush()
})
if err == nil {
if ht.stats != nil {
ht.stats.HandleRPC(s.Context(), &stats.OutHeader{})
}
}
return err
}
func (ht *serverHandlerTransport) HandleStreams(startStream func(*Stream), traceCtx func(context.Context, string) context.Context) {
// With this transport type there will be exactly 1 stream: this HTTP request.
ctx := ht.req.Context()
var cancel context.CancelFunc
if ht.timeoutSet {
ctx, cancel = context.WithTimeout(ctx, ht.timeout)
} else {
ctx, cancel = context.WithCancel(ctx)
}
// requestOver is closed when the status has been written via WriteStatus.
requestOver := make(chan struct{})
go func() {
select {
case <-requestOver:
case <-ht.closedCh:
case <-ht.req.Context().Done():
}
cancel()
ht.Close()
}()
req := ht.req
s := &Stream{
id: 0, // irrelevant
requestRead: func(int) {},
cancel: cancel,
buf: newRecvBuffer(),
st: ht,
method: req.URL.Path,
recvCompress: req.Header.Get("grpc-encoding"),
contentSubtype: ht.contentSubtype,
}
pr := &peer.Peer{
Addr: ht.RemoteAddr(),
}
if req.TLS != nil {
pr.AuthInfo = credentials.TLSInfo{State: *req.TLS}
}
ctx = metadata.NewIncomingContext(ctx, ht.headerMD)
s.ctx = peer.NewContext(ctx, pr)
if ht.stats != nil {
s.ctx = ht.stats.TagRPC(s.ctx, &stats.RPCTagInfo{FullMethodName: s.method})
inHeader := &stats.InHeader{
FullMethod: s.method,
RemoteAddr: ht.RemoteAddr(),
Compression: s.recvCompress,
}
ht.stats.HandleRPC(s.ctx, inHeader)
}
s.trReader = &transportReader{
reader: &recvBufferReader{ctx: s.ctx, ctxDone: s.ctx.Done(), recv: s.buf},
windowHandler: func(int) {},
}
// readerDone is closed when the Body.Read-ing goroutine exits.
readerDone := make(chan struct{})
go func() {
defer close(readerDone)
// TODO: minimize garbage, optimize recvBuffer code/ownership
const readSize = 8196
for buf := make([]byte, readSize); ; {
n, err := req.Body.Read(buf)
if n > 0 {
s.buf.put(recvMsg{data: buf[:n:n]})
buf = buf[n:]
}
if err != nil {
s.buf.put(recvMsg{err: mapRecvMsgError(err)})
return
}
if len(buf) == 0 {
buf = make([]byte, readSize)
}
}
}()
// startStream is provided by the *grpc.Server's serveStreams.
// It starts a goroutine serving s and exits immediately.
// The goroutine that is started is the one that then calls
// into ht, calling WriteHeader, Write, WriteStatus, Close, etc.
startStream(s)
ht.runStream()
close(requestOver)
// Wait for reading goroutine to finish.
req.Body.Close()
<-readerDone
}
func (ht *serverHandlerTransport) runStream() {
for {
select {
case fn := <-ht.writes:
fn()
case <-ht.closedCh:
return
}
}
}
func (ht *serverHandlerTransport) IncrMsgSent() {}
func (ht *serverHandlerTransport) IncrMsgRecv() {}
func (ht *serverHandlerTransport) Drain() {
panic("Drain() is not implemented")
}
// mapRecvMsgError returns the non-nil err into the appropriate
// error value as expected by callers of *grpc.parser.recvMsg.
// In particular, in can only be:
// * io.EOF
// * io.ErrUnexpectedEOF
// * of type transport.ConnectionError
// * an error from the status package
func mapRecvMsgError(err error) error {
if err == io.EOF || err == io.ErrUnexpectedEOF {
return err
}
if se, ok := err.(http2.StreamError); ok {
if code, ok := http2ErrConvTab[se.Code]; ok {
return status.Error(code, se.Error())
}
}
if strings.Contains(err.Error(), "body closed by handler") {
return status.Error(codes.Canceled, err.Error())
}
return connectionErrorf(true, err, err.Error())
}

File diff suppressed because it is too large Load diff

File diff suppressed because it is too large Load diff

View file

@ -0,0 +1,676 @@
/*
*
* Copyright 2014 gRPC authors.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*/
package transport
import (
"bufio"
"bytes"
"encoding/base64"
"fmt"
"io"
"math"
"net"
"net/http"
"strconv"
"strings"
"time"
"unicode/utf8"
"github.com/golang/protobuf/proto"
"golang.org/x/net/http2"
"golang.org/x/net/http2/hpack"
spb "google.golang.org/genproto/googleapis/rpc/status"
"google.golang.org/grpc/codes"
"google.golang.org/grpc/status"
)
const (
// http2MaxFrameLen specifies the max length of a HTTP2 frame.
http2MaxFrameLen = 16384 // 16KB frame
// http://http2.github.io/http2-spec/#SettingValues
http2InitHeaderTableSize = 4096
// baseContentType is the base content-type for gRPC. This is a valid
// content-type on it's own, but can also include a content-subtype such as
// "proto" as a suffix after "+" or ";". See
// https://github.com/grpc/grpc/blob/master/doc/PROTOCOL-HTTP2.md#requests
// for more details.
baseContentType = "application/grpc"
)
var (
clientPreface = []byte(http2.ClientPreface)
http2ErrConvTab = map[http2.ErrCode]codes.Code{
http2.ErrCodeNo: codes.Internal,
http2.ErrCodeProtocol: codes.Internal,
http2.ErrCodeInternal: codes.Internal,
http2.ErrCodeFlowControl: codes.ResourceExhausted,
http2.ErrCodeSettingsTimeout: codes.Internal,
http2.ErrCodeStreamClosed: codes.Internal,
http2.ErrCodeFrameSize: codes.Internal,
http2.ErrCodeRefusedStream: codes.Unavailable,
http2.ErrCodeCancel: codes.Canceled,
http2.ErrCodeCompression: codes.Internal,
http2.ErrCodeConnect: codes.Internal,
http2.ErrCodeEnhanceYourCalm: codes.ResourceExhausted,
http2.ErrCodeInadequateSecurity: codes.PermissionDenied,
http2.ErrCodeHTTP11Required: codes.Internal,
}
statusCodeConvTab = map[codes.Code]http2.ErrCode{
codes.Internal: http2.ErrCodeInternal,
codes.Canceled: http2.ErrCodeCancel,
codes.Unavailable: http2.ErrCodeRefusedStream,
codes.ResourceExhausted: http2.ErrCodeEnhanceYourCalm,
codes.PermissionDenied: http2.ErrCodeInadequateSecurity,
}
// HTTPStatusConvTab is the HTTP status code to gRPC error code conversion table.
