diff options
Diffstat (limited to 'vendor/google.golang.org/grpc/transport')
| -rw-r--r-- | vendor/google.golang.org/grpc/transport/bdp_estimator.go | 143 | ||||
| -rw-r--r-- | vendor/google.golang.org/grpc/transport/control.go | 174 | ||||
| -rw-r--r-- | vendor/google.golang.org/grpc/transport/go16.go | 46 | ||||
| -rw-r--r-- | vendor/google.golang.org/grpc/transport/go17.go | 46 | ||||
| -rw-r--r-- | vendor/google.golang.org/grpc/transport/handler_server.go | 116 | ||||
| -rw-r--r-- | vendor/google.golang.org/grpc/transport/http2_client.go | 969 | ||||
| -rw-r--r-- | vendor/google.golang.org/grpc/transport/http2_server.go | 860 | ||||
| -rw-r--r-- | vendor/google.golang.org/grpc/transport/http_util.go | 326 | ||||
| -rw-r--r-- | vendor/google.golang.org/grpc/transport/log.go | 50 | ||||
| -rw-r--r-- | vendor/google.golang.org/grpc/transport/transport.go | 384 |
10 files changed, 1834 insertions, 1280 deletions
diff --git a/vendor/google.golang.org/grpc/transport/bdp_estimator.go b/vendor/google.golang.org/grpc/transport/bdp_estimator.go new file mode 100644 index 0000000..8dd2ed4 --- /dev/null +++ b/vendor/google.golang.org/grpc/transport/bdp_estimator.go @@ -0,0 +1,143 @@ +/* + * + * 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. + bdpLimit = (1 << 20) * 4 + // 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 +) + +var ( + // Adding arbitrary data to ping so that its ack can be + // identified. + // Easter-egg: what does the ping message say? + 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() +} diff --git a/vendor/google.golang.org/grpc/transport/control.go b/vendor/google.golang.org/grpc/transport/control.go index 8d29aee..dd1a8d4 100644 --- a/vendor/google.golang.org/grpc/transport/control.go +++ b/vendor/google.golang.org/grpc/transport/control.go @@ -1,33 +1,18 @@ /* * - * Copyright 2014, Google Inc. - * All rights reserved. + * Copyright 2014 gRPC authors. * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: + * 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 * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. + * http://www.apache.org/licenses/LICENSE-2.0 * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + * 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. * */ @@ -37,17 +22,18 @@ import ( "fmt" "math" "sync" + "sync/atomic" "time" "golang.org/x/net/http2" + "golang.org/x/net/http2/hpack" ) 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 - initialConnWindowSize = defaultWindowSize * 16 // for a connection + initialWindowSize = defaultWindowSize // for an RPC infinity = time.Duration(math.MaxInt64) defaultClientKeepaliveTime = infinity defaultClientKeepaliveTimeout = time.Duration(20 * time.Second) @@ -58,11 +44,43 @@ const ( defaultServerKeepaliveTime = time.Duration(2 * time.Hour) defaultServerKeepaliveTimeout = time.Duration(20 * time.Second) defaultKeepalivePolicyMinTime = time.Duration(5 * time.Minute) + // max window limit set by HTTP2 Specs. + maxWindowSize = math.MaxInt32 + // defaultLocalSendQuota sets is default value for number of data + // bytes that each stream can schedule before some of it being + // flushed out. + defaultLocalSendQuota = 64 * 1024 ) // 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. + +type headerFrame struct { + streamID uint32 + hf []hpack.HeaderField + endStream bool +} + +func (*headerFrame) item() {} + +type continuationFrame struct { + streamID uint32 + endHeaders bool + headerBlockFragment []byte +} + +type dataFrame struct { + streamID uint32 + endStream bool + d []byte + f func() +} + +func (*dataFrame) item() {} + +func (*continuationFrame) item() {} + type windowUpdate struct { streamID uint32 increment uint32 @@ -87,6 +105,8 @@ func (*resetStream) item() {} type goAway struct { code http2.ErrCode debugData []byte + headsUp bool + closeConn bool } func (*goAway) item() {} @@ -108,8 +128,9 @@ func (*ping) item() {} type quotaPool struct { c chan int - mu sync.Mutex - quota int + mu sync.Mutex + version uint32 + quota int } // newQuotaPool creates a quotaPool which has quota q available to consume. @@ -130,6 +151,10 @@ func newQuotaPool(q int) *quotaPool { func (qb *quotaPool) add(v int) { qb.mu.Lock() defer qb.mu.Unlock() + qb.lockedAdd(v) +} + +func (qb *quotaPool) lockedAdd(v int) { select { case n := <-qb.c: qb.quota += n @@ -150,6 +175,35 @@ func (qb *quotaPool) add(v int) { } } +func (qb *quotaPool) addAndUpdate(v int) { + qb.mu.Lock() + defer qb.mu.Unlock() + qb.lockedAdd(v) + // Update the version only after having added to the quota + // so that if acquireWithVesrion sees the new vesrion it is + // guaranteed to have seen the updated quota. + // Also, still keep this inside of the lock, so that when + // compareAndExecute is processing, this function doesn't + // get executed partially (quota gets updated but the version + // doesn't). + atomic.AddUint32(&(qb.version), 1) +} + +func (qb *quotaPool) acquireWithVersion() (<-chan int, uint32) { + return qb.c, atomic.LoadUint32(&(qb.version)) +} + +func (qb *quotaPool) compareAndExecute(version uint32, success, failure func()) bool { + qb.mu.Lock() + defer qb.mu.Unlock() + if version == atomic.LoadUint32(&(qb.version)) { + success() + return true + } + failure() + return false +} + // acquire returns the channel on which available quota amounts are sent. func (qb *quotaPool) acquire() <-chan int { return qb.c @@ -157,16 +211,59 @@ func (qb *quotaPool) acquire() <-chan int { // inFlow deals with inbound flow control type inFlow struct { + mu sync.Mutex // The inbound flow control limit for pending data. limit uint32 - - mu sync.Mutex // 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() + defer f.mu.Unlock() + d := n - f.limit + f.limit = n + return d +} + +func (f *inFlow) maybeAdjust(n uint32) uint32 { + if n > uint32(math.MaxInt32) { + n = uint32(math.MaxInt32) + } + f.mu.Lock() + defer f.mu.Unlock() + // 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 speecified 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 + } + return f.delta + } + return 0 } // onData is invoked when some data frame is received. It updates pendingData. @@ -174,7 +271,7 @@ func (f *inFlow) onData(n uint32) error { f.mu.Lock() defer f.mu.Unlock() f.pendingData += n - if f.pendingData+f.pendingUpdate > f.limit { + if f.pendingData+f.pendingUpdate > f.limit+f.delta { return fmt.Errorf("received %d-bytes data exceeding the limit %d bytes", f.pendingData+f.pendingUpdate, f.limit) } return nil @@ -189,6 +286,13 @@ func (f *inFlow) onRead(n uint32) uint32 { 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 @@ -198,10 +302,10 @@ func (f *inFlow) onRead(n uint32) uint32 { return 0 } -func (f *inFlow) resetPendingData() uint32 { +func (f *inFlow) resetPendingUpdate() uint32 { f.mu.Lock() defer f.mu.Unlock() - n := f.pendingData - f.pendingData = 0 + n := f.pendingUpdate + f.pendingUpdate = 0 return n } diff --git a/vendor/google.golang.org/grpc/transport/go16.go b/vendor/google.golang.org/grpc/transport/go16.go deleted file mode 100644 index ee1c46b..0000000 --- a/vendor/google.golang.org/grpc/transport/go16.go +++ /dev/null @@ -1,46 +0,0 @@ -// +build go1.6,!go1.7 - -/* - * Copyright 2016, Google Inc. - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - * - */ - -package transport - -import ( - "net" - - "golang.org/x/net/context" -) - -// dialContext connects to the address on the named network. -func dialContext(ctx context.Context, network, address string) (net.Conn, error) { - return (&net.Dialer{Cancel: ctx.Done()}).Dial(network, address) -} diff --git a/vendor/google.golang.org/grpc/transport/go17.go b/vendor/google.golang.org/grpc/transport/go17.go deleted file mode 100644 index 356f13f..0000000 --- a/vendor/google.golang.org/grpc/transport/go17.go +++ /dev/null @@ -1,46 +0,0 @@ -// +build go1.7 - -/* - * Copyright 2016, Google Inc. - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - * - */ - -package transport - -import ( - "net" - - "golang.org/x/net/context" -) - -// dialContext connects to the address on the named network. -func dialContext(ctx context.Context, network, address string) (net.Conn, error) { - return (&net.Dialer{}).DialContext(ctx, network, address) -} diff --git a/vendor/google.golang.org/grpc/transport/handler_server.go b/vendor/google.golang.org/grpc/transport/handler_server.go index 28c9ce0..f1f6caf 100644 --- a/vendor/google.golang.org/grpc/transport/handler_server.go +++ b/vendor/google.golang.org/grpc/transport/handler_server.go @@ -1,32 +1,18 @@ /* - * Copyright 2016, Google Inc. - * All rights reserved. * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: + * Copyright 2016 gRPC authors. * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. + * 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 * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + * 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. * */ @@ -47,6 +33,7 @@ import ( "sync" "time" + "github.com/golang/protobuf/proto" "golang.org/x/net/context" "golang.org/x/net/http2" "google.golang.org/grpc/codes" @@ -102,14 +89,9 @@ func NewServerHandlerTransport(w http.ResponseWriter, r *http.Request) (ServerTr continue } for _, v := range vv { - if k == "user-agent" { - // user-agent is special. Copying logic of http_util.go. - if i := strings.LastIndex(v, " "); i == -1 { - // There is no application user agent string being set - continue - } else { - v = v[:i] - } + v, err := decodeMetadataHeader(k, v) + if err != nil { + return nil, streamErrorf(codes.InvalidArgument, "malformed binary metadata: %v", err) } metakv = append(metakv, k, v) } @@ -140,6 +122,11 @@ type serverHandlerTransport struct { // ServeHTTP (HandleStreams) goroutine. The channel is closed // when WriteStatus is called. writes chan func() + + mu sync.Mutex + // streamDone indicates whether WriteStatus has been called and writes channel + // has been closed. + streamDone bool } func (ht *serverHandlerTransport) Close() error { @@ -175,15 +162,28 @@ func (a strAddr) String() string { return string(a) } // do runs fn in the ServeHTTP goroutine. func (ht *serverHandlerTransport) do(fn func()) error { + // Avoid a panic writing to closed channel. Imperfect but maybe good enough. select { - case ht.writes <- fn: - return nil case <-ht.closedCh: return ErrConnClosing + default: + select { + case ht.writes <- fn: + return nil + case <-ht.closedCh: + return ErrConnClosing + } } } func (ht *serverHandlerTransport) WriteStatus(s *Stream, st *status.Status) error { + ht.mu.Lock() + if ht.streamDone { + ht.mu.Unlock() + return nil + } + ht.streamDone = true + ht.mu.Unlock() err := ht.do(func() { ht.writeCommonHeaders(s) @@ -198,7 +198,15 @@ func (ht *serverHandlerTransport) WriteStatus(s *Stream, st *status.Status) erro h.Set("Grpc-Message", encodeGrpcMessage(m)) } - // TODO: Support Grpc-Status-Details-Bin + 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 { @@ -207,10 +215,9 @@ func (ht *serverHandlerTransport) WriteStatus(s *Stream, st *status.Status) erro continue } for _, v := range vv { - // http2 ResponseWriter mechanism to - // send undeclared Trailers after the - // headers have possibly been written. - h.Add(http2.TrailerPrefix+k, v) + // http2 ResponseWriter mechanism to send undeclared Trailers after + // the headers have possibly been written. + h.Add(http2.TrailerPrefix+k, encodeMetadataHeader(k, v)) } } } @@ -238,16 +245,17 @@ func (ht *serverHandlerTransport) writeCommonHeaders(s *Stream) { // and https://golang.org/pkg/net/http/#example_ResponseWriter_trailers h.Add("Trailer", "Grpc-Status") h.Add("Trailer", "Grpc-Message") - // TODO: Support Grpc-Status-Details-Bin + h.Add("Trailer", "Grpc-Status-Details-Bin") if s.sendCompress != "" { h.Set("Grpc-Encoding", s.sendCompress) } } -func (ht *serverHandlerTransport) Write(s *Stream, data []byte, opts *Options) error { +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) if !opts.Delay { ht.rw.