gRPC是基于HTTP/2标准和proto协议开发的,gRPC的很多特性都依赖于HTTP/2标准提供。gRPC设计的四种模式是基于底层HTTP/2的流的概念。transport包是基于HTTP/2标准的实现,提供了流控等特性。
流控
transport提供基于connection和stream的两级流控。
-------------------------------------gRPC流控默认值----------------------------------------------
defaultWindowSize = 65535 //64K
initialWindowSize = defaultWindowSize // for an RPC
initialConnWindowSize = defaultWindowSize * 16 // for a connection
-------------------------------------流控数据结构------------------------------------------------
type inFlow struct {
//流控限制未处理的数据的数量
limit uint32
mu sync.Mutex
//pendingData包含所有收到但未被应用消费的数据
pendingData uint32
//pendingUpdate包含被消费但为发送更新窗口的数量,减少窗口更新的频率
pendingUpdate uint32
}
//真实的流控处理函数,server在接收到client的请求后会先
//检查pendingData+pendingUpdate是否超过limit限制
func (f *inFlow) onData(n uint32) error {
f.mu.Lock()
defer f.mu.Unlock()
f.pendingData += n
if f.pendingData+f.pendingUpdate > f.limit {
return fmt.Errorf("received %d-bytes data exceeding the limit %d bytes", f.pendingData+f.pendingUpdate, f.limit)
}
return nil
}
//http2标准中规定:针对控制类的frame,为了确保能够得到高优先级的处理不做流控。DataFrame的流控处理在如下的函数中进行处理。
----------------------------------server端处理流------------------------------------------------
//server端handleData用于接收dataFrame
func (t *http2Server) handleData(f *http2.DataFrame) {
size := len(f.Data())
//针对connection的流控,如果client和server在该connection的负载大于16 * 64K,server会主动断开与client之间的连接。
if err := t.fc.onData(uint32(size)); err != nil {
//onData函数实现见流控的数据结构
grpclog.Printf("transport: http2Server %v", err)
//超过负载,直接关闭connection
t.Close()
return
}
// 选择正确的流进行处理
s, ok := t.getStream(f)
if !ok {
if w := t.fc.onRead(uint32(size)); w > 0 {
//更新流控窗口的大小
t.controlBuf.put(&windowUpdate{0, w})
}
return
}
if size > 0 {
s.mu.Lock()
if s.state == streamDone {
s.mu.Unlock()
// stream已经被关闭,需要更新流控窗口
if w := t.fc.onRead(uint32(size)); w > 0 {
t.controlBuf.put(&windowUpdate{0, w})
}
return
}
//同一连接上的不同stream具有竞争关系,提供了strean级的流控
if err := s.fc.onData(uint32(size)); err != nil {
//onData()函数实现见流控数据结构
s.mu.Unlock()
//关闭超过流控限制的stream
t.closeStream(s)
//通知client再建立streamID相同的stream
t.controlBuf.put(&resetStream{s.id, http2.ErrCodeFlowControl})
return
}
s.mu.Unlock()
data := make([]byte, size)
copy(data, f.Data())
s.write(recvMsg{data: data})
}
if f.Header().Flags.Has(http2.FlagDataEndStream) {
s.mu.Lock()
if s.state != streamDone {
s.state = streamReadDone
}
s.mu.Unlock()
s.write(recvMsg{err: io.EOF})
}
}
RPC调用的执行过程
以unary模式的rpc调用为例分析一次RPC请求在gRPC中的流转过程,其他三种模式底层调用的函数与unary模式相同(四种模式从底层的HTTP/2分析都是stream,并且仍然是一套request和response的实现)。
注: 以下源码分析部分均是以grpc/example/route_guide为例进行分析。对其他模式感兴趣的读者可自行分析。
unary模式的RPC请求在gRPC中的执行过程
------------------------------------------proto的声明-------------------------------------------
service RouteGuide {
rpc GetFeature(Point) returns (Feature) {}
}
------------------------------------------pb.go源码---------------------------------------------
func (c *routeGuideClient) GetFeature(ctx context.Context, in *Point, opts ...grpc.CallOption) (*Feature, error) {
out := new(Feature)
// -->/routeguide.RouteGuide/GetFeature ->/package/server/method
err := grpc.Invoke(ctx, "/routeguide.RouteGuide/GetFeature", in, out, c.cc, opts...)