HTTPStatusConvTab = map[int]codes.Code{
// 400 Bad Request - INTERNAL.
http.StatusBadRequest: codes.Internal,
// 401 Unauthorized - UNAUTHENTICATED.
http.StatusUnauthorized: codes.Unauthenticated,
// 403 Forbidden - PERMISSION_DENIED.
http.StatusForbidden: codes.PermissionDenied,
// 404 Not Found - UNIMPLEMENTED.
http.StatusNotFound: codes.Unimplemented,
// 429 Too Many Requests - UNAVAILABLE.
http.StatusTooManyRequests: codes.Unavailable,
// 502 Bad Gateway - UNAVAILABLE.
http.StatusBadGateway: codes.Unavailable,
// 503 Service Unavailable - UNAVAILABLE.
http.StatusServiceUnavailable: codes.Unavailable,
// 504 Gateway timeout - UNAVAILABLE.
http.StatusGatewayTimeout: codes.Unavailable,
}
)
type parsedHeaderData struct {
encoding string
// statusGen caches the stream status received from the trailer the server
// sent. Client side only. Do not access directly. After all trailers are
// parsed, use the status method to retrieve the status.
statusGen *status.Status
// rawStatusCode and rawStatusMsg are set from the raw trailer fields and are not
// intended for direct access outside of parsing.
rawStatusCode *int
rawStatusMsg string
httpStatus *int
// Server side only fields.
timeoutSet bool
timeout time.Duration
method string
// key-value metadata map from the peer.
mdata map[string][]string
statsTags []byte
statsTrace []byte
contentSubtype string
// isGRPC field indicates whether the peer is speaking gRPC (otherwise HTTP).
//
// We are in gRPC mode (peer speaking gRPC) if:
// * We are client side and have already received a HEADER frame that indicates gRPC peer.
// * The header contains valid a content-type, i.e. a string starts with "application/grpc"
// And we should handle error specific to gRPC.
//
// Otherwise (i.e. a content-type string starts without "application/grpc", or does not exist), we
// are in HTTP fallback mode, and should handle error specific to HTTP.
isGRPC bool
grpcErr error
httpErr error
contentTypeErr string
}
// decodeState configures decoding criteria and records the decoded data.
type decodeState struct {
// whether decoding on server side or not
serverSide bool
// Records the states during HPACK decoding. It will be filled with info parsed from HTTP HEADERS
// frame once decodeHeader function has been invoked and returned.
data parsedHeaderData
}
// isReservedHeader checks whether hdr belongs to HTTP2 headers
// reserved by gRPC protocol. Any other headers are classified as the
// user-specified metadata.
func isReservedHeader(hdr string) bool {
if hdr != "" && hdr[0] == ':' {
return true
}
switch hdr {
case "content-type",
"user-agent",
"grpc-message-type",
"grpc-encoding",
"grpc-message",
"grpc-status",
"grpc-timeout",
"grpc-status-details-bin",
// Intentionally exclude grpc-previous-rpc-attempts and
// grpc-retry-pushback-ms, which are "reserved", but their API
// intentionally works via metadata.
"te":
return true
default:
return false
}
}
// isWhitelistedHeader checks whether hdr should be propagated into metadata
// visible to users, even though it is classified as "reserved", above.
func isWhitelistedHeader(hdr string) bool {
switch hdr {
case ":authority", "user-agent":
return true
default:
return false
}
}
// contentSubtype returns the content-subtype for the given content-type. The
// given content-type must be a valid content-type that starts with
// "application/grpc". A content-subtype will follow "application/grpc" after a
// "+" or ";". See
// https://github.com/grpc/grpc/blob/master/doc/PROTOCOL-HTTP2.md#requests for
// more details.
//
// If contentType is not a valid content-type for gRPC, the boolean
// will be false, otherwise true. If content-type == "application/grpc",
// "application/grpc+", or "application/grpc;", the boolean will be true,
// but no content-subtype will be returned.
//
// contentType is assumed to be lowercase already.
func contentSubtype(contentType string) (string, bool) {
if contentType == baseContentType {
return "", true
}
if !strings.HasPrefix(contentType, baseContentType) {
return "", false
}
// guaranteed since != baseContentType and has baseContentType prefix
switch contentType[len(baseContentType)] {
case '+', ';':
// this will return true for "application/grpc+" or "application/grpc;"
// which the previous validContentType function tested to be valid, so we
// just say that no content-subtype is specified in this case
return contentType[len(baseContentType)+1:], true
default:
return "", false
}
}
// contentSubtype is assumed to be lowercase
func contentType(contentSubtype string) string {
if contentSubtype == "" {
return baseContentType
}
return baseContentType + "+" + contentSubtype
}
func (d *decodeState) status() *status.Status {
if d.data.statusGen == nil {
// No status-details were provided; generate status using code/msg.
d.data.statusGen = status.New(codes.Code(int32(*(d.data.rawStatusCode))), d.data.rawStatusMsg)
}
return d.data.statusGen
}
const binHdrSuffix = "-bin"
func encodeBinHeader(v []byte) string {
return base64.RawStdEncoding.EncodeToString(v)
}
func decodeBinHeader(v string) ([]byte, error) {
if len(v)%4 == 0 {
// Input was padded, or padding was not necessary.