(http.Flusher).Flush() @@ -265,6 +273,7 @@ func (ht *serverHandlerTransport) WriteHeader(s *Stream, md metadata.MD) error { continue } for _, v := range vv { + v = encodeMetadataHeader(k, v) h.Add(k, v) } } @@ -305,13 +314,13 @@ func (ht *serverHandlerTransport) HandleStreams(startStream func(*Stream), trace req := ht.req s := &Stream{ - id: 0, // irrelevant - windowHandler: func(int) {}, // nothing - cancel: cancel, - buf: newRecvBuffer(), - st: ht, - method: req.URL.Path, - recvCompress: req.Header.Get("grpc-encoding"), + id: 0, // irrelevant + requestRead: func(int) {}, + cancel: cancel, + buf: newRecvBuffer(), + st: ht, + method: req.URL.Path, + recvCompress: req.Header.Get("grpc-encoding"), } pr := &peer.Peer{ Addr: ht.RemoteAddr(), @@ -322,7 +331,10 @@ func (ht *serverHandlerTransport) HandleStreams(startStream func(*Stream), trace ctx = metadata.NewIncomingContext(ctx, ht.headerMD) ctx = peer.NewContext(ctx, pr) s.ctx = newContextWithStream(ctx, s) - s.dec = &recvBufferReader{ctx: s.ctx, recv: s.buf} + s.trReader = &transportReader{ + reader: &recvBufferReader{ctx: s.ctx, recv: s.buf}, + windowHandler: func(int) {}, + } // readerDone is closed when the Body.Read-ing goroutine exits. readerDone := make(chan struct{}) @@ -334,11 +346,11 @@ func (ht *serverHandlerTransport) HandleStreams(startStream func(*Stream), trace for buf := make([]byte, readSize); ; { n, err := req.Body.Read(buf) if n > 0 { - s.buf.put(&recvMsg{data: buf[:n:n]}) + s.buf.put(recvMsg{data: buf[:n:n]}) buf = buf[n:] } if err != nil { - s.buf.put(&recvMsg{err: mapRecvMsgError(err)}) + s.buf.put(recvMsg{err: mapRecvMsgError(err)}) return } if len(buf) == 0 { diff --git a/vendor/google.golang.org/grpc/transport/http2_client.go b/vendor/google.golang.org/grpc/transport/http2_client.go index 5fc6b75..6ca6cc6 100644 --- a/vendor/google.golang.org/grpc/transport/http2_client.go +++ b/vendor/google.golang.org/grpc/transport/http2_client.go @@ -1,33 +1,18 @@ /* * - * Copyright 2014, Google Inc. - * All rights reserved. + * Copyright 2014 gRPC authors. * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: + * 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 * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. + * http://www.apache.org/licenses/LICENSE-2.0 * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + * 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. * */ @@ -48,7 +33,6 @@ import ( "golang.org/x/net/http2/hpack" "google.golang.org/grpc/codes" "google.golang.org/grpc/credentials" - "google.golang.org/grpc/grpclog" "google.golang.org/grpc/keepalive" "google.golang.org/grpc/metadata" "google.golang.org/grpc/peer" @@ -59,6 +43,7 @@ import ( // http2Client implements the ClientTransport interface with HTTP2. type http2Client struct { ctx context.Context + cancel context.CancelFunc target string // server name/addr userAgent string md interface{} @@ -68,17 +53,6 @@ type http2Client struct { authInfo credentials.AuthInfo // auth info about the connection nextID uint32 // the next stream ID to be used - // writableChan synchronizes write access to the transport. - // A writer acquires the write lock by sending a value on writableChan - // and releases it by receiving from writableChan. - writableChan chan int - // shutdownChan is closed when Close is called. - // Blocking operations should select on shutdownChan to avoid - // blocking forever after Close. - // TODO(zhaoq): Maybe have a channel context? - shutdownChan chan struct{} - // errorChan is closed to notify the I/O error to the caller. - errorChan chan struct{} // goAway is closed to notify the upper layer (i.e., addrConn.transportMonitor) // that the server sent GoAway on this transport. goAway chan struct{} @@ -91,7 +65,7 @@ type http2Client struct { // controlBuf delivers all the control related tasks (e.g., window // updates, reset streams, and various settings) to the controller. - controlBuf *recvBuffer + controlBuf *controlBuffer fc *inFlow // sendQuotaPool provides flow control to outbound message. sendQuotaPool *quotaPool @@ -101,6 +75,8 @@ type http2Client struct { // The scheme used: https if TLS is on, http otherwise. scheme string + isSecure bool + creds []credentials.PerRPCCredentials // Boolean to keep track of reading activity on transport. @@ -110,6 +86,11 @@ type http2Client struct { statsHandler stats.Handler + initialWindowSize int32 + + bdpEst *bdpEstimator + outQuotaVersion uint32 + mu sync.Mutex // guard the following variables state transportState // the state of underlying connection activeStreams map[uint32]*Stream @@ -117,17 +98,18 @@ type http2Client struct { maxStreams int // the per-stream outbound flow control window size set by the peer. streamSendQuota uint32 - // goAwayID records the Last-Stream-ID in the GoAway frame from the server. - goAwayID uint32 // prevGoAway ID records the Last-Stream-ID in the previous GOAway frame. prevGoAwayID uint32 + // goAwayReason records the http2.ErrCode and debug data received with the + // GoAway frame. + goAwayReason GoAwayReason } func dial(ctx context.Context, fn func(context.Context, string) (net.Conn, error), addr string) (net.Conn, error) { if fn != nil { return fn(ctx, addr) } - return dialContext(ctx, "tcp", addr) + return (&net.Dialer{}).DialContext(ctx, "tcp", addr) } func isTemporary(err error) bool { @@ -161,14 +143,23 @@ func isTemporary(err error) bool { // newHTTP2Client constructs a connected ClientTransport to addr based on HTTP2 // and starts to receive messages on it. Non-nil error returns if construction // fails. -func newHTTP2Client(ctx context.Context, addr TargetInfo, opts ConnectOptions) (_ ClientTransport, err error) { +func newHTTP2Client(ctx context.Context, addr TargetInfo, opts ConnectOptions, timeout time.Duration) (_ ClientTransport, err error) { scheme := "http" - conn, err := dial(ctx, opts.Dialer, addr.Addr) + ctx, cancel := context.WithCancel(ctx) + connectCtx, connectCancel := context.WithTimeout(ctx, timeout) + defer func() { + connectCancel() + if err != nil { + cancel() + } + }() + + conn, err := dial(connectCtx, opts.Dialer, addr.Addr) if err != nil { if opts.FailOnNonTempDialError { - return nil, connectionErrorf(isTemporary(err), err, "transport: %v", err) + return nil, connectionErrorf(isTemporary(err), err, "transport: error while dialing: %v", err) } - return nil, connectionErrorf(true, err, "transport: %v", err) + return nil, connectionErrorf(true, err, "transport: Error while dialing %v", err) } // Any further errors will close the underlying connection defer func(conn net.Conn) { @@ -176,16 +167,20 @@ func newHTTP2Client(ctx context.Context, addr TargetInfo, opts ConnectOptions) ( conn.Close() } }(conn) - var authInfo credentials.AuthInfo + var ( + isSecure bool + authInfo credentials.AuthInfo + ) if creds := opts.TransportCredentials; creds != nil { scheme = "https" - conn, authInfo, err = creds.ClientHandshake(ctx, addr.Addr, conn) + conn, authInfo, err = creds.ClientHandshake(connectCtx, addr.Addr, conn) if err != nil { // Credentials handshake errors are typically considered permanent // to avoid retrying on e.g. bad certificates. temp := isTemporary(err) - return nil, connectionErrorf(temp, err, "transport: %v", err) + return nil, connectionErrorf(temp, err, "transport: authentication handshake failed: %v", err) } + isSecure = true } kp := opts.KeepaliveParams // Validate keepalive parameters. @@ -195,9 +190,24 @@ func newHTTP2Client(ctx context.Context, addr TargetInfo, opts ConnectOptions) ( if kp.Timeout == 0 { kp.Timeout = defaultClientKeepaliveTimeout } + dynamicWindow := true + icwz := int32(initialWindowSize) + if opts.InitialConnWindowSize >= defaultWindowSize { + icwz = opts.InitialConnWindowSize + dynamicWindow = false + } var buf bytes.Buffer + writeBufSize := defaultWriteBufSize + if opts.WriteBufferSize > 0 { + writeBufSize = opts.WriteBufferSize + } + readBufSize := defaultReadBufSize + if opts.ReadBufferSize > 0 { + readBufSize = opts.ReadBufferSize + } t := &http2Client{ ctx: ctx, + cancel: cancel, target: addr.Addr, userAgent: opts.UserAgent, md: addr.Metadata, @@ -206,27 +216,36 @@ func newHTTP2Client(ctx context.Context, addr TargetInfo, opts ConnectOptions) ( localAddr: conn.LocalAddr(), authInfo: authInfo, // The client initiated stream id is odd starting from 1. - nextID: 1, - writableChan: make(chan int, 1), - shutdownChan: make(chan struct{}), - errorChan: make(chan struct{}), - goAway: make(chan struct{}), - awakenKeepalive: make(chan struct{}, 1), - framer: newFramer(conn), - hBuf: &buf, - hEnc: hpack.NewEncoder(&buf), - controlBuf: newRecvBuffer(), - fc: &inFlow{limit: initialConnWindowSize}, - sendQuotaPool: newQuotaPool(defaultWindowSize), - scheme: scheme, - state: reachable, - activeStreams: make(map[uint32]*Stream), - creds: opts.PerRPCCredentials, - maxStreams: defaultMaxStreamsClient, - streamsQuota: newQuotaPool(defaultMaxStreamsClient), - streamSendQuota: defaultWindowSize, - kp: kp, - statsHandler: opts.StatsHandler, + nextID: 1, + goAway: make(chan struct{}), + awakenKeepalive: make(chan struct{}, 1), + hBuf: &buf, + hEnc: hpack.NewEncoder(&buf), + framer: newFramer(conn, writeBufSize, readBufSize), + controlBuf: newControlBuffer(), + fc: &inFlow{limit: uint32(icwz)}, + sendQuotaPool: newQuotaPool(defaultWindowSize), + scheme: scheme, + state: reachable, + activeStreams: make(map[uint32]*Stream), + isSecure: isSecure, + creds: opts.PerRPCCredentials, + maxStreams: defaultMaxStreamsClient, + streamsQuota: newQuotaPool(defaultMaxStreamsClient), + streamSendQuota: defaultWindowSize, + kp: kp, + statsHandler: opts.StatsHandler, + initialWindowSize: initialWindowSize, + } + if opts.InitialWindowSize >= defaultWindowSize { + t.initialWindowSize = opts.InitialWindowSize + dynamicWindow = false + } + if dynamicWindow { + t.bdpEst = &bdpEstimator{ + bdp: initialWindowSize, + updateFlowControl: t.updateFlowControl, + } } // Make sure awakenKeepalive can't be written upon. // keepalive routine will make it writable, if need be. @@ -249,65 +268,75 @@ func newHTTP2Client(ctx context.Context, addr TargetInfo, opts ConnectOptions) ( n, err := t.conn.Write(clientPreface) if err != nil { t.Close() - return nil, connectionErrorf(true, err, "transport: %v", err) + return nil, connectionErrorf(true, err, "transport: failed to write client preface: %v", err) } if n != len(clientPreface) { t.Close() return nil, connectionErrorf(true, err, "transport: preface mismatch, wrote %d bytes; want %d", n, len(clientPreface)) } - if initialWindowSize != defaultWindowSize { - err = t.framer.writeSettings(true, http2.Setting{ + if t.initialWindowSize != defaultWindowSize { + err = t.framer.fr.WriteSettings(http2.Setting{ ID: http2.SettingInitialWindowSize, - Val: uint32(initialWindowSize), + Val: uint32(t.initialWindowSize), }) } else { - err = t.framer.writeSettings(true) + err = t.framer.fr.WriteSettings() } if err != nil { t.Close() - return nil, connectionErrorf(true, err, "transport: %v", err) + return nil, connectionErrorf(true, err, "transport: failed to write initial settings frame: %v", err) } // Adjust the connection flow control window if needed. - if delta := uint32(initialConnWindowSize - defaultWindowSize); delta > 0 { - if err := t.framer.writeWindowUpdate(true, 0, delta); err != nil { + if delta := uint32(icwz - defaultWindowSize); delta > 0 { + if err := t.framer.fr.WriteWindowUpdate(0, delta); err != nil { t.Close() - return nil, connectionErrorf(true, err, "transport: %v", err) + return nil, connectionErrorf(true, err, "transport: failed to write window update: %v", err) } } - go t.controller() + t.framer.writer.Flush() + go func() { + loopyWriter(t.ctx, t.controlBuf, t.itemHandler) + t.Close() + }() if t.kp.Time != infinity { go t.keepalive() } - t.writableChan <- 0 return t, nil } func (t *http2Client) newStream(ctx context.Context, callHdr *CallHdr) *Stream { // TODO(zhaoq): Handle uint32 overflow of Stream.id. s := &Stream{ - id: t.nextID, - done: make(chan struct{}), - goAway: make(chan struct{}), - method: callHdr.Method, - sendCompress: callHdr.SendCompress, - buf: newRecvBuffer(), - fc: &inFlow{limit: initialWindowSize}, - sendQuotaPool: newQuotaPool(int(t.streamSendQuota)), - headerChan: make(chan struct{}), + id: t.nextID, + done: make(chan struct{}), + goAway: make(chan struct{}), + method: callHdr.Method, + sendCompress: callHdr.SendCompress, + buf: newRecvBuffer(), + fc: &inFlow{limit: uint32(t.initialWindowSize)}, + sendQuotaPool: newQuotaPool(int(t.streamSendQuota)), + localSendQuota: newQuotaPool(defaultLocalSendQuota), + headerChan: make(chan struct{}), } t.nextID += 2 - s.windowHandler = func(n int) { - t.updateWindow(s, uint32(n)) + s.requestRead = func(n int) { + t.adjustWindow(s, uint32(n)) } // The client side stream context should have exactly the same life cycle with the user provided context. // That means, s.ctx should be read-only. And s.ctx is done iff ctx is done. // So we use the original context here instead of creating a copy. s.ctx = ctx - s.dec = &recvBufferReader{ - ctx: s.ctx, - goAway: s.goAway, - recv: s.buf, + s.trReader = &transportReader{ + reader: &recvBufferReader{ + ctx: s.ctx, + goAway: s.goAway, + recv: s.buf, + }, + windowHandler: func(n int) { + t.updateWindow(s, uint32(n)) + }, } + return s } @@ -321,31 +350,51 @@ func (t *http2Client) NewStream(ctx context.Context, callHdr *CallHdr) (_ *Strea if t.authInfo != nil { pr.AuthInfo = t.authInfo } - userCtx := ctx ctx = peer.NewContext(ctx, pr) - authData := make(map[string]string) - for _, c := range t.creds { + var ( + authData = make(map[string]string) + audience string + ) + // Create an audience string only if needed. + if len(t.creds) > 0 || callHdr.Creds != nil { // Construct URI required to get auth request metadata. - var port string - if pos := strings.LastIndex(t.target, ":"); pos != -1 { - // Omit port if it is the default one. - if t.target[pos+1:] != "443" { - port = ":" + t.target[pos+1:] - } - } + // Omit port if it is the default one. + host := strings.TrimSuffix(callHdr.Host, ":443") pos := strings.LastIndex(callHdr.Method, "/") if pos == -1 { - return nil, streamErrorf(codes.InvalidArgument, "transport: malformed method name: %q", callHdr.Method) + pos = len(callHdr.Method) } - audience := "https://" + callHdr.Host + port + callHdr.Method[:pos] + audience = "https://" + host + callHdr.Method[:pos] + } + for _, c := range t.creds { data, err := c.GetRequestMetadata(ctx, audience) if err != nil { - return nil, streamErrorf(codes.InvalidArgument, "transport: %v", err) + return nil, streamErrorf(codes.Internal, "transport: %v", err) } for k, v := range data { + // Capital header names are illegal in HTTP/2. + k = strings.ToLower(k) authData[k] = v } } + callAuthData := map[string]string{} + // Check if credentials.PerRPCCredentials were provided via call options. + // Note: if these credentials are provided both via dial options and call + // options, then both sets of credentials will be applied. + if callCreds := callHdr.Creds; callCreds != nil { + if !t.isSecure && callCreds.RequireTransportSecurity() { + return nil, streamErrorf(codes.Unauthenticated, "transport: cannot send secure credentials on an insecure connection") + } + data, err := callCreds.GetRequestMetadata(ctx, audience) + if err != nil { + return nil, streamErrorf(codes.Internal, "transport: %v", err) + } + for k, v := range data { + // Capital header names are illegal in HTTP/2 + k = strings.ToLower(k) + callAuthData[k] = v + } + } t.mu.Lock() if t.activeStreams == nil { t.mu.Unlock() @@ -360,7 +409,7 @@ func (t *http2Client) NewStream(ctx context.Context, callHdr *CallHdr) (_ *Strea return nil, ErrConnClosing } t.mu.Unlock() - sq, err := wait(ctx, nil, nil, t.shutdownChan, t.streamsQuota.acquire()) + sq, err := wait(ctx, t.ctx, nil, nil, t.streamsQuota.acquire()) if err != nil { return nil, err } @@ -368,140 +417,108 @@ func (t *http2Client) NewStream(ctx context.Context, callHdr *CallHdr) (_ *Strea if sq > 1 { t.streamsQuota.add(sq - 1) } - if _, err := wait(ctx, nil, nil, t.shutdownChan, t.writableChan); err != nil { - // Return the quota back now because there is no stream returned to the caller. - if _, ok := err.