if err != nil {
return nil, err
}
return out, nil
}
//以下代码去掉错误处理和非关键函数的调用
//以下代码分析的是grpc client端如何发送request到server
-----------------------------------------grpc-client代码----------------------------------------
func invoke(ctx context.Context, method string, args, reply interface{}, cc *ClientConn, opts ...CallOption) (err error) {
c := defaultCallInfo //构造rpc调用的defaultCallInfo并根据用户传入的信息进行填充
topts := &transport.Options{
Last: true,
Delay: false,
}
for {
var (
err error
t transport.ClientTransport
stream *transport.Stream
put func()
)
//callHdr携带详细的RPC调用信息,如Method->/routeguide.RouteGuide/GetFeature
callHdr := &transport.CallHdr{
Host: cc.authority,
Method: method,
}
gopts := BalancerGetOptions{
BlockingWait: !c.failFast,
}
t, put, err = cc.getTransport(ctx, gopts)
if err != nil {
if _, ok := err.(*rpcError); ok {
return err
}
//非failFast情况下,err为以下两种情况会重试
if err == errConnClosing || err == errConnUnavailable {
if c.failFast {
return Errorf(codes.Unavailable, "%v", err)
}
continue
}
return Errorf(codes.Internal, "%v", err)
}
//将client请求信息发送,并等待server返回
stream, err = sendRequest(ctx, cc.dopts.codec, cc.dopts.cp, callHdr, t, args, topts)
if err != nil {
if put != nil {
put()
put = nil
}
if _, ok := err.(transport.ConnectionError); ok || err == transport.ErrStreamDrain {
if c.failFast {
return toRPCErr(err)
}
continue
}
return toRPCErr(err)
}
//在sendRequest创建的stream上等待server返回response
err = recvResponse(cc.dopts, t, &c, stream, reply)
if err != nil {
if put != nil {
put()
put = nil
}
if _, ok := err.(transport.ConnectionError); ok || err == transport.ErrStreamDrain {
if c.failFast {
return toRPCErr(err)
}
continue
}
return toRPCErr(err)
}
//关闭创建的stream
t.CloseStream(stream, nil)
if put != nil {
put()
put = nil
}
return Errorf(stream.StatusCode(), "%s", stream.StatusDesc())
}
}
----------------------------------------------sendRequest()说明--------------------------------
func sendRequest(ctx context.Context, codec Codec, compressor Compressor, callHdr *transport.CallHdr, t transport.ClientTransport, args interface{}, opts *transport.Options) (_ *transport.Stream, err error) {
//根据callHdr中包含的host和method信息创建对应的stream
//函数具体实现-transport/http2_client.go/http2Client.NewStream()
stream, err := t.NewStream(ctx, callHdr)
//序列化消息并定义消息头
//消息头=5yte=1byte(msg是否压缩) + 4byte(msg长度)
//函数具体实现-rpc_util.go
outBuf, err := encode(codec, args, compressor, cbuf)
//将outBuf按照http2帧的大小分帧并发送到对端,下面会对该函数具体分析
err = t.Write(stream, outBuf, opts)
//发送成功,返回该stream,用于接收response
return stream, nil
}
------------------------------------ClientTransport.Write()说明---------------------------------
//真正将message分帧在指定的stream上传输的函数如下,将对该函数进行详细分析
func (t *http2Client) Write(s *Stream, data []byte, opts *Options) error {
r := bytes.NewBuffer(data)
for {
var p []byte
if r.Len() > 0 {
size := http2MaxFrameLen
s.sendQuotaPool.add(0)
// 等待stream的流控上有配额发送数据,stream.sendQuotaPool=65535
sq, err := wait(s.ctx, s.done, s.goAway, t.shutdownChan, s.sendQuotaPool.acquire())