return base64.StdEncoding.DecodeString(v)
}
return base64.RawStdEncoding.DecodeString(v)
}
func encodeMetadataHeader(k, v string) string {
if strings.HasSuffix(k, binHdrSuffix) {
return encodeBinHeader(([]byte)(v))
}
return v
}
func decodeMetadataHeader(k, v string) (string, error) {
if strings.HasSuffix(k, binHdrSuffix) {
b, err := decodeBinHeader(v)
return string(b), err
}
return v, nil
}
func (d *decodeState) decodeHeader(frame *http2.MetaHeadersFrame) error {
// frame.Truncated is set to true when framer detects that the current header
// list size hits MaxHeaderListSize limit.
if frame.Truncated {
return status.Error(codes.Internal, "peer header list size exceeded limit")
}
for _, hf := range frame.Fields {
d.processHeaderField(hf)
}
if d.data.isGRPC {
if d.data.grpcErr != nil {
return d.data.grpcErr
}
if d.serverSide {
return nil
}
if d.data.rawStatusCode == nil && d.data.statusGen == nil {
// gRPC status doesn't exist.
// Set rawStatusCode to be unknown and return nil error.
// So that, if the stream has ended this Unknown status
// will be propagated to the user.
// Otherwise, it will be ignored. In which case, status from
// a later trailer, that has StreamEnded flag set, is propagated.
code := int(codes.Unknown)
d.data.rawStatusCode = &code
}
return nil
}
// HTTP fallback mode
if d.data.httpErr != nil {
return d.data.httpErr
}
var (
code = codes.Internal // when header does not include HTTP status, return INTERNAL
ok bool
)
if d.data.httpStatus != nil {
code, ok = HTTPStatusConvTab[*(d.data.httpStatus)]
if !ok {
code = codes.Unknown
}
}
return status.Error(code, d.constructHTTPErrMsg())
}
// constructErrMsg constructs error message to be returned in HTTP fallback mode.
// Format: HTTP status code and its corresponding message + content-type error message.
func (d *decodeState) constructHTTPErrMsg() string {
var errMsgs []string
if d.data.httpStatus == nil {
errMsgs = append(errMsgs, "malformed header: missing HTTP status")
} else {
errMsgs = append(errMsgs, fmt.Sprintf("%s: HTTP status code %d", http.StatusText(*(d.data.httpStatus)), *d.data.httpStatus))
}
if d.data.contentTypeErr == "" {
errMsgs = append(errMsgs, "transport: missing content-type field")
} else {
errMsgs = append(errMsgs, d.data.contentTypeErr)
}
return strings.Join(errMsgs, "; ")
}
func (d *decodeState) addMetadata(k, v string) {
if d.data.mdata == nil {
d.data.mdata = make(map[string][]string)
}
d.data.mdata[k] = append(d.data.mdata[k], v)
}
func (d *decodeState) processHeaderField(f hpack.HeaderField) {
switch f.Name {
case "content-type":
contentSubtype, validContentType := contentSubtype(f.Value)
if !validContentType {
d.data.contentTypeErr = fmt.Sprintf("transport: received the unexpected content-type %q", f.Value)
return
}
d.data.contentSubtype = contentSubtype
// TODO: do we want to propagate the whole content-type in the metadata,
// or come up with a way to just propagate the content-subtype if it was set?
// ie {"content-type": "application/grpc+proto"} or {"content-subtype": "proto"}
// in the metadata?
d.addMetadata(f.Name, f.Value)
d.data.isGRPC = true
case "grpc-encoding":
d.data.encoding = f.Value
case "grpc-status":
code, err := strconv.Atoi(f.Value)
if err != nil {
d.data.grpcErr = status.Errorf(codes.Internal, "transport: malformed grpc-status: %v", err)
return
}
d.data.rawStatusCode = &code
case "grpc-message":
d.data.rawStatusMsg = decodeGrpcMessage(f.Value)
case "grpc-status-details-bin":
v, err := decodeBinHeader(f.Value)
if err != nil {
d.data.grpcErr = status.Errorf(codes.Internal, "transport: malformed grpc-status-details-bin: %v", err)
return
}
s := &spb.Status{}
if err := proto.Unmarshal(v, s); err != nil {
d.data.grpcErr = status.Errorf(codes.Internal, "transport: malformed grpc-status-details-bin: %v", err)
return
}
d.data.statusGen = status.FromProto(s)
case "grpc-timeout":
d.data.timeoutSet = true
var err error
if d.data.timeout, err = decodeTimeout(f.Value); err != nil {
d.data.grpcErr = status.Errorf(codes.Internal, "transport: malformed time-out: %v", err)
}
case ":path":
d.data.method = f.Value
case ":status":
code, err := strconv.Atoi(f.Value)
if err != nil {
d.data.httpErr = status.Errorf(codes.Internal, "transport: malformed http-status: %v", err)
return
}
d.data.httpStatus = &code
case "grpc-tags-bin":
v, err := decodeBinHeader(f.Value)
if err != nil {
d.data.grpcErr = status.Errorf(codes.Internal, "transport: malformed grpc-tags-bin: %v", err)
return
}
d.data.statsTags = v
d.addMetadata(f.Name, string(v))
case "grpc-trace-bin":
v, err := decodeBinHeader(f.Value)
if err != nil {
d.data.grpcErr = status.Errorf(codes.Internal, "transport: malformed grpc-trace-bin: %v", err)
return
}
d.data.statsTrace = v
d.addMetadata(f.Name, string(v))
default:
if isReservedHeader(f.Name) && !isWhitelistedHeader(f.Name) {
break
}
v, err := decodeMetadataHeader(f.Name, f.Value)
if err != nil {
errorf("Failed to decode metadata header (%q, %q): %v", f.Name, f.Value, err)
return
}
d.addMetadata(f.Name, v)
}
}
type timeoutUnit uint8
const (
hour timeoutUnit = 'H'
minute timeoutUnit = 'M'
second timeoutUnit = 'S'
millisecond timeoutUnit = 'm'
microsecond timeoutUnit = 'u'
nanosecond timeoutUnit = 'n'
)
func timeoutUnitToDuration(u timeoutUnit) (d time.Duration, ok bool) {
switch u {
case hour:
return time.Hour, true
case minute:
return time.Minute, true
case second:
return time.Second, true
case millisecond:
return time.Millisecond, true
case microsecond:
return time.Microsecond, true
case nanosecond:
return time.Nanosecond, true
default:
}
return
}
const maxTimeoutValue int64 = 100000000 - 1
// div does integer division and round-up the result. Note that this is
// equivalent to (d+r-1)/r but has less chance to overflow.