(StreamError); ok { - t.streamsQuota.add(1) - } - return nil, err - } - t.mu.Lock() - if t.state == draining { - t.mu.Unlock() - t.streamsQuota.add(1) - // Need to make t writable again so that the rpc in flight can still proceed. - t.writableChan <- 0 - return nil, ErrStreamDrain - } - if t.state != reachable { - t.mu.Unlock() - return nil, ErrConnClosing - } - s := t.newStream(ctx, callHdr) - s.clientStatsCtx = userCtx - t.activeStreams[s.id] = s - // If the number of active streams change from 0 to 1, then check if keepalive - // has gone dormant. If so, wake it up. - if len(t.activeStreams) == 1 { - select { - case t.awakenKeepalive <- struct{}{}: - t.framer.writePing(false, false, [8]byte{}) - default: - } - } - - t.mu.Unlock() - - // HPACK encodes various headers. Note that once WriteField(...) is - // called, the corresponding headers/continuation frame has to be sent - // because hpack.Encoder is stateful. - t.hBuf.Reset() - t.hEnc.WriteField(hpack.HeaderField{Name: ":method", Value: "POST"}) - t.hEnc.WriteField(hpack.HeaderField{Name: ":scheme", Value: t.scheme}) - t.hEnc.WriteField(hpack.HeaderField{Name: ":path", Value: callHdr.Method}) - t.hEnc.WriteField(hpack.HeaderField{Name: ":authority", Value: callHdr.Host}) - t.hEnc.WriteField(hpack.HeaderField{Name: "content-type", Value: "application/grpc"}) - t.hEnc.WriteField(hpack.HeaderField{Name: "user-agent", Value: t.userAgent}) - t.hEnc.WriteField(hpack.HeaderField{Name: "te", Value: "trailers"}) + // TODO(mmukhi): Benchmark if the perfomance gets better if count the metadata and other header fields + // first and create a slice of that exact size. + // Make the slice of certain predictable size to reduce allocations made by append. + hfLen := 7 // :method, :scheme, :path, :authority, content-type, user-agent, te + hfLen += len(authData) + len(callAuthData) + headerFields := make([]hpack.HeaderField, 0, hfLen) + headerFields = append(headerFields, hpack.HeaderField{Name: ":method", Value: "POST"}) + headerFields = append(headerFields, hpack.HeaderField{Name: ":scheme", Value: t.scheme}) + headerFields = append(headerFields, hpack.HeaderField{Name: ":path", Value: callHdr.Method}) + headerFields = append(headerFields, hpack.HeaderField{Name: ":authority", Value: callHdr.Host}) + headerFields = append(headerFields, hpack.HeaderField{Name: "content-type", Value: "application/grpc"}) + headerFields = append(headerFields, hpack.HeaderField{Name: "user-agent", Value: t.userAgent}) + headerFields = append(headerFields, hpack.HeaderField{Name: "te", Value: "trailers"}) if callHdr.SendCompress != "" { - t.hEnc.WriteField(hpack.HeaderField{Name: "grpc-encoding", Value: callHdr.SendCompress}) + headerFields = append(headerFields, hpack.HeaderField{Name: "grpc-encoding", Value: callHdr.SendCompress}) } if dl, ok := ctx.Deadline(); ok { // Send out timeout regardless its value. The server can detect timeout context by itself. + // TODO(mmukhi): Perhaps this field should be updated when actually writing out to the wire. timeout := dl.Sub(time.Now()) - t.hEnc.WriteField(hpack.HeaderField{Name: "grpc-timeout", Value: encodeTimeout(timeout)}) + headerFields = append(headerFields, hpack.HeaderField{Name: "grpc-timeout", Value: encodeTimeout(timeout)}) } - for k, v := range authData { - // Capital header names are illegal in HTTP/2. - k = strings.ToLower(k) - t.hEnc.WriteField(hpack.HeaderField{Name: k, Value: v}) + headerFields = append(headerFields, hpack.HeaderField{Name: k, Value: encodeMetadataHeader(k, v)}) + } + for k, v := range callAuthData { + headerFields = append(headerFields, hpack.HeaderField{Name: k, Value: encodeMetadataHeader(k, v)}) + } + if b := stats.OutgoingTags(ctx); b != nil { + headerFields = append(headerFields, hpack.HeaderField{Name: "grpc-tags-bin", Value: encodeBinHeader(b)}) + } + if b := stats.OutgoingTrace(ctx); b != nil { + headerFields = append(headerFields, hpack.HeaderField{Name: "grpc-trace-bin", Value: encodeBinHeader(b)}) } - var ( - hasMD bool - endHeaders bool - ) if md, ok := metadata.FromOutgoingContext(ctx); ok { - hasMD = true - for k, v := range md { + for k, vv := range md { // HTTP doesn't allow you to set pseudoheaders after non pseudoheaders were set. if isReservedHeader(k) { continue } - for _, entry := range v { - t.hEnc.WriteField(hpack.HeaderField{Name: k, Value: entry}) + for _, v := range vv { + headerFields = append(headerFields, hpack.HeaderField{Name: k, Value: encodeMetadataHeader(k, v)}) } } } if md, ok := t.md.(*metadata.MD); ok { - for k, v := range *md { + for k, vv := range *md { if isReservedHeader(k) { continue } - for _, entry := range v { - t.hEnc.WriteField(hpack.HeaderField{Name: k, Value: entry}) + for _, v := range vv { + headerFields = append(headerFields, hpack.HeaderField{Name: k, Value: encodeMetadataHeader(k, v)}) } } } - first := true - bufLen := t.hBuf.Len() - // Sends the headers in a single batch even when they span multiple frames. - for !endHeaders { - size := t.hBuf.Len() - if size > http2MaxFrameLen { - size = http2MaxFrameLen - } else { - endHeaders = true - } - var flush bool - if endHeaders && (hasMD || callHdr.Flush) { - flush = true - } - if first { - // Sends a HeadersFrame to server to start a new stream. - p := http2.HeadersFrameParam{ - StreamID: s.id, - BlockFragment: t.hBuf.Next(size), - EndStream: false, - EndHeaders: endHeaders, - } - // Do a force flush for the buffered frames iff it is the last headers frame - // and there is header metadata to be sent. Otherwise, there is flushing until - // the corresponding data frame is written. - err = t.framer.writeHeaders(flush, p) - first = false - } else { - // Sends Continuation frames for the leftover headers. - err = t.framer.writeContinuation(flush, s.id, endHeaders, t.hBuf.Next(size)) - } - if err != nil { - t.notifyError(err) - return nil, connectionErrorf(true, err, "transport: %v", err) + t.mu.Lock() + if t.state == draining { + t.mu.Unlock() + t.streamsQuota.add(1) + return nil, ErrStreamDrain + } + if t.state != reachable { + t.mu.Unlock() + return nil, ErrConnClosing + } + s := t.newStream(ctx, callHdr) + t.activeStreams[s.id] = s + // If the number of active streams change from 0 to 1, then check if keepalive + // has gone dormant. If so, wake it up. + if len(t.activeStreams) == 1 { + select { + case t.awakenKeepalive <- struct{}{}: + t.controlBuf.put(&ping{data: [8]byte{}}) + // Fill the awakenKeepalive channel again as this channel must be + // kept non-writable except at the point that the keepalive() + // goroutine is waiting either to be awaken or shutdown. + t.awakenKeepalive <- struct{}{} + default: } } + t.controlBuf.put(&headerFrame{ + streamID: s.id, + hf: headerFields, + endStream: false, + }) + t.mu.Unlock() + + s.mu.Lock() + s.bytesSent = true + s.mu.Unlock() + if t.statsHandler != nil { outHeader := &stats.OutHeader{ Client: true, - WireLength: bufLen, FullMethod: callHdr.Method, RemoteAddr: t.remoteAddr, LocalAddr: t.localAddr, Compression: callHdr.SendCompress, } - t.statsHandler.HandleRPC(s.clientStatsCtx, outHeader) + t.statsHandler.HandleRPC(s.ctx, outHeader) } - t.writableChan <- 0 return s, nil } @@ -513,6 +530,10 @@ func (t *http2Client) CloseStream(s *Stream, err error) { t.mu.Unlock() return } + if err != nil { + // notify in-flight streams, before the deletion + s.write(recvMsg{err: err}) + } delete(t.activeStreams, s.id) if t.state == draining && len(t.activeStreams) == 0 { // The transport is draining and s is the last live stream on t. @@ -542,11 +563,6 @@ func (t *http2Client) CloseStream(s *Stream, err error) { s.mu.Lock() rstStream = s.rstStream rstError = s.rstError - if q := s.fc.resetPendingData(); q > 0 { - if n := t.fc.onRead(q); n > 0 { - t.controlBuf.put(&windowUpdate{0, n}) - } - } if s.state == streamDone { s.mu.Unlock() return @@ -572,12 +588,9 @@ func (t *http2Client) Close() (err error) { t.mu.Unlock() return } - if t.state == reachable || t.state == draining { - close(t.errorChan) - } t.state = closing t.mu.Unlock() - close(t.shutdownChan) + t.cancel() err = t.conn.Close() t.mu.Lock() streams := t.activeStreams @@ -599,41 +612,18 @@ func (t *http2Client) Close() (err error) { } t.statsHandler.HandleConn(t.ctx, connEnd) } - return + return err } +// GracefulClose sets the state to draining, which prevents new streams from +// being created and causes the transport to be closed when the last active +// stream is closed. If there are no active streams, the transport is closed +// immediately. This does nothing if the transport is already draining or +// closing. func (t *http2Client) GracefulClose() error { t.mu.Lock() switch t.state { - case unreachable: - // The server may close the connection concurrently. t is not available for - // any streams. Close it now. - t.mu.Unlock() - t.Close() - return nil - case closing: - t.mu.Unlock() - return nil - } - // Notify the streams which were initiated after the server sent GOAWAY. - select { - case <-t.goAway: - n := t.prevGoAwayID - if n == 0 && t.nextID > 1 { - n = t.nextID - 2 - } - m := t.goAwayID + 2 - if m == 2 { - m = 1 - } - for i := m; i <= n; i += 2 { - if s, ok := t.activeStreams[i]; ok { - close(s.goAway) - } - } - default: - } - if t.state == draining { + case closing, draining: t.mu.Unlock() return nil } @@ -648,21 +638,38 @@ func (t *http2Client) GracefulClose() error { // Write formats the data into HTTP2 data frame(s) and sends it out. The caller // should proceed only if Write returns nil. -// TODO(zhaoq): opts.Delay is ignored in this implementation. Support it later -// if it improves the performance. -func (t *http2Client) Write(s *Stream, data []byte, opts *Options) error { - r := bytes.NewBuffer(data) - for { - var p []byte - if r.Len() > 0 { +func (t *http2Client) Write(s *Stream, hdr []byte, data []byte, opts *Options) error { + select { + case <-s.ctx.Done(): + return ContextErr(s.ctx.Err()) + case <-t.ctx.Done(): + return ErrConnClosing + default: + } + + if hdr == nil && data == nil && opts.Last { + // stream.CloseSend uses this to send an empty frame with endStream=True + t.controlBuf.put(&dataFrame{streamID: s.id, endStream: true, f: func() {}}) + return nil + } + // Add data to header frame so that we can equally distribute data across frames. + emptyLen := http2MaxFrameLen - len(hdr) + if emptyLen > len(data) { + emptyLen = len(data) + } + hdr = append(hdr, data[:emptyLen]...) + data = data[emptyLen:] + for idx, r := range [][]byte{hdr, data} { + for len(r) > 0 { size := http2MaxFrameLen // Wait until the stream has some quota to send the data. - sq, err := wait(s.ctx, s.done, s.goAway, t.shutdownChan, s.sendQuotaPool.acquire()) + quotaChan, quotaVer := s.sendQuotaPool.acquireWithVersion() + sq, err := wait(s.ctx, t.ctx, s.done, s.goAway, quotaChan) if err != nil { return err } // Wait until the transport has some quota to send the data. - tq, err := wait(s.ctx, s.done, s.goAway, t.shutdownChan, t.sendQuotaPool.acquire()) + tq, err := wait(s.ctx, t.ctx, s.done, s.goAway, t.sendQuotaPool.acquire()) if err != nil { return err } @@ -672,69 +679,51 @@ func (t *http2Client) Write(s *Stream, data []byte, opts *Options) error { if tq < size { size = tq } - p = r.Next(size) - ps := len(p) - if ps < sq { - // Overbooked stream quota. Return it back. - s.sendQuotaPool.add(sq - ps) + if size > len(r) { + size = len(r) } + p := r[:size] + ps := len(p) if ps < tq { // Overbooked transport quota. Return it back. t.sendQuotaPool.add(tq - ps) } - } - var ( - endStream bool - forceFlush bool - ) - if opts.Last && r.Len() == 0 { - endStream = true - } - // Indicate there is a writer who is about to write a data frame. - t.framer.adjustNumWriters(1) - // Got some quota. Try to acquire writing privilege on the transport. - if _, err := wait(s.ctx, s.done, s.goAway, t.shutdownChan, t.writableChan); err != nil { - if _, ok := err.(StreamError); ok || err == io.EOF { - // Return the connection quota back. - t.sendQuotaPool.add(len(p)) + // Acquire local send quota to be able to write to the controlBuf. + ltq, err := wait(s.ctx, t.ctx, s.done, s.goAway, s.localSendQuota.acquire()) + if err != nil { + if _, ok := err.