if err != nil {
return err
}
t.sendQuotaPool.add(0)
// 等待connection的流控有配额去发送数据,t.sendQuotaPool= 65535 * 16
tq, err := wait(s.ctx, s.done, s.goAway, t.shutdownChan, t.sendQuotaPool.acquire())
if err != nil {
if _, ok := err.(StreamError); ok || err == io.EOF {
t.sendQuotaPool.cancel()
}
return err
}
if sq < size {
size = sq
}
if tq < size {
size = tq
}
p = r.Next(size)
ps := len(p)
if ps < sq {
// 返回stream预留超额的配额数量
s.sendQuotaPool.add(sq - ps)
}
if ps < tq {
// 返回connection预留超额的配额数量
t.sendQuotaPool.add(tq - ps)
}
}
var (
endStream bool
forceFlush bool
)
//判断是否为最后一帧l
if opts.Last && r.Len() == 0 {
endStream = true
}
// 表明这将有一个writer将要去写data frame
t.framer.adjustNumWriters(1)
// 释放t.writableChan上加的锁,获得在该transport上写的权利,确保只有一个调用者可以调用t.framer.writeData()函数。
if _, err := wait(s.ctx, s.done, s.goAway, t.shutdownChan, t.writableChan); err != nil {
if _, ok := err.(StreamError); ok || err == io.EOF {
// 释放connection上预留的配额数量
t.sendQuotaPool.add(len(p))
}
if t.framer.adjustNumWriters(-1) == 0 {
// 如果该Writer是这一批的最后一个有责任去刷新http2.frames的缓存区
//将刷新的请求排入一个队列而不是直接刷新合一避免和其他的Writer或者刷新请求的竞争
t.controlBuf.put(&flushIO{})
}
return err
}
select {
case <-s.ctx.Done():
t.sendQuotaPool.add(len(p))
if t.framer.adjustNumWriters(-1) == 0 {
t.controlBuf.put(&flushIO{})
}
//再次为该transport加锁
t.writableChan <- 0
return ContextErr(s.ctx.Err())
default:
}
if r.Len() == 0 && t.framer.adjustNumWriters(0) == 1 {
// 强制刷新因为这是grpc message的最后一个数据帧
//对于调用者来说此刻仅仅只有一个writer
forceFlush = true
}
//如果t.framer.writeData失败,所有等待处理的stream将会在http2Clinet.Close()函数中进行处理,此处不必显示调用CloseStream()
//writeData()不会并发被调用,确保server端收到的frame不会乱序(不会出现dataframe早于headerframe先到)
if err := t.framer.writeData(forceFlush, s.id, endStream, p); err != nil {
//writeData()增加二进制帧的头部,函数实现-net/http2/frame.go
t.notifyError(err)
return connectionErrorf(true, err, "transport: %v", err)
}
if t.framer.adjustNumWriters(-1) == 0 {
t.framer.flushWrite()
}
//再次为该transport加锁
t.writableChan <- 0
if r.Len() == 0 {
break
}
}
if !opts.Last {
return nil
}
s.mu.Lock()
if s.state != streamDone {
//更新stream的状态
s.state = streamWriteDone
}
s.mu.Unlock()
return nil
}
//以下代码是分析grpc-server接收client的请求后内部的处理流程
---------------------------------------grpc-server代码------------------------------------------
//serve函数在net.Listener接收客户端的连接,创建一个新的ServerTransport和service goroutine为每个连接,服务goroutine读取gRPC请求,然后调用server中注册的函数。
func (s *Server) Serve(lis net.Listener) error {
s.lis[lis] = true
for {
rawConn, err := lis.Accept()
if err != nil {
s.mu.Lock()
s.printf("done serving; Accept = %v", err)
s.mu.Unlock()
return err
}
//开始一个单独的goroutine处理client的连接-rawConn
//继续for循环等待其他client的到来
go s.handleRawConn(rawConn)
}
}
//handleRawConn运行在独立的goroutine,并且处理已经接收连接但未执行任何I/O操作的连接
func (s *Server) handleRawConn(rawConn net.Conn) {
conn, authInfo, err := s.useTransportAuthenticator(rawConn)
if err != credentials.ErrConnDispatched {
rawConn.Close()
}
return
}
if s.opts.useHandlerImpl {