func div(d, r time.Duration) int64 {
if m := d % r; m > 0 {
return int64(d/r + 1)
}
return int64(d / r)
}
// TODO(zhaoq): It is the simplistic and not bandwidth efficient. Improve it.
func encodeTimeout(t time.Duration) string {
if t <= 0 {
return "0n"
}
if d := div(t, time.Nanosecond); d <= maxTimeoutValue {
return strconv.FormatInt(d, 10) + "n"
}
if d := div(t, time.Microsecond); d <= maxTimeoutValue {
return strconv.FormatInt(d, 10) + "u"
}
if d := div(t, time.Millisecond); d <= maxTimeoutValue {
return strconv.FormatInt(d, 10) + "m"
}
if d := div(t, time.Second); d <= maxTimeoutValue {
return strconv.FormatInt(d, 10) + "S"
}
if d := div(t, time.Minute); d <= maxTimeoutValue {
return strconv.FormatInt(d, 10) + "M"
}
// Note that maxTimeoutValue * time.Hour > MaxInt64.
return strconv.FormatInt(div(t, time.Hour), 10) + "H"
}
func decodeTimeout(s string) (time.Duration, error) {
size := len(s)
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 {
return 0, fmt.Errorf("transport: timeout unit is not recognized: %q", s)
}
t, err := strconv.ParseInt(s[:size-1], 10, 64)
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
}
const (
spaceByte = ' '
tildeByte = '~'
percentByte = '%'
)
// encodeGrpcMessage is used to encode status code in header field
// "grpc-message". It does percent encoding and also replaces invalid utf-8
// characters with Unicode replacement character.
//
// It checks to see if each individual byte in msg is an allowable byte, and
// then either percent encoding or passing it through. When percent encoding,
// the byte is converted into hexadecimal notation with a '%' prepended.
func encodeGrpcMessage(msg string) string {
if msg == "" {
return ""
}
lenMsg := len(msg)
for i := 0; i < lenMsg; i++ {
c := msg[i]
if !(c >= spaceByte && c <= tildeByte && c != percentByte) {
return encodeGrpcMessageUnchecked(msg)
}
}
return msg
}
func encodeGrpcMessageUnchecked(msg string) string {
var buf bytes.Buffer
for len(msg) > 0 {
r, size := utf8.DecodeRuneInString(msg)
for _, b := range []byte(string(r)) {
if size > 1 {
// If size > 1, r is not ascii. Always do percent encoding.
buf.WriteString(fmt.Sprintf("%%%02X", b))
continue
}
// The for loop is necessary even if size == 1. r could be
// utf8.RuneError.
//
// fmt.Sprintf("%%%02X", utf8.RuneError) gives "%FFFD".
if b >= spaceByte && b <= tildeByte && b != percentByte {
buf.WriteByte(b)
} else {
buf.WriteString(fmt.Sprintf("%%%02X", b))
}
}
msg = msg[size:]
}
return buf.String()
}
// decodeGrpcMessage decodes the msg encoded by encodeGrpcMessage.
func decodeGrpcMessage(msg string) string {
if msg == "" {
return ""
}
lenMsg := len(msg)
for i := 0; i < lenMsg; i++ {
if msg[i] == percentByte && i+2 < lenMsg {
return decodeGrpcMessageUnchecked(msg)
}
}
return msg
}
func decodeGrpcMessageUnchecked(msg string) string {
var buf bytes.Buffer
lenMsg := len(msg)
for i := 0; i < lenMsg; i++ {
c := msg[i]
if c == percentByte && i+2 < lenMsg {
parsed, err := strconv.ParseUint(msg[i+1:i+3], 16, 8)
if err != nil {
buf.WriteByte(c)
} else {
buf.WriteByte(byte(parsed))
i += 2
}
} else {
buf.WriteByte(c)
}
}
return buf.String()
}
type bufWriter struct {
buf []byte
offset int
batchSize int
conn net.Conn
err error
onFlush func()
}
func newBufWriter(conn net.Conn, batchSize int) *bufWriter {
return &bufWriter{
buf: make([]byte, batchSize*2),
batchSize: batchSize,
conn: conn,
}
}
func (w *bufWriter) Write(b []byte) (n int, err error) {
if w.err != nil {
return 0, w.err
}
if w.batchSize == 0 { // Buffer has been disabled.
return w.conn.Write(b)
}
for len(b) > 0 {
nn := copy(w.buf[w.offset:], b)
b = b[nn:]
w.offset += nn
n += nn
if w.offset >= w.batchSize {
err = w.Flush()
}
}
return n, err
}
func (w *bufWriter) Flush() error {
if w.err != nil {
return w.err
}
if w.offset == 0 {
return nil
}
if w.onFlush != nil {
w.onFlush()
}
_, w.err = w.conn.Write(w.buf[:w.offset])
w.offset = 0
return w.err
}
type framer struct {
writer *bufWriter
fr *http2.Framer
}
func newFramer(conn net.Conn, writeBufferSize, readBufferSize int, maxHeaderListSize uint32) *framer {
if writeBufferSize < 0 {
writeBufferSize = 0
}
var r io.Reader = conn
if readBufferSize > 0 {
r = bufio.NewReaderSize(r, readBufferSize)
}
w := newBufWriter(conn, writeBufferSize)
f := &framer{
writer: w,
fr: http2.NewFramer(w, r),
}
// Opt-in to Frame reuse API on framer to reduce garbage.