(ConnectionError); !ok { + t.sendQuotaPool.add(ps) + } + return err } - if t.framer.adjustNumWriters(-1) == 0 { - // This writer is the last one in this batch and has the - // responsibility to flush the buffered frames. It queues - // a flush request to controlBuf instead of flushing directly - // in order to avoid the race with other writing or flushing. - t.controlBuf.put(&flushIO{}) + s.localSendQuota.add(ltq - ps) // It's ok if we make it negative. + var endStream bool + // See if this is the last frame to be written. + if opts.Last { + if len(r)-size == 0 { // No more data in r after this iteration. + if idx == 0 { // We're writing data header. + if len(data) == 0 { // There's no data to follow. + endStream = true + } + } else { // We're writing data. + endStream = true + } + } } - return err - } - select { - case <-s.ctx.Done(): - t.sendQuotaPool.add(len(p)) - if t.framer.adjustNumWriters(-1) == 0 { - t.controlBuf.put(&flushIO{}) + success := func() { + t.controlBuf.put(&dataFrame{streamID: s.id, endStream: endStream, d: p, f: func() { s.localSendQuota.add(ps) }}) + if ps < sq { + s.sendQuotaPool.lockedAdd(sq - ps) + } + r = r[ps:] + } + failure := func() { + s.sendQuotaPool.lockedAdd(sq) + } + if !s.sendQuotaPool.compareAndExecute(quotaVer, success, failure) { + t.sendQuotaPool.add(ps) + s.localSendQuota.add(ps) } - t.writableChan <- 0 - return ContextErr(s.ctx.Err()) - default: - } - if r.Len() == 0 && t.framer.adjustNumWriters(0) == 1 { - // Do a force flush iff this is last frame for the entire gRPC message - // and the caller is the only writer at this moment. - forceFlush = true - } - // If WriteData fails, all the pending streams will be handled - // by http2Client.Close(). No explicit CloseStream() needs to be - // invoked. - if err := t.framer.writeData(forceFlush, s.id, endStream, p); err != nil { - t.notifyError(err) - return connectionErrorf(true, err, "transport: %v", err) - } - if t.framer.adjustNumWriters(-1) == 0 { - t.framer.flushWrite() - } - t.writableChan <- 0 - if r.Len() == 0 { - break } } if !opts.Last { @@ -755,6 +744,24 @@ func (t *http2Client) getStream(f http2.Frame) (*Stream, bool) { return s, ok } +// adjustWindow sends out extra window update over the initial window size +// of stream if the application is requesting data larger in size than +// the window. +func (t *http2Client) adjustWindow(s *Stream, n uint32) { + s.mu.Lock() + defer s.mu.Unlock() + if s.state == streamDone { + return + } + if w := s.fc.maybeAdjust(n); w > 0 { + // Piggyback connection's window update along. + if cw := t.fc.resetPendingUpdate(); cw > 0 { + t.controlBuf.put(&windowUpdate{0, cw}) + } + t.controlBuf.put(&windowUpdate{s.id, w}) + } +} + // updateWindow adjusts the inbound quota for the stream and the transport. // Window updates will deliver to the controller for sending when // the cumulative quota exceeds the corresponding threshold. @@ -764,41 +771,76 @@ func (t *http2Client) updateWindow(s *Stream, n uint32) { if s.state == streamDone { return } - if w := t.fc.onRead(n); w > 0 { - t.controlBuf.put(&windowUpdate{0, w}) - } if w := s.fc.onRead(n); w > 0 { + if cw := t.fc.resetPendingUpdate(); cw > 0 { + t.controlBuf.put(&windowUpdate{0, cw}) + } t.controlBuf.put(&windowUpdate{s.id, w}) } } +// updateFlowControl updates the incoming flow control windows +// for the transport and the stream based on the current bdp +// estimation. +func (t *http2Client) updateFlowControl(n uint32) { + t.mu.Lock() + for _, s := range t.activeStreams { + s.fc.newLimit(n) + } + t.initialWindowSize = int32(n) + t.mu.Unlock() + t.controlBuf.put(&windowUpdate{0, t.fc.newLimit(n)}) + t.controlBuf.put(&settings{ + ack: false, + ss: []http2.Setting{ + { + ID: http2.SettingInitialWindowSize, + Val: uint32(n), + }, + }, + }) +} + func (t *http2Client) handleData(f *http2.DataFrame) { size := f.Header().Length - if err := t.fc.onData(uint32(size)); err != nil { - t.notifyError(connectionErrorf(true, err, "%v", err)) - return + var sendBDPPing bool + if t.bdpEst != nil { + sendBDPPing = t.bdpEst.add(uint32(size)) + } + // Decouple connection's flow control from application's read. + // An update on connection's flow control should not depend on + // whether user application has read the data or not. Such a + // restriction is already imposed on the stream's flow control, + // and therefore the sender will be blocked anyways. + // Decoupling the connection flow control will prevent other + // active(fast) streams from starving in presence of slow or + // inactive streams. + // + // Furthermore, if a bdpPing is being sent out we can piggyback + // connection's window update for the bytes we just received. + if sendBDPPing { + if size != 0 { // Could've been an empty data frame. + t.controlBuf.put(&windowUpdate{0, uint32(size)}) + } + t.controlBuf.put(bdpPing) + } else { + if err := t.fc.onData(uint32(size)); err != nil { + t.Close() + return + } + if w := t.fc.onRead(uint32(size)); w > 0 { + t.controlBuf.put(&windowUpdate{0, w}) + } } // Select the right stream to dispatch. 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 { - if f.Header().Flags.Has(http2.FlagDataPadded) { - if w := t.fc.onRead(uint32(size) - uint32(len(f.Data()))); w > 0 { - t.controlBuf.put(&windowUpdate{0, w}) - } - } s.mu.Lock() if s.state == streamDone { s.mu.Unlock() - // The stream has been closed. Release the corresponding quota. - 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 { @@ -854,10 +896,10 @@ func (t *http2Client) handleRSTStream(f *http2.RSTStreamFrame) { } statusCode, ok := http2ErrConvTab[http2.ErrCode(f.ErrCode)] if !ok { - grpclog.Println("transport: http2Client.handleRSTStream found no mapped gRPC status for the received http2 error ", f.ErrCode) + warningf("transport: http2Client.handleRSTStream found no mapped gRPC status for the received http2 error %v", f.ErrCode) statusCode = codes.Unknown } - s.finish(status.Newf(statusCode, "stream terminated by RST_STREAM with error code: %d", f.ErrCode)) + s.finish(status.Newf(statusCode, "stream terminated by RST_STREAM with error code: %v", f.ErrCode)) s.mu.Unlock() s.write(recvMsg{err: io.EOF}) } @@ -876,7 +918,11 @@ func (t *http2Client) handleSettings(f *http2.SettingsFrame) { } func (t *http2Client) handlePing(f *http2.PingFrame) { - if f.IsAck() { // Do nothing. + if f.IsAck() { + // Maybe it's a BDP ping. + if t.bdpEst != nil { + t.bdpEst.calculate(f.Data) + } return } pingAck := &ping{ack: true} @@ -885,35 +931,76 @@ func (t *http2Client) handlePing(f *http2.PingFrame) { } func (t *http2Client) handleGoAway(f *http2.GoAwayFrame) { + t.mu.Lock() + if t.state != reachable && t.state != draining { + t.mu.Unlock() + return + } if f.ErrCode == http2.ErrCodeEnhanceYourCalm { - grpclog.Printf("Client received GoAway with http2.ErrCodeEnhanceYourCalm.") + infof("Client received GoAway with http2.ErrCodeEnhanceYourCalm.") } - t.mu.Lock() - if t.state == reachable || t.state == draining { - if f.LastStreamID > 0 && f.LastStreamID%2 != 1 { - t.mu.Unlock() - t.notifyError(connectionErrorf(true, nil, "received illegal http2 GOAWAY frame: stream ID %d is even", f.LastStreamID)) - return - } - select { - case <-t.goAway: - id := t.goAwayID - // t.goAway has been closed (i.e.,multiple GoAways). - if id < f.LastStreamID { - t.mu.Unlock() - t.notifyError(connectionErrorf(true, nil, "received illegal http2 GOAWAY frame: previously recv GOAWAY frame with LastStramID %d, currently recv %d", id, f.LastStreamID)) - return - } - t.prevGoAwayID = id - t.goAwayID = f.LastStreamID + id := f.LastStreamID + if id > 0 && id%2 != 1 { + t.mu.Unlock() + t.Close() + return + } + // A client can receive multiple GoAways from server (look at https://github.com/grpc/grpc-go/issues/1387). + // The idea is that the first GoAway will be sent with an ID of MaxInt32 and the second GoAway will be sent after an RTT delay + // with the ID of the last stream the server will process. + // Therefore, when we get the first GoAway we don't really close any streams. While in case of second GoAway we + // close all streams created after the second GoAwayId. This way streams that were in-flight while the GoAway from server + // was being sent don't get killed. + select { + case <-t.goAway: // t.goAway has been closed (i.e.,multiple GoAways). + // If there are multiple GoAways the first one should always have an ID greater than the following ones. + if id > t.prevGoAwayID { t.mu.Unlock() + t.Close() return - default: } - t.goAwayID = f.LastStreamID + default: + t.setGoAwayReason(f) close(t.goAway) + t.state = draining } + // All streams with IDs greater than the GoAwayId + // and smaller than the previous GoAway ID should be killed. + upperLimit := t.prevGoAwayID + if upperLimit == 0 { // This is the first GoAway Frame. + upperLimit = math.MaxUint32 // Kill all streams after the GoAway ID. + } + for streamID, stream := range t.activeStreams { + if streamID > id && streamID <= upperLimit { + close(stream.goAway) + } + } + t.prevGoAwayID = id + active := len(t.activeStreams) t.mu.Unlock() + if active == 0 { + t.Close() + } +} + +// setGoAwayReason sets the value of t.goAwayReason based +// on the GoAway frame received. +// It expects a lock on transport's mutext to be held by +// the caller. +func (t *http2Client) setGoAwayReason(f *http2.GoAwayFrame) { + t.goAwayReason = NoReason + switch f.ErrCode { + case http2.ErrCodeEnhanceYourCalm: + if string(f.DebugData()) == "too_many_pings" { + t.goAwayReason = TooManyPings + } + } +} + +func (t *http2Client) GetGoAwayReason() GoAwayReason { + t.mu.Lock() + defer t.mu.Unlock() + return t.goAwayReason } func (t *http2Client) handleWindowUpdate(f *http2.WindowUpdateFrame) { @@ -934,19 +1021,20 @@ func (t *http2Client) operateHeaders(frame *http2.MetaHeadersFrame) { if !ok { return } + s.mu.Lock() + s.bytesReceived = true + s.mu.Unlock() var state decodeState - for _, hf := range frame.Fields { - if err := state.processHeaderField(hf); err != nil { - s.mu.Lock() - if !s.headerDone { - close(s.headerChan) - s.headerDone = true - } - s.mu.Unlock() - s.write(recvMsg{err: err}) - // Something wrong. Stops reading even when there is remaining. - return + if err := state.decodeResponseHeader(frame); err != nil { + s.mu.Lock() + if !s.headerDone { + close(s.headerChan) + s.headerDone = true } + s.mu.Unlock() + s.write(recvMsg{err: err}) + // Something wrong. Stops reading even when there is remaining. + return } endStream := frame.StreamEnded() @@ -958,13 +1046,13 @@ func (t *http2Client) operateHeaders(frame *http2.MetaHeadersFrame) { Client: true, WireLength: int(frame.Header().Length), } - t.statsHandler.HandleRPC(s.clientStatsCtx, inHeader) + t.statsHandler.HandleRPC(s.ctx, inHeader) } else { inTrailer := &stats.InTrailer{ Client: true, WireLength: int(frame.Header().Length), } - t.statsHandler.HandleRPC(s.clientStatsCtx, inTrailer) + t.statsHandler.HandleRPC(s.ctx, inTrailer) } } }() @@ -1012,22 +1100,22 @@ func handleMalformedHTTP2(s *Stream, err error) { // TODO(zhaoq): Check the validity of the incoming frame sequence. func (t *http2Client) reader() { // Check the validity of server preface. - frame, err := t.framer.readFrame() + frame, err := t.framer.fr.ReadFrame() if err != nil { - t.notifyError(err) + t.Close() return } atomic.CompareAndSwapUint32(&t.activity, 0, 1) sf, ok := frame.(*http2.SettingsFrame) if !ok { - t.notifyError(err) + t.Close() return } t.handleSettings(sf) // loop to keep reading incoming messages on this transport. for { - frame, err := t.framer.readFrame() + frame, err := t.framer.fr.ReadFrame() atomic.CompareAndSwapUint32(&t.activity, 0, 1) if err != nil { // Abort an active stream if the http2.Framer returns a @@ -1039,12 +1127,12 @@ func (t *http2Client) reader() { t.mu.Unlock() if s != nil { // use error detail to provide better err message - handleMalformedHTTP2(s, streamErrorf(http2ErrConvTab[se.Code], "%v", t.framer.errorDetail())) + handleMalformedHTTP2(s, streamErrorf(http2ErrConvTab[se.Code], "%v", t.framer.fr.ErrorDetail())) } continue } else { // Transport error. - t.notifyError(err) + t.Close() return } } @@ -1064,7 +1152,7 @@ func (t *http2Client) reader() { case *http2.WindowUpdateFrame: t.handleWindowUpdate(frame) default: - grpclog.Printf("transport: http2Client.reader got unhandled frame type %v.", frame) + errorf("transport: http2Client.reader got unhandled frame type %v.", frame) } } } @@ -1088,7 +1176,7 @@ func (t *http2Client) applySettings(ss []http2.Setting) { t.mu.Lock() for _, stream := range t.activeStreams { // Adjust the sending quota for each stream. - stream.sendQuotaPool.add(int(s.Val - t.streamSendQuota)) + stream.sendQuotaPool.addAndUpdate(int(s.Val) - int(t.streamSendQuota)) } t.streamSendQuota = s.Val t.mu.Unlock() @@ -1096,49 +1184,78 @@ func (t *http2Client) applySettings(ss []http2.Setting) { } } -// controller running in a separate goroutine takes charge of sending control -// frames (e.g., window update, reset stream, setting, etc.) to the server. -func (t *http2Client) controller() { - for { - select { - case i := <-t.controlBuf.get(): - t.controlBuf.load() - select { - case <-t.writableChan: - switch i := i.