s.serveUsingHandler(conn)
} else {
s.serveNewHTTP2Transport(conn, authInfo)
}
}
//serveNewHTTP2Transport建立一个新的HTTP/2 tranport并且为在该transport上的流提供服务
func (s *Server) serveNewHTTP2Transport(c net.Conn, authInfo credentials.AuthInfo) {
//调用transport/http2_server.go
st, err := transport.NewServerTransport("http2", c, 2, authInfo)
if !s.addConn(st) {
st.Close()
return
}
//在transport上接收client发送stream并进行处理的函数
s.serveStreams(st)
}
func (s *Server) serveStreams(st transport.ServerTransport) {
defer s.removeConn(st)
defer st.Close()
var wg sync.WaitGroup
//transport.ServerTranport下的st.HandleStreams处理client发送的stream
st.HandleStreams(func(stream *transport.Stream) {
wg.Add(1)
go func() {
defer wg.Done()
s.handleStream(st, stream, s.traceInfo(st, stream))
}()
})
wg.Wait()
}
----------------------------transport/http2Server.HanleStreams()分析----------------------------
func (t *http2Server) HandleStreams(handle func(*Stream)) {
// 检查client 发送的preface是否合法
preface := make([]byte, len(clientPreface))
if _, err := io.ReadFull(t.conn, preface); err != nil {
grpclog.Printf("transport: http2Server.HandleStreams failed to receive the preface from client: %v", err)
t.Close()
return
}
if !bytes.Equal(preface, clientPreface) {
grpclog.Printf("transport: http2Server.HandleStreams received bogus greeting from client: %q", preface)
t.Close()
return
}
frame, err := t.framer.readFrame()
if err == io.EOF || err == io.ErrUnexpectedEOF {
t.Close()
return
}
if err != nil {
grpclog.Printf("transport: http2Server.HandleStreams failed to read frame: %v", err)
t.Close()
return
}
//读取client发送的SettingFrame
sf, ok := frame.(*http2.SettingsFrame)
if !ok {
grpclog.Printf("transport: http2Server.HandleStreams saw invalid preface type %T from client", frame)
t.Close()
return
}
//根据SettingFrame的内容进行设置
t.handleSettings(sf)
//读取client发送的request内容
for {
frame, err := t.framer.readFrame()
if err != nil {
if se, ok := err.(http2.StreamError); ok {
t.mu.Lock()
s := t.activeStreams[se.StreamID]
t.mu.Unlock()
if s != nil {
t.closeStream(s)
}
t.controlBuf.put(&resetStream{se.StreamID, se.Code})
continue
}
if err == io.EOF || err == io.ErrUnexpectedEOF {
t.Close()
return
}
grpclog.Printf("transport: http2Server.HandleStreams failed to read frame: %v", err)
t.Close()
return
}
switch frame := frame.(type) {
case *http2.MetaHeadersFrame:
//t.operateHeaders函数解码headers内容,并将传输该frame的stream进行记录
//函数实现包括根据stream携带的callHdr信息,如何路由到grpc.Server中注册server具体实现method的过程
//函数实现-transport/http2_server.go operateHeader()函数
if t.operateHeaders(frame, handle) {
t.Close()
break
}
case *http2.DataFrame:
t.handleData(frame)
case *http2.RSTStreamFrame:
t.handleRSTStream(frame)
case *http2.SettingsFrame:
t.handleSettings(frame)
case *http2.PingFrame:
t.handlePing(frame)
case *http2.WindowUpdateFrame:
t.handleWindowUpdate(frame)
case *http2.GoAwayFrame:
default:
grpclog.Printf("transport: http2Server.HandleStreams found unhandled frame type %v.", frame)
}
}
}