// Frames aren't safe to read from after a subsequent call to ReadFrame.
f.fr.SetReuseFrames()
f.fr.MaxHeaderListSize = maxHeaderListSize
f.fr.ReadMetaHeaders = hpack.NewDecoder(http2InitHeaderTableSize, nil)
return f
}

View file

@ -0,0 +1,44 @@
/*
*
* Copyright 2017 gRPC authors.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*/
// This file contains wrappers for grpclog functions.
// The transport package only logs to verbose level 2 by default.
package transport
import "google.golang.org/grpc/grpclog"
const logLevel = 2
func infof(format string, args ...interface{}) {
if grpclog.V(logLevel) {
grpclog.Infof(format, args...)
}
}
func warningf(format string, args ...interface{}) {
if grpclog.V(logLevel) {
grpclog.Warningf(format, args...)
}
}
func errorf(format string, args ...interface{}) {
if grpclog.V(logLevel) {
grpclog.Errorf(format, args...)
}
}

View file

@ -0,0 +1,760 @@
/*
*
* Copyright 2014 gRPC authors.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*/
// Package transport defines and implements message oriented communication
// channel to complete various transactions (e.g., an RPC). It is meant for
// grpc-internal usage and is not intended to be imported directly by users.
package transport
import (
"context"
"errors"
"fmt"
"io"
"net"
"sync"
"sync/atomic"
"google.golang.org/grpc/codes"
"google.golang.org/grpc/credentials"
"google.golang.org/grpc/keepalive"
"google.golang.org/grpc/metadata"
"google.golang.org/grpc/stats"
"google.golang.org/grpc/status"
"google.golang.org/grpc/tap"
)
// recvMsg represents the received msg from the transport. All transport
// protocol specific info has been removed.
type recvMsg struct {
data []byte
// nil: received some data
// io.EOF: stream is completed. data is nil.
// other non-nil error: transport failure. data is nil.
err error
}
// recvBuffer is an unbounded channel of recvMsg structs.
// Note recvBuffer differs from controlBuffer only in that recvBuffer
// holds a channel of only recvMsg structs instead of objects implementing "item" interface.
// recvBuffer is written to much more often than
// controlBuffer and using strict recvMsg structs helps avoid allocation in "recvBuffer.put"
type recvBuffer struct {
c chan recvMsg
mu sync.Mutex
backlog []recvMsg
err error
}
func newRecvBuffer() *recvBuffer {
b := &recvBuffer{
c: make(chan recvMsg, 1),
}
return b
}
func (b *recvBuffer) put(r recvMsg) {
b.mu.Lock()
if b.err != nil {
b.mu.Unlock()
// An error had occurred earlier, don't accept more
// data or errors.
return
}
b.err = r.err
if len(b.backlog) == 0 {
select {
case b.c <- r:
b.mu.Unlock()
return
default:
}
}
b.backlog = append(b.backlog, r)
b.mu.Unlock()
}
func (b *recvBuffer) load() {
b.mu.Lock()
if len(b.backlog) > 0 {
select {
case b.c <- b.backlog[0]:
b.backlog[0] = recvMsg{}
b.backlog = b.backlog[1:]
default:
}
}
b.mu.Unlock()
}
// get returns the channel that receives a recvMsg in the buffer.
//
// Upon receipt of a recvMsg, the caller should call load to send another
// recvMsg onto the channel if there is any.
func (b *recvBuffer) get() <-chan recvMsg {
return b.c
}
// recvBufferReader implements io.Reader interface to read the data from
// recvBuffer.
type recvBufferReader struct {
closeStream func(error) // Closes the client transport stream with the given error and nil trailer metadata.
ctx context.Context
ctxDone <-chan struct{} // cache of ctx.Done() (for performance).
recv *recvBuffer
last []byte // Stores the remaining data in the previous calls.
err error
}
// Read reads the next len(p) bytes from last. If last is drained, it tries to
// read additional data from recv. It blocks if there no additional data available
// in recv. If Read returns any non-nil error, it will continue to return that error.
func (r *recvBufferReader) Read(p []byte) (n int, err error) {
if r.err != nil {
return 0, r.err
}
if r.last != nil && len(r.last) > 0 {
// Read remaining data left in last call.
copied := copy(p, r.last)
r.last = r.last[copied:]
return copied, nil
}
if r.closeStream != nil {
n, r.err = r.readClient(p)
} else {
n, r.err = r.read(p)
}
return n, r.err
}
func (r *recvBufferReader) read(p []byte) (n int, err error) {
select {
case <-r.ctxDone:
return 0, ContextErr(r.ctx.Err())
case m := <-r.recv.get():
return r.readAdditional(m, p)
}
}
func (r *recvBufferReader) readClient(p []byte) (n int, err error) {
// If the context is canceled, then closes the stream with nil metadata.
// closeStream writes its error parameter to r.recv as a recvMsg.
// r.readAdditional acts on that message and returns the necessary error.
select {
case <-r.ctxDone:
r.closeStream(ContextErr(r.ctx.Err()))
m := <-r.recv.get()
return r.readAdditional(m, p)
case m := <-r.recv.get():
return r.readAdditional(m, p)
}
}
func (r *recvBufferReader) readAdditional(m recvMsg, p []byte) (n int, err error) {
r.recv.load()
if m.err != nil {
return 0, m.err
}
copied := copy(p, m.data)
r.last = m.data[copied:]
return copied, nil
}
type streamState uint32
const (
streamActive streamState = iota
streamWriteDone // EndStream sent
streamReadDone // EndStream received
streamDone // the entire stream is finished.