(type) { - case *windowUpdate: - t.framer.writeWindowUpdate(true, i.streamID, i.increment) - case *settings: - if i.ack { - t.framer.writeSettingsAck(true) - t.applySettings(i.ss) - } else { - t.framer.writeSettings(true, i.ss...) - } - case *resetStream: - // If the server needs to be to intimated about stream closing, - // then we need to make sure the RST_STREAM frame is written to - // the wire before the headers of the next stream waiting on - // streamQuota. We ensure this by adding to the streamsQuota pool - // only after having acquired the writableChan to send RST_STREAM. - t.streamsQuota.add(1) - t.framer.writeRSTStream(true, i.streamID, i.code) - case *flushIO: - t.framer.flushWrite() - case *ping: - t.framer.writePing(true, i.ack, i.data) - default: - grpclog.Printf("transport: http2Client.controller got unexpected item type %v\n", i) - } - t.writableChan <- 0 - continue - case <-t.shutdownChan: - return +// TODO(mmukhi): A lot of this code(and code in other places in the tranpsort layer) +// is duplicated between the client and the server. +// The transport layer needs to be refactored to take care of this. +func (t *http2Client) itemHandler(i item) error { + var err error + switch i := i.(type) { + case *dataFrame: + err = t.framer.fr.WriteData(i.streamID, i.endStream, i.d) + if err == nil { + i.f() + } + case *headerFrame: + t.hBuf.Reset() + for _, f := range i.hf { + t.hEnc.WriteField(f) + } + endHeaders := false + first := true + for !endHeaders { + size := t.hBuf.Len() + if size > http2MaxFrameLen { + size = http2MaxFrameLen + } else { + endHeaders = true } - case <-t.shutdownChan: - return + if first { + first = false + err = t.framer.fr.WriteHeaders(http2.HeadersFrameParam{ + StreamID: i.streamID, + BlockFragment: t.hBuf.Next(size), + EndStream: i.endStream, + EndHeaders: endHeaders, + }) + } else { + err = t.framer.fr.WriteContinuation( + i.streamID, + endHeaders, + t.hBuf.Next(size), + ) + } + if err != nil { + return err + } + } + case *windowUpdate: + err = t.framer.fr.WriteWindowUpdate(i.streamID, i.increment) + case *settings: + if i.ack { + t.applySettings(i.ss) + err = t.framer.fr.WriteSettingsAck() + } else { + err = t.framer.fr.WriteSettings(i.ss...) + } + case *resetStream: + // If the server needs to be to intimated about stream closing, + // then we need to make sure the RST_STREAM frame is written to + // the wire before the headers of the next stream waiting on + // streamQuota. We ensure this by adding to the streamsQuota pool + // only after having acquired the writableChan to send RST_STREAM. + err = t.framer.fr.WriteRSTStream(i.streamID, i.code) + t.streamsQuota.add(1) + case *flushIO: + err = t.framer.writer.Flush() + case *ping: + if !i.ack { + t.bdpEst.timesnap(i.data) } + err = t.framer.fr.WritePing(i.ack, i.data) + default: + errorf("transport: http2Client.controller got unexpected item type %v", i) } + return err } // keepalive running in a separate goroutune makes sure the connection is alive by sending pings. @@ -1162,7 +1279,7 @@ func (t *http2Client) keepalive() { case <-t.awakenKeepalive: // If the control gets here a ping has been sent // need to reset the timer with keepalive.Timeout. - case <-t.shutdownChan: + case <-t.ctx.Done(): return } } else { @@ -1181,13 +1298,13 @@ func (t *http2Client) keepalive() { } t.Close() return - case <-t.shutdownChan: + case <-t.ctx.Done(): if !timer.Stop() { <-timer.C } return } - case <-t.shutdownChan: + case <-t.ctx.Done(): if !timer.Stop() { <-timer.C } @@ -1197,25 +1314,9 @@ func (t *http2Client) keepalive() { } func (t *http2Client) Error() <-chan struct{} { - return t.errorChan + return t.ctx.Done() } func (t *http2Client) GoAway() <-chan struct{} { return t.goAway } - -func (t *http2Client) notifyError(err error) { - t.mu.Lock() - // make sure t.errorChan is closed only once. - if t.state == draining { - t.mu.Unlock() - t.Close() - return - } - if t.state == reachable { - t.state = unreachable - close(t.errorChan) - grpclog.Printf("transport: http2Client.notifyError got notified that the client transport was broken %v.", err) - } - t.mu.Unlock() -} diff --git a/vendor/google.golang.org/grpc/transport/http2_server.go b/vendor/google.golang.org/grpc/transport/http2_server.go index 31fefc7..bad29b8 100644 --- a/vendor/google.golang.org/grpc/transport/http2_server.go +++ b/vendor/google.golang.org/grpc/transport/http2_server.go @@ -1,33 +1,18 @@ /* * - * Copyright 2014, Google Inc. - * All rights reserved. + * Copyright 2014 gRPC authors. * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: + * 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 * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. + * http://www.apache.org/licenses/LICENSE-2.0 * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + * 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. * */ @@ -36,6 +21,7 @@ package transport import ( "bytes" "errors" + "fmt" "io" "math" "math/rand" @@ -51,7 +37,6 @@ import ( "golang.org/x/net/http2/hpack" "google.golang.org/grpc/codes" "google.golang.org/grpc/credentials" - "google.golang.org/grpc/grpclog" "google.golang.org/grpc/keepalive" "google.golang.org/grpc/metadata" "google.golang.org/grpc/peer" @@ -67,35 +52,25 @@ var ErrIllegalHeaderWrite = errors.New("transport: the stream is done or WriteHe // http2Server implements the ServerTransport interface with HTTP2. type http2Server struct { ctx context.Context + cancel context.CancelFunc conn net.Conn remoteAddr net.Addr localAddr net.Addr maxStreamID uint32 // max stream ID ever seen authInfo credentials.AuthInfo // auth info about the connection inTapHandle tap.ServerInHandle - // writableChan synchronizes write access to the transport. - // A writer acquires the write lock by receiving a value on writableChan - // and releases it by sending on writableChan. - writableChan chan int - // shutdownChan is closed when Close is called. - // Blocking operations should select on shutdownChan to avoid - // blocking forever after Close. - shutdownChan chan struct{} - framer *framer - hBuf *bytes.Buffer // the buffer for HPACK encoding - hEnc *hpack.Encoder // HPACK encoder - + framer *framer + hBuf *bytes.Buffer // the buffer for HPACK encoding + hEnc *hpack.Encoder // HPACK encoder // The max number of concurrent streams. maxStreams uint32 // controlBuf delivers all the control related tasks (e.g., window // updates, reset streams, and various settings) to the controller. - controlBuf *recvBuffer + controlBuf *controlBuffer fc *inFlow // sendQuotaPool provides flow control to outbound message. sendQuotaPool *quotaPool - - stats stats.Handler - + stats stats.Handler // Flag to keep track of reading activity on transport. // 1 is true and 0 is false. activity uint32 // Accessed atomically. @@ -111,15 +86,25 @@ type http2Server struct { // Flag to signify that number of ping strikes should be reset to 0. // This is set whenever data or header frames are sent. // 1 means yes. - resetPingStrikes uint32 // Accessed atomically. + resetPingStrikes uint32 // Accessed atomically. + initialWindowSize int32 + bdpEst *bdpEstimator + + mu sync.Mutex // guard the following - mu sync.Mutex // guard the following + // drainChan is initialized when drain(...) is called the first time. + // After which the server writes out the first GoAway(with ID 2^31-1) frame. + // Then an independent goroutine will be launched to later send the second GoAway. + // During this time we don't want to write another first GoAway(with ID 2^31 -1) frame. + // Thus call to drain(...) will be a no-op if drainChan is already initialized since draining is + // already underway. + drainChan chan struct{} state transportState activeStreams map[uint32]*Stream // the per-stream outbound flow control window size set by the peer. streamSendQuota uint32 // idle is the time instant when the connection went idle. - // This is either the begining of the connection or when the number of + // This is either the beginning of the connection or when the number of // RPCs go down to 0. // When the connection is busy, this value is set to 0. idle time.Time @@ -128,31 +113,50 @@ type http2Server struct { // newHTTP2Server constructs a ServerTransport based on HTTP2. ConnectionError is // returned if something goes wrong. func newHTTP2Server(conn net.Conn, config *ServerConfig) (_ ServerTransport, err error) { - framer := newFramer(conn) + writeBufSize := defaultWriteBufSize + if config.WriteBufferSize > 0 { + writeBufSize = config.WriteBufferSize + } + readBufSize := defaultReadBufSize + if config.ReadBufferSize > 0 { + readBufSize = config.ReadBufferSize + } + framer := newFramer(conn, writeBufSize, readBufSize) // Send initial settings as connection preface to client. - var settings []http2.Setting + var isettings []http2.Setting // TODO(zhaoq): Have a better way to signal "no limit" because 0 is // permitted in the HTTP2 spec. maxStreams := config.MaxStreams if maxStreams == 0 { maxStreams = math.MaxUint32 } else { - settings = append(settings, http2.Setting{ + isettings = append(isettings, http2.Setting{ ID: http2.SettingMaxConcurrentStreams, Val: maxStreams, }) } - if initialWindowSize != defaultWindowSize { - settings = append(settings, http2.Setting{ + dynamicWindow := true + iwz := int32(initialWindowSize) + if config.InitialWindowSize >= defaultWindowSize { + iwz = config.InitialWindowSize + dynamicWindow = false + } + icwz := int32(initialWindowSize) + if config.InitialConnWindowSize >= defaultWindowSize { + icwz = config.InitialConnWindowSize + dynamicWindow = false + } + if iwz != defaultWindowSize { + isettings = append(isettings, http2.Setting{ ID: http2.SettingInitialWindowSize, - Val: uint32(initialWindowSize)}) + Val: uint32(iwz)}) } - if err := framer.writeSettings(true, settings...); err != nil { + if err := framer.fr.WriteSettings(isettings...); err != nil { return nil, connectionErrorf(true, err, "transport: %v", err) } // Adjust the connection flow control window if needed. - if delta := uint32(initialConnWindowSize - defaultWindowSize); delta > 0 { - if err := framer.writeWindowUpdate(true, 0, delta); err != nil { + if delta := uint32(icwz - defaultWindowSize); delta > 0 { + if err := framer.fr.WriteWindowUpdate(0, delta); err != nil { return nil, connectionErrorf(true, err, "transport: %v", err) } } @@ -179,29 +183,36 @@ func newHTTP2Server(conn net.Conn, config *ServerConfig) (_ ServerTransport, err kep.MinTime = defaultKeepalivePolicyMinTime } var buf bytes.Buffer + ctx, cancel := context.WithCancel(context.Background()) t := &http2Server{ - ctx: context.Background(), - conn: conn, - remoteAddr: conn.RemoteAddr(), - localAddr: conn.LocalAddr(), - authInfo: config.AuthInfo, - framer: framer, - hBuf: &buf, - hEnc: hpack.NewEncoder(&buf), - maxStreams: maxStreams, - inTapHandle: config.InTapHandle, - controlBuf: newRecvBuffer(), - fc: &inFlow{limit: initialConnWindowSize}, - sendQuotaPool: newQuotaPool(defaultWindowSize), - state: reachable, - writableChan: make(chan int, 1), - shutdownChan: make(chan struct{}), - activeStreams: make(map[uint32]*Stream), - streamSendQuota: defaultWindowSize, - stats: config.StatsHandler, - kp: kp, - idle: time.Now(), - kep: kep, + ctx: ctx, + cancel: cancel, + conn: conn, + remoteAddr: conn.RemoteAddr(), + localAddr: conn.LocalAddr(), + authInfo: config.AuthInfo, + framer: framer, + hBuf: &buf, + hEnc: hpack.NewEncoder(&buf), + maxStreams: maxStreams, + inTapHandle: config.InTapHandle, + controlBuf: newControlBuffer(), + fc: &inFlow{limit: uint32(icwz)}, + sendQuotaPool: newQuotaPool(defaultWindowSize), + state: reachable, + activeStreams: make(map[uint32]*Stream), + streamSendQuota: defaultWindowSize, + stats: config.StatsHandler, + kp: kp, + idle: time.Now(), + kep: kep, + initialWindowSize: iwz, + } + if dynamicWindow { + t.bdpEst = &bdpEstimator{ + bdp: initialWindowSize, + updateFlowControl: t.updateFlowControl, + } } if t.stats != nil { t.ctx = t.stats.TagConn(t.ctx, &stats.ConnTagInfo{ @@ -211,37 +222,43 @@ func newHTTP2Server(conn net.Conn, config *ServerConfig) (_ ServerTransport, err connBegin := &stats.ConnBegin{} t.stats.HandleConn(t.ctx, connBegin) } - go t.controller() + t.framer.writer.Flush() + go func() { + loopyWriter(t.ctx, t.controlBuf, t.itemHandler) + t.Close() + }() go t.keepalive() - t.writableChan <- 0 return t, nil } // operateHeader takes action on the decoded headers. func (t *http2Server) operateHeaders(frame *http2.MetaHeadersFrame, handle func(*Stream), traceCtx func(context.Context, string) context.Context) (close bool) { - buf := newRecvBuffer() - s := &Stream{ - id: frame.Header().StreamID, - st: t, - buf: buf, - fc: &inFlow{limit: initialWindowSize}, - } + streamID := frame.Header().StreamID var state decodeState for _, hf := range frame.Fields { if err := state.processHeaderField(hf); err != nil { if se, ok := err.