func (t *http2Server) operateHeaders(frame *http2.MetaHeadersFrame, handle func(*Stream)) (close bool) {
buf := newRecvBuffer()
//保存client传输的stream信息
s := &Stream{
id: frame.Header().StreamID,
st: t,
buf: buf,
fc: &inFlow{limit: initialWindowSize},
}
var state decodeState
for _, hf := range frame.Fields {
state.processHeaderField(hf)
}
if err := state.err; err != nil {
if se, ok := err.(StreamError); ok {
t.controlBuf.put(&resetStream{s.id, statusCodeConvTab[se.Code]})
}
return
}
if frame.StreamEnded() {
s.state = streamReadDone
}
s.recvCompress = state.encoding
if state.timeoutSet {
s.ctx, s.cancel = context.WithTimeout(context.TODO(), state.timeout)
} else {
s.ctx, s.cancel = context.WithCancel(context.TODO())
}
if uint32(len(t.activeStreams)) >= t.maxStreams {
t.mu.Unlock()
t.controlBuf.put(&resetStream{s.id, http2.ErrCodeRefusedStream})
return
}
//对stream的合法性进行检查
if s.id%2 != 1 || s.id <= t.maxStreamID {
t.mu.Unlock()
grpclog.Println("transport: http2Server.HandleStreams received an illegal stream id: ", s.id)
return true
}
t.maxStreamID = s.id
s.sendQuotaPool = newQuotaPool(int(t.streamSendQuota))
t.activeStreams[s.id] = s
t.mu.Unlock()
s.windowHandler = func(n int) {
t.updateWindow(s, uint32(n))
}
//调用server.go serveStreams()传入的handle去处理server端接收的stream
//handle()会调用server.go handleStream()路由到server端真正实现的函数
handle(s)
return
}
//handleData处理server端接收到数据帧
func (t *http2Server) handleData(f *http2.DataFrame) {
size := len(f.Data())
//检查transport的流控
if err := t.fc.onData(uint32(size)); err != nil {
grpclog.Printf("transport: http2Server %v", err)
t.Close()
return
}
s, ok := t.getStream(f)
if !ok {
if w := t.fc.onRead(uint32(size)); w > 0 {
t.controlBuf.put(&windowUpdate{0, w})
}
return
}
if size > 0 {
s.mu.Lock()
if s.state == streamDone {
s.mu.Unlock()
//检查stream的流控
if w := t.fc.onRead(uint32(size)); w > 0 {
t.controlBuf.put(&windowUpdate{0, w})
}
return
}
if err := s.fc.onData(uint32(size)); err != nil {
s.mu.Unlock()
t.closeStream(s)
t.controlBuf.put(&resetStream{s.id, http2.ErrCodeFlowControl})
return
}
s.mu.Unlock()
data := make([]byte, size)
copy(data, f.Data())
s.write(recvMsg{data: data})
}
if f.Header().Flags.Has(http2.FlagDataEndStream) {
s.mu.Lock()
if s.state != streamDone {
s.state = streamReadDone
}
s.mu.Unlock()
s.write(recvMsg{err: io.EOF})
}
}
以上源码分析一次gRPC调用,从client端如何发送请求到grpc.server端如何路由到server端注册函数的所有过程。
问题总结:
1.grpc的http/2的stream流是如何变化的?
答:unary模式的stream的创建、删除都是由gRPC控制的,剩下的三种模式是将stream的很多操作暴露给用户层,由用户自行控制,但sendRequset和recvResponse的流程和unary模式处理相同。笔者测试发现grpc用到的都是client端的stream,server端的stream在gRPC中并未使用。client端发起的stream都是基数开始的,并且最大值为2^31-1,如果client的streamID超过限制,server端会断开与client的连接。测试结果如下:
//2^31的最大取值2147483648
client stream id 2147483649
2017/08/04 10:44:17 transport: http2Client.notifyError got notified that the client transport was broken invalid stream ID.
2017/08/04 10:44:17 &{0xc4201787e0}.RouteChat(_) = _, rpc error: code = 13 desc = transport: invalid stream ID
exit status 1