)
// Stream represents an RPC in the transport layer.
type Stream struct {
id uint32
st ServerTransport // nil for client side Stream
ctx context.Context // the associated context of the stream
cancel context.CancelFunc // always nil for client side Stream
done chan struct{} // closed at the end of stream to unblock writers. On the client side.
ctxDone <-chan struct{} // same as done chan but for server side. Cache of ctx.Done() (for performance)
method string // the associated RPC method of the stream
recvCompress string
sendCompress string
buf *recvBuffer
trReader io.Reader
fc *inFlow
wq *writeQuota
// Callback to state application's intentions to read data. This
// is used to adjust flow control, if needed.
requestRead func(int)
headerChan chan struct{} // closed to indicate the end of header metadata.
headerDone uint32 // set when headerChan is closed. Used to avoid closing headerChan multiple times.
// hdrMu protects header and trailer metadata on the server-side.
hdrMu sync.Mutex
// On client side, header keeps the received header metadata.
//
// On server side, header keeps the header set by SetHeader(). The complete
// header will merged into this after t.WriteHeader() is called.
header metadata.MD
trailer metadata.MD // the key-value map of trailer metadata.
noHeaders bool // set if the client never received headers (set only after the stream is done).
// On the server-side, headerSent is atomically set to 1 when the headers are sent out.
headerSent uint32
state streamState
// On client-side it is the status error received from the server.
// On server-side it is unused.
status *status.Status
bytesReceived uint32 // indicates whether any bytes have been received on this stream
unprocessed uint32 // set if the server sends a refused stream or GOAWAY including this stream
// contentSubtype is the content-subtype for requests.
// this must be lowercase or the behavior is undefined.
contentSubtype string
}
// isHeaderSent is only valid on the server-side.
func (s *Stream) isHeaderSent() bool {
return atomic.LoadUint32(&s.headerSent) == 1
}
// updateHeaderSent updates headerSent and returns true
// if it was alreay set. It is valid only on server-side.
func (s *Stream) updateHeaderSent() bool {
return atomic.SwapUint32(&s.headerSent, 1) == 1
}
func (s *Stream) swapState(st streamState) streamState {
return streamState(atomic.SwapUint32((*uint32)(&s.state), uint32(st)))
}
func (s *Stream) compareAndSwapState(oldState, newState streamState) bool {
return atomic.CompareAndSwapUint32((*uint32)(&s.state), uint32(oldState), uint32(newState))
}
func (s *Stream) getState() streamState {
return streamState(atomic.LoadUint32((*uint32)(&s.state)))
}
func (s *Stream) waitOnHeader() error {
if s.headerChan == nil {
// On the server headerChan is always nil since a stream originates
// only after having received headers.
return nil
}
select {
case <-s.ctx.Done():
return ContextErr(s.ctx.Err())
case <-s.headerChan:
return nil
}
}
// RecvCompress returns the compression algorithm applied to the inbound
// message. It is empty string if there is no compression applied.
func (s *Stream) RecvCompress() string {
if err := s.waitOnHeader(); err != nil {
return ""
}
return s.recvCompress
}
// SetSendCompress sets the compression algorithm to the stream.
func (s *Stream) SetSendCompress(str string) {
s.sendCompress = str
}
// Done returns a channel which is closed when it receives the final status
// from the server.
func (s *Stream) Done() <-chan struct{} {
return s.done
}
// Header returns the header metadata of the stream.
//
// On client side, it acquires the key-value pairs of header metadata once it is
// available. It blocks until i) the metadata is ready or ii) there is no header
// metadata or iii) the stream is canceled/expired.
//
// On server side, it returns the out header after t.WriteHeader is called.
func (s *Stream) Header() (metadata.MD, error) {
if s.headerChan == nil && s.header != nil {
// On server side, return the header in stream. It will be the out
// header after t.WriteHeader is called.
return s.header.Copy(), nil
}
err := s.waitOnHeader()
// Even if the stream is closed, header is returned if available.
select {
case <-s.headerChan:
if s.header == nil {
return nil, nil
}
return s.header.Copy(), nil
default:
}
return nil, err
}
// TrailersOnly blocks until a header or trailers-only frame is received and
// then returns true if the stream was trailers-only. If the stream ends
// before headers are received, returns true, nil. If a context error happens
// first, returns it as a status error. Client-side only.
func (s *Stream) TrailersOnly() (bool, error) {
err := s.waitOnHeader()
if err != nil {
return false, err
}
return s.noHeaders, nil
}
// Trailer returns the cached trailer metedata. Note that if it is not called
// after the entire stream is done, it could return an empty MD. Client
// side only.
// It can be safely read only after stream has ended that is either read
// or write have returned io.EOF.
func (s *Stream) Trailer() metadata.MD {
c := s.trailer.Copy()
return c
}
// ContentSubtype returns the content-subtype for a request. For example, a
// content-subtype of "proto" will result in a content-type of
// "application/grpc+proto". This will always be lowercase. See
// https://github.com/grpc/grpc/blob/master/doc/PROTOCOL-HTTP2.md#requests for
// more details.
func (s *Stream) ContentSubtype() string {
return s.contentSubtype
}
// Context returns the context of the stream.
func (s *Stream) Context() context.Context {
return s.ctx
}
// Method returns the method for the stream.
func (s *Stream) Method() string {
return s.method
}
// Status returns the status received from the server.
// Status can be read safely only after the stream has ended,
// that is, after Done() is closed.
func (s *Stream) Status() *status.Status {
return s.status
}
// SetHeader sets the header metadata. This can be called multiple times.
// Server side only.
// This should not be called in parallel to other data writes.
func (s *Stream) SetHeader(md metadata.MD) error {
if md.Len() == 0 {
return nil
}
if s.isHeaderSent() || s.getState() == streamDone {
return ErrIllegalHeaderWrite
}
s.hdrMu.Lock()
s.header = metadata.Join(s.header, md)
s.hdrMu.Unlock()
return nil
}
// SendHeader sends the given header metadata. The given metadata is
// combined with any metadata set by previous calls to SetHeader and
// then written to the transport stream.
func (s *Stream) SendHeader(md metadata.MD) error {
return s.st.WriteHeader(s, md)
}
// SetTrailer sets the trailer metadata which will be sent with the RPC status
// by the server. This can be called multiple times. Server side only.