(StreamError); ok { - t.controlBuf.put(&resetStream{s.id, statusCodeConvTab[se.Code]}) + t.controlBuf.put(&resetStream{streamID, statusCodeConvTab[se.Code]}) } return } } + buf := newRecvBuffer() + s := &Stream{ + id: streamID, + st: t, + buf: buf, + fc: &inFlow{limit: uint32(t.initialWindowSize)}, + recvCompress: state.encoding, + method: state.method, + } + if frame.StreamEnded() { // s is just created by the caller. No lock needed. s.state = streamReadDone } - s.recvCompress = state.encoding if state.timeoutSet { s.ctx, s.cancel = context.WithTimeout(t.ctx, state.timeout) } else { @@ -263,13 +280,12 @@ func (t *http2Server) operateHeaders(frame *http2.MetaHeadersFrame, handle func( if len(state.mdata) > 0 { s.ctx = metadata.NewIncomingContext(s.ctx, state.mdata) } - - s.dec = &recvBufferReader{ - ctx: s.ctx, - recv: s.buf, + if state.statsTags != nil { + s.ctx = stats.SetIncomingTags(s.ctx, state.statsTags) + } + if state.statsTrace != nil { + s.ctx = stats.SetIncomingTrace(s.ctx, state.statsTrace) } - s.recvCompress = state.encoding - s.method = state.method if t.inTapHandle != nil { var err error info := &tap.Info{ @@ -277,7 +293,7 @@ func (t *http2Server) operateHeaders(frame *http2.MetaHeadersFrame, handle func( } s.ctx, err = t.inTapHandle(s.ctx, info) if err != nil { - // TODO: Log the real error. + warningf("transport: http2Server.operateHeaders got an error from InTapHandle: %v", err) t.controlBuf.put(&resetStream{s.id, http2.ErrCodeRefusedStream}) return } @@ -289,24 +305,25 @@ func (t *http2Server) operateHeaders(frame *http2.MetaHeadersFrame, handle func( } if uint32(len(t.activeStreams)) >= t.maxStreams { t.mu.Unlock() - t.controlBuf.put(&resetStream{s.id, http2.ErrCodeRefusedStream}) + t.controlBuf.put(&resetStream{streamID, http2.ErrCodeRefusedStream}) return } - if s.id%2 != 1 || s.id <= t.maxStreamID { + if streamID%2 != 1 || streamID <= t.maxStreamID { t.mu.Unlock() // illegal gRPC stream id. - grpclog.Println("transport: http2Server.HandleStreams received an illegal stream id: ", s.id) + errorf("transport: http2Server.HandleStreams received an illegal stream id: %v", streamID) return true } - t.maxStreamID = s.id + t.maxStreamID = streamID s.sendQuotaPool = newQuotaPool(int(t.streamSendQuota)) - t.activeStreams[s.id] = s + s.localSendQuota = newQuotaPool(defaultLocalSendQuota) + t.activeStreams[streamID] = s if len(t.activeStreams) == 1 { t.idle = time.Time{} } t.mu.Unlock() - s.windowHandler = func(n int) { - t.updateWindow(s, uint32(n)) + s.requestRead = func(n int) { + t.adjustWindow(s, uint32(n)) } s.ctx = traceCtx(s.ctx, s.method) if t.stats != nil { @@ -320,6 +337,15 @@ func (t *http2Server) operateHeaders(frame *http2.MetaHeadersFrame, handle func( } t.stats.HandleRPC(s.ctx, inHeader) } + s.trReader = &transportReader{ + reader: &recvBufferReader{ + ctx: s.ctx, + recv: s.buf, + }, + windowHandler: func(n int) { + t.updateWindow(s, uint32(n)) + }, + } handle(s) return } @@ -331,37 +357,40 @@ func (t *http2Server) HandleStreams(handle func(*Stream), traceCtx func(context. // Check the validity of client preface. 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) + // Only log if it isn't a simple tcp accept check (ie: tcp balancer doing open/close socket) + if err != io.EOF { + errorf("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) + errorf("transport: http2Server.HandleStreams received bogus greeting from client: %q", preface) t.Close() return } - frame, err := t.framer.readFrame() + frame, err := t.framer.fr.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) + errorf("transport: http2Server.HandleStreams failed to read initial settings frame: %v", err) t.Close() return } atomic.StoreUint32(&t.activity, 1) sf, ok := frame.(*http2.SettingsFrame) if !ok { - grpclog.Printf("transport: http2Server.HandleStreams saw invalid preface type %T from client", frame) + errorf("transport: http2Server.HandleStreams saw invalid preface type %T from client", frame) t.Close() return } t.handleSettings(sf) for { - frame, err := t.framer.readFrame() + frame, err := t.framer.fr.ReadFrame() atomic.StoreUint32(&t.activity, 1) if err != nil { if se, ok := err.(http2.StreamError); ok { @@ -378,7 +407,7 @@ func (t *http2Server) HandleStreams(handle func(*Stream), traceCtx func(context. t.Close() return } - grpclog.Printf("transport: http2Server.HandleStreams failed to read frame: %v", err) + warningf("transport: http2Server.HandleStreams failed to read frame: %v", err) t.Close() return } @@ -401,7 +430,7 @@ func (t *http2Server) HandleStreams(handle func(*Stream), traceCtx func(context. case *http2.GoAwayFrame: // TODO: Handle GoAway from the client appropriately. default: - grpclog.Printf("transport: http2Server.HandleStreams found unhandled frame type %v.", frame) + errorf("transport: http2Server.HandleStreams found unhandled frame type %v.", frame) } } } @@ -421,6 +450,23 @@ func (t *http2Server) getStream(f http2.Frame) (*Stream, bool) { return s, true } +// adjustWindow sends out extra window update over the initial window size +// of stream if the application is requesting data larger in size than +// the window. +func (t *http2Server) adjustWindow(s *Stream, n uint32) { + s.mu.Lock() + defer s.mu.Unlock() + if s.state == streamDone { + return + } + if w := s.fc.maybeAdjust(n); w > 0 { + if cw := t.fc.resetPendingUpdate(); cw > 0 { + t.controlBuf.put(&windowUpdate{0, cw}) + } + t.controlBuf.put(&windowUpdate{s.id, w}) + } +} + // updateWindow adjusts the inbound quota for the stream and the transport. // Window updates will deliver to the controller for sending when // the cumulative quota exceeds the corresponding threshold. @@ -430,42 +476,78 @@ func (t *http2Server) updateWindow(s *Stream, n uint32) { if s.state == streamDone { return } - if w := t.fc.onRead(n); w > 0 { - t.controlBuf.put(&windowUpdate{0, w}) - } if w := s.fc.onRead(n); w > 0 { + if cw := t.fc.resetPendingUpdate(); cw > 0 { + t.controlBuf.put(&windowUpdate{0, cw}) + } t.controlBuf.put(&windowUpdate{s.id, w}) } } +// updateFlowControl updates the incoming flow control windows +// for the transport and the stream based on the current bdp +// estimation. +func (t *http2Server) updateFlowControl(n uint32) { + t.mu.Lock() + for _, s := range t.activeStreams { + s.fc.newLimit(n) + } + t.initialWindowSize = int32(n) + t.mu.Unlock() + t.controlBuf.put(&windowUpdate{0, t.fc.newLimit(n)}) + t.controlBuf.put(&settings{ + ack: false, + ss: []http2.Setting{ + { + ID: http2.SettingInitialWindowSize, + Val: uint32(n), + }, + }, + }) + +} + func (t *http2Server) handleData(f *http2.DataFrame) { size := f.Header().Length - if err := t.fc.onData(uint32(size)); err != nil { - grpclog.Printf("transport: http2Server %v", err) - t.Close() - return + var sendBDPPing bool + if t.bdpEst != nil { + sendBDPPing = t.bdpEst.add(uint32(size)) + } + // Decouple connection's flow control from application's read. + // An update on connection's flow control should not depend on + // whether user application has read the data or not. Such a + // restriction is already imposed on the stream's flow control, + // and therefore the sender will be blocked anyways. + // Decoupling the connection flow control will prevent other + // active(fast) streams from starving in presence of slow or + // inactive streams. + // + // Furthermore, if a bdpPing is being sent out we can piggyback + // connection's window update for the bytes we just received. + if sendBDPPing { + if size != 0 { // Could be an empty frame. + t.controlBuf.put(&windowUpdate{0, uint32(size)}) + } + t.controlBuf.put(bdpPing) + } else { + if err := t.fc.onData(uint32(size)); err != nil { + errorf("transport: http2Server %v", err) + t.Close() + return + } + if w := t.fc.onRead(uint32(size)); w > 0 { + t.controlBuf.put(&windowUpdate{0, w}) + } } // Select the right stream to dispatch. 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 { - if f.Header().Flags.Has(http2.FlagDataPadded) { - if w := t.fc.onRead(uint32(size) - uint32(len(f.Data()))); w > 0 { - t.controlBuf.put(&windowUpdate{0, w}) - } - } s.mu.Lock() if s.state == streamDone { s.mu.Unlock() - // The stream has been closed. Release the corresponding quota. - 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 { @@ -517,17 +599,38 @@ func (t *http2Server) handleSettings(f *http2.SettingsFrame) { ss = append(ss, s) return nil }) - // The settings will be applied once the ack is sent. t.controlBuf.put(&settings{ack: true, ss: ss}) } +func (t *http2Server) applySettings(ss []http2.Setting) { + for _, s := range ss { + if s.ID == http2.SettingInitialWindowSize { + t.mu.Lock() + for _, stream := range t.activeStreams { + stream.sendQuotaPool.addAndUpdate(int(s.Val) - int(t.streamSendQuota)) + } + t.streamSendQuota = s.Val + t.mu.Unlock() + } + + } +} + const ( maxPingStrikes = 2 defaultPingTimeout = 2 * time.Hour ) func (t *http2Server) handlePing(f *http2.PingFrame) { - if f.IsAck() { // Do nothing. + if f.IsAck() { + if f.Data == goAwayPing.data && t.drainChan != nil { + close(t.drainChan) + return + } + // Maybe it's a BDP ping. + if t.bdpEst != nil { + t.bdpEst.calculate(f.Data) + } return } pingAck := &ping{ack: true} @@ -550,7 +653,7 @@ func (t *http2Server) handlePing(f *http2.PingFrame) { t.mu.Unlock() if ns < 1 && !t.kep.PermitWithoutStream { // Keepalive shouldn't be active thus, this new ping should - // have come after atleast defaultPingTimeout. + // have come after at least defaultPingTimeout. if t.lastPingAt.Add(defaultPingTimeout).After(now) { t.pingStrikes++ } @@ -563,7 +666,8 @@ func (t *http2Server) handlePing(f *http2.PingFrame) { if t.pingStrikes > maxPingStrikes { // Send goaway and close the connection. - t.controlBuf.put(&goAway{code: http2.ErrCodeEnhanceYourCalm, debugData: []byte("too_many_pings")}) + errorf("transport: Got to too many pings from the client, closing the connection.") + t.controlBuf.put(&goAway{code: http2.ErrCodeEnhanceYourCalm, debugData: []byte("too_many_pings"), closeConn: true}) } } @@ -579,47 +683,16 @@ func (t *http2Server) handleWindowUpdate(f *http2.WindowUpdateFrame) { } } -func (t *http2Server) writeHeaders(s *Stream, b *bytes.Buffer, endStream bool) error { - first := true - endHeaders := false - var err error - defer func() { - if err == nil { - // Reset ping strikes when seding headers since that might cause the - // peer to send ping. - atomic.StoreUint32(&t.resetPingStrikes, 1) - } - }() - // Sends the headers in a single batch. - for !endHeaders { - size := t.hBuf.Len() - if size > http2MaxFrameLen { - size = http2MaxFrameLen - } else { - endHeaders = true - } - if first { - p := http2.HeadersFrameParam{ - StreamID: s.id, - BlockFragment: b.Next(size), - EndStream: endStream, - EndHeaders: endHeaders, - } - err = t.framer.writeHeaders(endHeaders, p) - first = false - } else { - err = t.framer.writeContinuation(endHeaders, s.id, endHeaders, b.Next(size)) - } - if err != nil { - t.Close() - return connectionErrorf(true, err, "transport: %v", err) - } - } - return nil -} - // WriteHeader sends the header metedata md back to the client. func (t *http2Server) WriteHeader(s *Stream, md metadata.MD) error { + select { + case <-s.ctx.Done(): + return ContextErr(s.ctx.Err()) + case <-t.ctx.Done(): + return ErrConnClosing + default: + } + s.mu.Lock() if s.headerOk || s.state == streamDone { s.mu.Unlock() @@ -635,35 +708,34 @@ func (t *http2Server) WriteHeader(s *Stream, md metadata.MD) error { } md = s.header s.mu.Unlock() - if _, err := wait(s.ctx, nil, nil, t.shutdownChan, t.writableChan); err != nil { - return err - } - t.hBuf.Reset() - t.hEnc.WriteField(hpack.HeaderField{Name: ":status", Value: "200"}) - t.hEnc.WriteField(hpack.HeaderField{Name: "content-type", Value: "application/grpc"}) + // TODO(mmukhi): Benchmark if the perfomance gets better if count the metadata and other header fields + // first and create a slice of that exact size. + headerFields := make([]hpack.HeaderField, 0, 2) // at least :status, content-type will be there if none else. + headerFields = append(headerFields, hpack.HeaderField{Name: ":status", Value: "200"}) + headerFields = append(headerFields, hpack.HeaderField{Name: "content-type", Value: "application/grpc"}) if s.sendCompress != "" { - t.hEnc.WriteField(hpack.HeaderField{Name: "grpc-encoding", Value: s.sendCompress}) + headerFields = append(headerFields, hpack.HeaderField{Name: "grpc-encoding", Value: s.sendCompress}) } - for k, v := range md { + for k, vv := range md { if isReservedHeader(k) { // Clients don't tolerate reading restricted headers after some non restricted ones were sent. continue } - for _, entry := range v { - t.hEnc.WriteField(hpack.HeaderField{Name: k, Value: entry}) + for _, v := range vv { + headerFields = append(headerFields, hpack.HeaderField{Name: k, Value: encodeMetadataHeader(k, v)}) } } - bufLen := t.hBuf.Len() - if err := t.writeHeaders(s, t.hBuf, false); err != nil { - return err - } + t.controlBuf.put(&headerFrame{ + streamID: s.id, + hf: headerFields, + endStream: false, + }) if t.stats != nil { outHeader := &stats.OutHeader{ - WireLength: bufLen, + //WireLength: // TODO(mmukhi): Revisit this later, if needed. } t.stats.HandleRPC(s.Context(), outHeader) } - t.writableChan <- 0 return nil } @@ -672,6 +744,12 @@ func (t *http2Server) WriteHeader(s *Stream, md metadata.MD) error { // TODO(zhaoq): Now it indicates the end of entire stream. Revisit if early // OK is adopted. func (t *http2Server) WriteStatus(s *Stream, st *status.Status) error { + select { + case <-t.ctx.Done(): + return ErrConnClosing + default: + } + var headersSent, hasHeader bool s.mu.Lock() if s.state == streamDone { @@ -691,20 +769,15 @@ func (t *http2Server) WriteStatus(s *Stream, st *status.Status) error { headersSent = true } - if _, err := wait(s.ctx, nil, nil, t.shutdownChan, t.writableChan); err != nil { - return err - } - t.hBuf.Reset() + // TODO(mmukhi): Benchmark if the perfomance gets better if count the metadata and other header fields + // first and create a slice of that exact size. + headerFields := make([]hpack.HeaderField, 0, 2) // grpc-status and grpc-message will be there if none else. if !headersSent { - t.hEnc.WriteField(hpack.HeaderField{Name: ":status", Value: "200"}) - t.hEnc.WriteField(hpack.HeaderField{Name: "content-type", Value: "application/grpc"}) + headerFields = append(headerFields, hpack.HeaderField{Name: ":status", Value: "200"}) + headerFields = append(headerFields, hpack.HeaderField{Name: "content-type", Value: "application/grpc"}) } - t.hEnc.WriteField( - hpack.HeaderField{ - Name: "grpc-status", - Value: strconv.Itoa(int(st.Code())), - }) - t.hEnc.WriteField(hpack.HeaderField{Name: "grpc-message", Value: encodeGrpcMessage(st.Message())}) + headerFields = append(headerFields, hpack.HeaderField{Name: "grpc-status", Value: strconv.Itoa(int(st.Code()))}) + headerFields = append(headerFields, hpack.HeaderField{Name: "grpc-message", Value: encodeGrpcMessage(st.Message())}) if p := st.Proto(); p != nil && len(p.Details) > 0 { stBytes, err := proto.Marshal(p) @@ -713,43 +786,42 @@ func (t *http2Server) WriteStatus(s *Stream, st *status.Status) error { panic(err) } - for k, v := range metadata.New(map[string]string{"grpc-status-details-bin": (string)(stBytes)}) { - for _, entry := range v { - t.hEnc.WriteField(hpack.HeaderField{Name: k, Value: entry}) - } - } + headerFields = append(headerFields, hpack.HeaderField{Name: "grpc-status-details-bin", Value: encodeBinHeader(stBytes)}) } // Attach the trailer metadata. - for k, v := range s.trailer { + for k, vv := range s.trailer { // Clients don't tolerate reading restricted headers after some non restricted ones were sent. if isReservedHeader(k) { continue } - for _, entry := range v { - t.hEnc.WriteField(hpack.HeaderField{Name: k, Value: entry}) + for _, v := range vv { + headerFields = append(headerFields, hpack.HeaderField{Name: k, Value: encodeMetadataHeader(k, v)}) } } - bufLen := t.hBuf.Len() - if err := t.writeHeaders(s, t.hBuf, true); err != nil { - t.Close() - return err - } + t.controlBuf.put(&headerFrame{ + streamID: s.id, + hf: headerFields, + endStream: true, + }) if t.stats != nil { - outTrailer := &stats.OutTrailer{ - WireLength: bufLen, - } - t.stats.HandleRPC(s.Context(), outTrailer) + t.stats.HandleRPC(s.Context(), &stats.OutTrailer{}) } t.closeStream(s) - t.writableChan <- 0 return nil } // Write converts the data into HTTP2 data frame and sends it out. Non-nil error // is returns if it fails (e.g., framing error, transport error). -func (t *http2Server) Write(s *Stream, data []byte, opts *Options) (err error) { - // TODO(zhaoq): Support multi-writers for a single stream. +func (t *http2Server) Write(s *Stream, hdr []byte, data []byte, opts *Options) (err error) { + select { + case <-s.ctx.Done(): + return ContextErr(s.ctx.Err()) + case <-t.ctx.Done(): + return ErrConnClosing + default: + } + var writeHeaderFrame bool s.mu.Lock() if s.state == streamDone { @@ -763,107 +835,81 @@ func (t *http2Server) Write(s *Stream, data []byte, opts *Options) (err error) { if writeHeaderFrame { t.WriteHeader(s, nil) } - defer func() { - if err == nil { + // Add data to header frame so that we can equally distribute data across frames. + emptyLen := http2MaxFrameLen - len(hdr) + if emptyLen > len(data) { + emptyLen = len(data) + } + hdr = append(hdr, data[:emptyLen]...) + data = data[emptyLen:] + for _, r := range [][]byte{hdr, data} { + for len(r) > 0 { + size := http2MaxFrameLen + // Wait until the stream has some quota to send the data. + quotaChan, quotaVer := s.sendQuotaPool.acquireWithVersion() + sq, err := wait(s.ctx, t.ctx, nil, nil, quotaChan) + if err != nil { + return err + } + // Wait until the transport has some quota to send the data. + tq, err := wait(s.ctx, t.ctx, nil, nil, t.sendQuotaPool.acquire()) + if err != nil { + return err + } + if sq < size { + size = sq + } + if tq < size { + size = tq + } + if size > len(r) { + size = len(r) + } + p := r[:size] + ps := len(p) + if ps < tq { + // Overbooked transport quota. Return it back. + t.sendQuotaPool.add(tq - ps) + } + // Acquire local send quota to be able to write to the controlBuf. + ltq, err := wait(s.ctx, t.ctx, nil, nil, s.localSendQuota.acquire()) + if err != nil { + if _, ok := err.(ConnectionError); !ok { + t.sendQuotaPool.add(ps) + } + return err + } + s.localSendQuota.add(ltq - ps) // It's ok we make this negative. // Reset ping strikes when sending data since this might cause // the peer to send ping. atomic.StoreUint32(&t.resetPingStrikes, 1) - } - }() - r := bytes.NewBuffer(data) - for { - if r.Len() == 0 { - return nil - } - size := http2MaxFrameLen - // Wait until the stream has some quota to send the data. - sq, err := wait(s.ctx, nil, nil, t.shutdownChan, s.sendQuotaPool.acquire()) - if err != nil { - return err - } - // Wait until the transport has some quota to send the data. - tq, err := wait(s.ctx, nil, nil, t.shutdownChan, t.sendQuotaPool.acquire()) - if err != nil { - return err - } - if sq < size { - size = sq - } - if tq < size { - size = tq - } - p := r.Next(size) - ps := len(p) - if ps < sq { - // Overbooked stream quota. Return it back. - s.sendQuotaPool.add(sq - ps) - } - if ps < tq { - // Overbooked transport quota. Return it back. - t.sendQuotaPool.add(tq - ps) - } - t.framer.adjustNumWriters(1) - // Got some quota. Try to acquire writing privilege on the - // transport. - if _, err := wait(s.ctx, nil, nil, t.shutdownChan, t.writableChan); err != nil { - if _, ok := err.(StreamError); ok { - // Return the connection quota back. - t.sendQuotaPool.add(ps) - } - if t.framer.adjustNumWriters(-1) == 0 { - // This writer is the last one in this batch and has the - // responsibility to flush the buffered frames. It queues - // a flush request to controlBuf instead of flushing directly - // in order to avoid the race with other writing or flushing. - t.controlBuf.put(&flushIO{}) + success := func() { + t.controlBuf.put(&dataFrame{streamID: s.id, endStream: false, d: p, f: func() { + s.localSendQuota.add(ps) + }}) + if ps < sq { + // Overbooked stream quota. Return it back. + s.sendQuotaPool.lockedAdd(sq - ps) + } + r = r[ps:] } - return err - } - select { - case <-s.ctx.Done(): - t.sendQuotaPool.add(ps) - if t.framer.adjustNumWriters(-1) == 0 { - t.controlBuf.put(&flushIO{}) + failure := func() { + s.sendQuotaPool.lockedAdd(sq) } - t.writableChan <- 0 - return ContextErr(s.ctx.Err()) - default: - } - var forceFlush bool - if r.Len() == 0 && t.framer.adjustNumWriters(0) == 1 && !opts.Last { - forceFlush = true - } - if err := t.framer.writeData(forceFlush, s.id, false, p); err != nil { - t.Close() - return connectionErrorf(true, err, "transport: %v", err) - } - if t.framer.adjustNumWriters(-1) == 0 { - t.framer.flushWrite() - } - t.writableChan <- 0 - } - -} - -func (t *http2Server) applySettings(ss []http2.Setting) { - for _, s := range ss { - if s.ID == http2.SettingInitialWindowSize { - t.mu.Lock() - defer t.mu.Unlock() - for _, stream := range t.activeStreams { - stream.sendQuotaPool.add(int(s.Val - t.streamSendQuota)) + if !s.sendQuotaPool.compareAndExecute(quotaVer, success, failure) { + t.sendQuotaPool.add(ps) + s.localSendQuota.add(ps) } - t.streamSendQuota = s.Val } - } + return nil } // keepalive running in a separate goroutine does the following: // 1. Gracefully closes an idle connection after a duration of keepalive.MaxConnectionIdle. // 2. Gracefully closes any connection after a duration of keepalive.MaxConnectionAge. // 3. Forcibly closes a connection after an additive period of keepalive.MaxConnectionAgeGrace over keepalive.MaxConnectionAge. -// 4. Makes sure a connection is alive by sending pings with a frequency of keepalive.Time and closes a non-resposive connection +// 4. Makes sure a connection is alive by sending pings with a frequency of keepalive.Time and closes a non-responsive connection // after an additional duration of keepalive.Timeout. func (t *http2Server) keepalive() { p := &ping{} @@ -872,7 +918,7 @@ func (t *http2Server) keepalive() { maxAge := time.NewTimer(t.kp.MaxConnectionAge) keepalive := time.NewTimer(t.kp.Time) // NOTE: All exit paths of this function should reset their - // respecitve timers. A failure to do so will cause the + // respective timers. A failure to do so will cause the // following clean-up to deadlock and eventually leak. defer func() { if !maxIdle.Stop() { @@ -896,23 +942,18 @@ func (t *http2Server) keepalive() { continue } val := t.kp.MaxConnectionIdle - time.Since(idle) + t.mu.Unlock() if val <= 0 { // The connection has been idle for a duration of keepalive.MaxConnectionIdle or more. // Gracefully close the connection. - t.state = draining - t.mu.Unlock() - t.Drain() + t.drain(http2.ErrCodeNo, []byte{}) // Reseting the timer so that the clean-up doesn't deadlock. maxIdle.Reset(infinity) return } - t.mu.Unlock() maxIdle.Reset(val) case <-maxAge.C: - t.mu.Lock() - t.state = draining - t.mu.Unlock() - t.Drain() + t.drain(http2.ErrCodeNo, []byte{}) maxAge.Reset(t.kp.MaxConnectionAgeGrace) select { case <-maxAge.C: @@ -920,7 +961,7 @@ func (t *http2Server) keepalive() { t.Close() // Reseting the timer so that the clean-up doesn't deadlock. maxAge.Reset(infinity) - case <-t.shutdownChan: + case <-t.ctx.Done(): } return case <-keepalive.C: @@ -938,69 +979,137 @@ func (t *http2Server) keepalive() { pingSent = true t.controlBuf.put(p) keepalive.Reset(t.kp.Timeout) - case <-t.shutdownChan: + case <-t.ctx.Done(): return } } } -// controller running in a separate goroutine takes charge of sending control -// frames (e.g., window update, reset stream, setting, etc.) to the server. -func (t *http2Server) controller() { - for { - select { - case i := <-t.controlBuf.get(): - t.controlBuf.load() +var goAwayPing = &ping{data: [8]byte{1, 6, 1, 8, 0, 3, 3, 9}} + +// TODO(mmukhi): A lot of this code(and code in other places in the tranpsort layer) +// is duplicated between the client and the server. +// The transport layer needs to be refactored to take care of this. +func (t *http2Server) itemHandler(i item) error { + switch i := i.(type) { + case *dataFrame: + if err := t.framer.fr.WriteData(i.streamID, i.endStream, i.d); err != nil { + return err + } + i.f() + return nil + case *headerFrame: + t.hBuf.Reset() + for _, f := range i.hf { + t.hEnc.WriteField(f) + } + first := true + endHeaders := false + for !endHeaders { + size := t.hBuf.Len() + if size > http2MaxFrameLen { + size = http2MaxFrameLen + } else { + endHeaders = true + } + var err error + if first { + first = false + err = t.framer.fr.WriteHeaders(http2.HeadersFrameParam{ + StreamID: i.streamID, + BlockFragment: t.hBuf.Next(size), + EndStream: i.endStream, + EndHeaders: endHeaders, + }) + } else { + err = t.framer.fr.WriteContinuation( + i.streamID, + endHeaders, + t.hBuf.Next(size), + ) + } + if err != nil { + return err + } + } + atomic.StoreUint32(&t.resetPingStrikes, 1) + return nil + case *windowUpdate: + return t.framer.fr.WriteWindowUpdate(i.streamID, i.increment) + case *settings: + if i.ack { + t.applySettings(i.ss) + return t.framer.fr.WriteSettingsAck() + } + return t.framer.fr.WriteSettings(i.ss...) + case *resetStream: + return t.framer.fr.WriteRSTStream(i.streamID, i.code) + case *goAway: + t.mu.Lock() + if t.state == closing { + t.mu.Unlock() + // The transport is closing. + return fmt.Errorf("transport: Connection closing") + } + sid := t.maxStreamID + if !i.headsUp { + // Stop accepting more streams now. + t.state = draining + t.mu.Unlock() + if err := t.framer.fr.WriteGoAway(sid, i.code, i.debugData); err != nil { + return err + } + if i.closeConn { + // Abruptly close the connection following the GoAway (via + // loopywriter). But flush out what's inside the buffer first. + t.framer.writer.Flush() + return fmt.Errorf("transport: Connection closing") + } + return nil + } + t.mu.Unlock() + // For a graceful close, send out a GoAway with stream ID of MaxUInt32, + // Follow that with a ping and wait for the ack to come back or a timer + // to expire. During this time accept new streams since they might have + // originated before the GoAway reaches the client. + // After getting the ack or timer expiration send out another GoAway this + // time with an ID of the max stream server intends to process. + if err := t.framer.fr.WriteGoAway(math.MaxUint32, http2.ErrCodeNo, []byte{}); err != nil { + return err + } + if err := t.framer.fr.WritePing(false, goAwayPing.data); err != nil { + return err + } + go func() { + timer := time.NewTimer(time.Minute) + defer timer.Stop() select { - case <-t.writableChan: - switch i := i.