// This should not be called parallel to other data writes.
func (s *Stream) SetTrailer(md metadata.MD) error {
if md.Len() == 0 {
return nil
}
if s.getState() == streamDone {
return ErrIllegalHeaderWrite
}
s.hdrMu.Lock()
s.trailer = metadata.Join(s.trailer, md)
s.hdrMu.Unlock()
return nil
}
func (s *Stream) write(m recvMsg) {
s.buf.put(m)
}
// Read reads all p bytes from the wire for this stream.
func (s *Stream) Read(p []byte) (n int, err error) {
// Don't request a read if there was an error earlier
if er := s.trReader.(*transportReader).er; er != nil {
return 0, er
}
s.requestRead(len(p))
return io.ReadFull(s.trReader, p)
}
// tranportReader reads all the data available for this Stream from the transport and
// passes them into the decoder, which converts them into a gRPC message stream.
// The error is io.EOF when the stream is done or another non-nil error if
// the stream broke.
type transportReader struct {
reader io.Reader
// The handler to control the window update procedure for both this
// particular stream and the associated transport.
windowHandler func(int)
er error
}
func (t *transportReader) Read(p []byte) (n int, err error) {
n, err = t.reader.Read(p)
if err != nil {
t.er = err
return
}
t.windowHandler(n)
return
}
// BytesReceived indicates whether any bytes have been received on this stream.
func (s *Stream) BytesReceived() bool {
return atomic.LoadUint32(&s.bytesReceived) == 1
}
// Unprocessed indicates whether the server did not process this stream --
// i.e. it sent a refused stream or GOAWAY including this stream ID.
func (s *Stream) Unprocessed() bool {
return atomic.LoadUint32(&s.unprocessed) == 1
}
// GoString is implemented by Stream so context.String() won't
// race when printing %#v.
func (s *Stream) GoString() string {
return fmt.Sprintf("<stream: %p, %v>", s, s.method)
}
// state of transport
type transportState int
const (
reachable transportState = iota
closing
draining
)
// ServerConfig consists of all the configurations to establish a server transport.
type ServerConfig struct {
MaxStreams uint32
AuthInfo credentials.AuthInfo
InTapHandle tap.ServerInHandle
StatsHandler stats.Handler
KeepaliveParams keepalive.ServerParameters
KeepalivePolicy keepalive.EnforcementPolicy
InitialWindowSize int32
InitialConnWindowSize int32
WriteBufferSize int
ReadBufferSize int
ChannelzParentID int64
MaxHeaderListSize *uint32
}
// NewServerTransport creates a ServerTransport with conn or non-nil error
// if it fails.
func NewServerTransport(protocol string, conn net.Conn, config *ServerConfig) (ServerTransport, error) {
return newHTTP2Server(conn, config)
}
// ConnectOptions covers all relevant options for communicating with the server.
type ConnectOptions struct {
// UserAgent is the application user agent.
UserAgent string
// Dialer specifies how to dial a network address.
Dialer func(context.Context, string) (net.Conn, error)
// FailOnNonTempDialError specifies if gRPC fails on non-temporary dial errors.
FailOnNonTempDialError bool
// PerRPCCredentials stores the PerRPCCredentials required to issue RPCs.
PerRPCCredentials []credentials.PerRPCCredentials
// 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.
StatsHandler stats.Handler
// InitialWindowSize sets the initial window size for a stream.
InitialWindowSize int32
// InitialConnWindowSize sets the initial window size for a connection.
InitialConnWindowSize int32
// WriteBufferSize sets the size of write buffer which in turn determines how much data can be batched before it's written on the wire.
WriteBufferSize int
// ReadBufferSize sets the size of read buffer, which in turn determines how much data can be read at most for one read syscall.
ReadBufferSize int
// ChannelzParentID sets the addrConn id which initiate the creation of this client transport.
ChannelzParentID int64
// MaxHeaderListSize sets the max (uncompressed) size of header list that is prepared to be received.
MaxHeaderListSize *uint32
}
// TargetInfo contains the information of the target such as network address and metadata.
type TargetInfo struct {
Addr string
Metadata interface{}
Authority string
}
// NewClientTransport establishes the transport with the required ConnectOptions
// and returns it to the caller.
func NewClientTransport(connectCtx, ctx context.Context, target TargetInfo, opts ConnectOptions, onPrefaceReceipt func(), onGoAway func(GoAwayReason), onClose func()) (ClientTransport, error) {
return newHTTP2Client(connectCtx, ctx, target, opts, onPrefaceReceipt, onGoAway, onClose)
}
// Options provides additional hints and information for message
// transmission.
type Options struct {
// Last indicates whether this write is the last piece for
// this stream.
Last bool
}
// CallHdr carries the information of a particular RPC.
type CallHdr struct {
// Host specifies the peer's host.
Host string
// Method specifies the operation to perform.
Method string
// SendCompress specifies the compression algorithm applied on
// outbound message.
SendCompress string
// Creds specifies credentials.PerRPCCredentials for a call.
Creds credentials.PerRPCCredentials
// ContentSubtype specifies the content-subtype for a request. For example, a
// content-subtype of "proto" will result in a content-type of
// "application/grpc+proto". The value of ContentSubtype must be all
// lowercase, otherwise the behavior is undefined. See
// https://github.com/grpc/grpc/blob/master/doc/PROTOCOL-HTTP2.md#requests
// for more details.
ContentSubtype string
PreviousAttempts int // value of grpc-previous-rpc-attempts header to set
}
// ClientTransport is the common interface for all gRPC client-side transport
// implementations.
type ClientTransport interface {
// Close tears down this transport. Once it returns, the transport
// should not be accessed any more. The caller must make sure this
// is called only once.
Close() error
// GracefulClose starts to tear down the transport. It stops accepting
// new RPCs and wait the completion of the pending RPCs.
GracefulClose() error
// Write sends the data for the given stream. A nil stream indicates
// the write is to be performed on the transport as a whole.
Write(s *Stream, hdr []byte, data []byte, opts *Options) error
// NewStream creates a Stream for an RPC.
NewStream(ctx context.Context, callHdr *CallHdr) (*Stream, error)
// CloseStream clears the footprint of a stream when the stream is
// not needed any more. The err indicates the error incurred when
// CloseStream is called. Must be called when a stream is finished
// unless the associated transport is closing.
CloseStream(stream *Stream, err error)
// Error returns a channel that is closed when some I/O error
// happens. Typically the caller should have a goroutine to monitor
// this in order to take action (e.g., close the current transport
// and create a new one) in error case. It should not return nil
// once the transport is initiated.
Error() <-chan struct{}
// GoAway returns a channel that is closed when ClientTransport
// receives the draining signal from the server (e.g., GOAWAY frame in
// HTTP/2).
GoAway() <-chan struct{}
// GetGoAwayReason returns the reason why GoAway frame was received.
GetGoAwayReason() GoAwayReason
// RemoteAddr returns the remote network address.
RemoteAddr() net.Addr
// IncrMsgSent increments the number of message sent through this transport.
IncrMsgSent()
// IncrMsgRecv increments the number of message received through this transport.
IncrMsgRecv()
}
// ServerTransport is the common interface for all gRPC server-side transport
// implementations.
//
// Methods may be called concurrently from multiple goroutines, but
// Write methods for a given Stream will be called serially.
type ServerTransport interface {
// HandleStreams receives incoming streams using the given handler.
HandleStreams(func(*Stream), func(context.Context, string) context.Context)
// WriteHeader sends the header metadata for the given stream.
// WriteHeader may not be called on all streams.
WriteHeader(s *Stream, md metadata.MD) error
// Write sends the data for the given stream.
// Write may not be called on all streams.
Write(s *Stream, hdr []byte, data []byte, opts *Options) error
// WriteStatus sends the status of a stream to the client. WriteStatus is
// the final call made on a stream and always occurs.
WriteStatus(s *Stream, st *status.Status) error
// Close tears down the transport. Once it is called, the transport
// should not be accessed any more. All the pending streams and their
// handlers will be terminated asynchronously.
Close() error
// RemoteAddr returns the remote network address.
RemoteAddr() net.Addr
// Drain notifies the client this ServerTransport stops accepting new RPCs.
Drain()
// IncrMsgSent increments the number of message sent through this transport.
IncrMsgSent()
// IncrMsgRecv increments the number of message received through this transport.
IncrMsgRecv()
}
// connectionErrorf creates an ConnectionError with the specified error description.
func connectionErrorf(temp bool, e error, format string, a ...interface{}) ConnectionError {
return ConnectionError{
Desc: fmt.Sprintf(format, a...),
temp: temp,
err: e,
}
}
// ConnectionError is an error that results in the termination of the
// entire connection and the retry of all the active streams.
type ConnectionError struct {
Desc string
temp bool
err error
}
func (e ConnectionError) Error() string {
return fmt.Sprintf("connection error: desc = %q", e.Desc)
}
// Temporary indicates if this connection error is temporary or fatal.
func (e ConnectionError) Temporary() bool {
return e.temp
}
// Origin returns the original error of this connection error.
func (e ConnectionError) Origin() error {
// Never return nil error here.
// If the original error is nil, return itself.
if e.err == nil {
return e
}
return e.err
}
var (
// ErrConnClosing indicates that the transport is closing.
ErrConnClosing = connectionErrorf(true, nil, "transport is closing")
// errStreamDrain indicates that the stream is rejected because the
// connection is draining. This could be caused by goaway or balancer
// removing the address.
errStreamDrain = status.Error(codes.Unavailable, "the connection is draining")
// errStreamDone is returned from write at the client side to indiacte application
// layer of an error.
errStreamDone = errors.New("the stream is done")
// StatusGoAway indicates that the server sent a GOAWAY that included this
// stream's ID in unprocessed RPCs.
statusGoAway = status.New(codes.Unavailable, "the stream is rejected because server is draining the connection")
)
// GoAwayReason contains the reason for the GoAway frame received.
type GoAwayReason uint8
const (
// GoAwayInvalid indicates that no GoAway frame is received.
GoAwayInvalid GoAwayReason = 0
// GoAwayNoReason is the default value when GoAway frame is received.
GoAwayNoReason GoAwayReason = 1
// GoAwayTooManyPings indicates that a GoAway frame with
// ErrCodeEnhanceYourCalm was received and that the debug data said
// "too_many_pings".
GoAwayTooManyPings GoAwayReason = 2
)
// channelzData is used to store channelz related data for http2Client and http2Server.
// These fields cannot be embedded in the original structs (e.g. http2Client), since to do atomic
// operation on int64 variable on 32-bit machine, user is responsible to enforce memory alignment.
// Here, by grouping those int64 fields inside a struct, we are enforcing the alignment.
type channelzData struct {
kpCount int64
// The number of streams that have started, including already finished ones.
streamsStarted int64
// Client side: The number of streams that have ended successfully by receiving
// EoS bit set frame from server.
// Server side: The number of streams that have ended successfully by sending
// frame with EoS bit set.
streamsSucceeded int64
streamsFailed int64
// lastStreamCreatedTime stores the timestamp that the last stream gets created. It is of int64 type
// instead of time.Time since it's more costly to atomically update time.Time variable than int64
// variable. The same goes for lastMsgSentTime and lastMsgRecvTime.
lastStreamCreatedTime int64
msgSent int64
msgRecv int64
lastMsgSentTime int64
lastMsgRecvTime int64
}
// ContextErr converts the error from context package into a status error.
func ContextErr(err error) error {
switch err {
case context.DeadlineExceeded:
return status.Error(codes.DeadlineExceeded, err.Error())
case context.Canceled:
return status.Error(codes.Canceled, err.Error())
}
return status.Errorf(codes.Internal, "Unexpected error from context packet: %v", err)
}