(type) { - case *windowUpdate: - t.framer.writeWindowUpdate(true, i.streamID, i.increment) - case *settings: - if i.ack { - t.framer.writeSettingsAck(true) - t.applySettings(i.ss) - } else { - t.framer.writeSettings(true, i.ss...) - } - case *resetStream: - t.framer.writeRSTStream(true, i.streamID, i.code) - case *goAway: - t.mu.Lock() - if t.state == closing { - t.mu.Unlock() - // The transport is closing. - return - } - sid := t.maxStreamID - t.state = draining - t.mu.Unlock() - t.framer.writeGoAway(true, sid, i.code, i.debugData) - if i.code == http2.ErrCodeEnhanceYourCalm { - t.Close() - } - case *flushIO: - t.framer.flushWrite() - case *ping: - t.framer.writePing(true, i.ack, i.data) - default: - grpclog.Printf("transport: http2Server.controller got unexpected item type %v\n", i) - } - t.writableChan <- 0 - continue - case <-t.shutdownChan: + case <-t.drainChan: + case <-timer.C: + case <-t.ctx.Done(): return } - case <-t.shutdownChan: - return + t.controlBuf.put(&goAway{code: i.code, debugData: i.debugData}) + }() + return nil + case *flushIO: + return t.framer.writer.Flush() + case *ping: + if !i.ack { + t.bdpEst.timesnap(i.data) } + return t.framer.fr.WritePing(i.ack, i.data) + default: + err := status.Errorf(codes.Internal, "transport: http2Server.controller got unexpected item type %t", i) + errorf("%v", err) + return err } } // Close starts shutting down the http2Server transport. // TODO(zhaoq): Now the destruction is not blocked on any pending streams. This // could cause some resource issue. Revisit this later. -func (t *http2Server) Close() (err error) { +func (t *http2Server) Close() error { t.mu.Lock() if t.state == closing { t.mu.Unlock() @@ -1010,8 +1119,8 @@ func (t *http2Server) Close() (err error) { streams := t.activeStreams t.activeStreams = nil t.mu.Unlock() - close(t.shutdownChan) - err = t.conn.Close() + t.cancel() + err := t.conn.Close() // Cancel all active streams. for _, s := range streams { s.cancel() @@ -1020,7 +1129,7 @@ func (t *http2Server) Close() (err error) { connEnd := &stats.ConnEnd{} t.stats.HandleConn(t.ctx, connEnd) } - return + return err } // closeStream clears the footprint of a stream when the stream is not needed @@ -1040,11 +1149,6 @@ func (t *http2Server) closeStream(s *Stream) { // called to interrupt the potential blocking on other goroutines. s.cancel() s.mu.Lock() - if q := s.fc.resetPendingData(); q > 0 { - if w := t.fc.onRead(q); w > 0 { - t.controlBuf.put(&windowUpdate{0, w}) - } - } if s.state == streamDone { s.mu.Unlock() return @@ -1058,7 +1162,17 @@ func (t *http2Server) RemoteAddr() net.Addr { } func (t *http2Server) Drain() { - t.controlBuf.put(&goAway{code: http2.ErrCodeNo}) + t.drain(http2.ErrCodeNo, []byte{}) +} + +func (t *http2Server) drain(code http2.ErrCode, debugData []byte) { + t.mu.Lock() + defer t.mu.Unlock() + if t.drainChan != nil { + return + } + t.drainChan = make(chan struct{}) + t.controlBuf.put(&goAway{code: code, debugData: debugData, headsUp: true}) } var rgen = rand.New(rand.NewSource(time.Now().UnixNano())) diff --git a/vendor/google.golang.org/grpc/transport/http_util.go b/vendor/google.golang.org/grpc/transport/http_util.go index 6adef77..39f878c 100644 --- a/vendor/google.golang.org/grpc/transport/http_util.go +++ b/vendor/google.golang.org/grpc/transport/http_util.go @@ -1,33 +1,18 @@ /* * - * Copyright 2014, Google Inc. - * All rights reserved. + * Copyright 2014 gRPC authors. * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: + * 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 * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. + * http://www.apache.org/licenses/LICENSE-2.0 * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + * 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. * */ @@ -36,12 +21,13 @@ package transport import ( "bufio" "bytes" + "encoding/base64" "fmt" "io" "net" + "net/http" "strconv" "strings" - "sync/atomic" "time" "github.com/golang/protobuf/proto" @@ -49,8 +35,6 @@ import ( "golang.org/x/net/http2/hpack" spb "google.golang.org/genproto/googleapis/rpc/status" "google.golang.org/grpc/codes" - "google.golang.org/grpc/grpclog" - "google.golang.org/grpc/metadata" "google.golang.org/grpc/status" ) @@ -60,7 +44,8 @@ const ( // http://http2.github.io/http2-spec/#SettingValues http2InitHeaderTableSize = 4096 // http2IOBufSize specifies the buffer size for sending frames. - http2IOBufSize = 32 * 1024 + defaultWriteBufSize = 32 * 1024 + defaultReadBufSize = 32 * 1024 ) var ( @@ -88,6 +73,24 @@ var ( codes.ResourceExhausted: http2.ErrCodeEnhanceYourCalm, codes.PermissionDenied: http2.ErrCodeInadequateSecurity, } + 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, + } ) // Records the states during HPACK decoding. Must be reset once the @@ -100,14 +103,17 @@ type decodeState struct { statusGen *status.Status // rawStatusCode and rawStatusMsg are set from the raw trailer fields and are not // intended for direct access outside of parsing. - rawStatusCode int32 + 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 + mdata map[string][]string + statsTags []byte + statsTrace []byte } // isReservedHeader checks whether hdr belongs to HTTP2 headers @@ -159,11 +165,85 @@ func validContentType(t string) bool { func (d *decodeState) status() *status.Status { if d.statusGen == nil { // No status-details were provided; generate status using code/msg. - d.statusGen = status.New(codes.Code(d.rawStatusCode), d.rawStatusMsg) + d.statusGen = status.New(codes.Code(int32(*(d.rawStatusCode))), d.rawStatusMsg) } return d.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) decodeResponseHeader(frame *http2.MetaHeadersFrame) error { + for _, hf := range frame.Fields { + if err := d.processHeaderField(hf); err != nil { + return err + } + } + + // If grpc status exists, no need to check further. + if d.rawStatusCode != nil || d.statusGen != nil { + return nil + } + + // If grpc status doesn't exist and http status doesn't exist, + // then it's a malformed header. + if d.httpStatus == nil { + return streamErrorf(codes.Internal, "malformed header: doesn't contain status(gRPC or HTTP)") + } + + if *(d.httpStatus) != http.StatusOK { + code, ok := httpStatusConvTab[*(d.httpStatus)] + if !ok { + code = codes.Unknown + } + return streamErrorf(code, http.StatusText(*(d.httpStatus))) + } + + // gRPC status doesn't exist and http status is OK. + // Set rawStatusCode to be unknown and return nil error. + // So that, if the stream has ended this Unknown status + // will be propogated to the user. + // Otherwise, it will be ignored. In which case, status from + // a later trailer, that has StreamEnded flag set, is propogated. + code := int(codes.Unknown) + d.rawStatusCode = &code + return nil + +} + +func (d *decodeState) addMetadata(k, v string) { + if d.mdata == nil { + d.mdata = make(map[string][]string) + } + d.mdata[k] = append(d.mdata[k], v) +} + func (d *decodeState) processHeaderField(f hpack.HeaderField) error { switch f.Name { case "content-type": @@ -177,16 +257,16 @@ func (d *decodeState) processHeaderField(f hpack.HeaderField) error { if err != nil { return streamErrorf(codes.Internal, "transport: malformed grpc-status: %v", err) } - d.rawStatusCode = int32(code) + d.rawStatusCode = &code case "grpc-message": d.rawStatusMsg = decodeGrpcMessage(f.Value) case "grpc-status-details-bin": - _, v, err := metadata.DecodeKeyValue("grpc-status-details-bin", f.Value) + v, err := decodeBinHeader(f.Value) if err != nil { return streamErrorf(codes.Internal, "transport: malformed grpc-status-details-bin: %v", err) } s := &spb.Status{} - if err := proto.Unmarshal([]byte(v), s); err != nil { + if err := proto.Unmarshal(v, s); err != nil { return streamErrorf(codes.Internal, "transport: malformed grpc-status-details-bin: %v", err) } d.statusGen = status.FromProto(s) @@ -198,18 +278,36 @@ func (d *decodeState) processHeaderField(f hpack.HeaderField) error { } case ":path": d.method = f.Value + case ":status": + code, err := strconv.Atoi(f.Value) + if err != nil { + return streamErrorf(codes.Internal, "transport: malformed http-status: %v", err) + } + d.httpStatus = &code + case "grpc-tags-bin": + v, err := decodeBinHeader(f.Value) + if err != nil { + return streamErrorf(codes.Internal, "transport: malformed grpc-tags-bin: %v", err) + } + d.statsTags = v + d.addMetadata(f.Name, string(v)) + case "grpc-trace-bin": + v, err := decodeBinHeader(f.Value) + if err != nil { + return streamErrorf(codes.Internal, "transport: malformed grpc-trace-bin: %v", err) + } + d.statsTrace = v + d.addMetadata(f.Name, string(v)) default: - if !isReservedHeader(f.Name) || isWhitelistedPseudoHeader(f.Name) { - if d.mdata == nil { - d.mdata = make(map[string][]string) - } - k, v, err := metadata.DecodeKeyValue(f.Name, f.Value) - if err != nil { - grpclog.Printf("Failed to decode (%q, %q): %v", f.Name, f.Value, err) - return nil - } - d.mdata[k] = append(d.mdata[k], v) + if isReservedHeader(f.Name) && !isWhitelistedPseudoHeader(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 nil + } + d.addMetadata(f.Name, string(v)) } return nil } @@ -377,141 +475,15 @@ type framer struct { fr *http2.Framer } -func newFramer(conn net.Conn) *framer { +func newFramer(conn net.Conn, writeBufferSize, readBufferSize int) *framer { f := &framer{ - reader: bufio.NewReaderSize(conn, http2IOBufSize), - writer: bufio.NewWriterSize(conn, http2IOBufSize), + reader: bufio.NewReaderSize(conn, readBufferSize), + writer: bufio.NewWriterSize(conn, writeBufferSize), } f.fr = http2.NewFramer(f.writer, f.reader) + // 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.ReadMetaHeaders = hpack.NewDecoder(http2InitHeaderTableSize, nil) return f } - -func (f *framer) adjustNumWriters(i int32) int32 { - return atomic.AddInt32(&f.numWriters, i) -} - -// The following writeXXX functions can only be called when the caller gets -// unblocked from writableChan channel (i.e., owns the privilege to write). - -func (f *framer) writeContinuation(forceFlush bool, streamID uint32, endHeaders bool, headerBlockFragment []byte) error { - if err := f.fr.WriteContinuation(streamID, endHeaders, headerBlockFragment); err != nil { - return err - } - if forceFlush { - return f.writer.Flush() - } - return nil -} - -func (f *framer) writeData(forceFlush bool, streamID uint32, endStream bool, data []byte) error { - if err := f.fr.WriteData(streamID, endStream, data); err != nil { - return err - } - if forceFlush { - return f.writer.Flush() - } - return nil -} - -func (f *framer) writeGoAway(forceFlush bool, maxStreamID uint32, code http2.ErrCode, debugData []byte) error { - if err := f.fr.WriteGoAway(maxStreamID, code, debugData); err != nil { - return err - } - if forceFlush { - return f.writer.Flush() - } - return nil -} - -func (f *framer) writeHeaders(forceFlush bool, p http2.HeadersFrameParam) error { - if err := f.fr.WriteHeaders(p); err != nil { - return err - } - if forceFlush { - return f.writer.Flush() - } - return nil -} - -func (f *framer) writePing(forceFlush, ack bool, data [8]byte) error { - if err := f.fr.WritePing(ack, data); err != nil { - return err - } - if forceFlush { - return f.writer.Flush() - } - return nil -} - -func (f *framer) writePriority(forceFlush bool, streamID uint32, p http2.PriorityParam) error { - if err := f.fr.WritePriority(streamID, p); err != nil { - return err - } - if forceFlush { - return f.writer.Flush() - } - return nil -} - -func (f *framer) writePushPromise(forceFlush bool, p http2.PushPromiseParam) error { - if err := f.fr.WritePushPromise(p); err != nil { - return err - } - if forceFlush { - return f.writer.Flush() - } - return nil -} - -func (f *framer) writeRSTStream(forceFlush bool, streamID uint32, code http2.ErrCode) error { - if err := f.fr.WriteRSTStream(streamID, code); err != nil { - return err - } - if forceFlush { - return f.writer.Flush() - } - return nil -} - -func (f *framer) writeSettings(forceFlush bool, settings ...http2.Setting) error { - if err := f.fr.WriteSettings(settings...); err != nil { - return err - } - if forceFlush { - return f.writer.Flush() - } - return nil -} - -func (f *framer) writeSettingsAck(forceFlush bool) error { - if err := f.fr.WriteSettingsAck(); err != nil { - return err - } - if forceFlush { - return f.writer.Flush() - } - return nil -} - -func (f *framer) writeWindowUpdate(forceFlush bool, streamID, incr uint32) error { - if err := f.fr.WriteWindowUpdate(streamID, incr); err != nil { - return err - } - if forceFlush { - return f.writer.Flush() - } - return nil -} - -func (f *framer) flushWrite() error { - return f.writer.Flush() -} - -func (f *framer) readFrame() (http2.Frame, error) { - return f.fr.ReadFrame() -} - -func (f *framer) errorDetail() error { - return f.fr.ErrorDetail() -} diff --git a/vendor/google.golang.org/grpc/transport/log.go b/vendor/google.golang.org/grpc/transport/log.go new file mode 100644 index 0000000..ac8e358 --- /dev/null +++ b/vendor/google.golang.org/grpc/transport/log.go @@ -0,0 +1,50 @@ +/* + * + * 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...) + } +} + +func fatalf(format string, args ...interface{}) { + if grpclog.V(logLevel) { + grpclog.Fatalf(format, args...) + } +} diff --git a/vendor/google.golang.org/grpc/transport/transport.go b/vendor/google.golang.org/grpc/transport/transport.go index b62f702..2cf9bd3 100644 --- a/vendor/google.golang.org/grpc/transport/transport.go +++ b/vendor/google.golang.org/grpc/transport/transport.go @@ -1,48 +1,32 @@ /* * - * Copyright 2014, Google Inc. - * All rights reserved. + * Copyright 2014 gRPC authors. * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: + * 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 * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. + * http://www.apache.org/licenses/LICENSE-2.0 * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + * 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). -*/ +// Package transport defines and implements message oriented communication +// channel to complete various transactions (e.g., an RPC). package transport // import "google.golang.org/grpc/transport" import ( - "bytes" + stdctx "context" "fmt" "io" "net" "sync" + "time" "golang.org/x/net/context" "golang.org/x/net/http2" @@ -65,57 +49,56 @@ type recvMsg struct { err error } -func (*recvMsg) item() {} - -// All items in an out of a recvBuffer should be the same type. -type item interface { - item() -} - -// recvBuffer is an unbounded channel of item. +// 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 item + c chan recvMsg mu sync.Mutex - backlog []item + backlog []recvMsg } func newRecvBuffer() *recvBuffer { b := &recvBuffer{ - c: make(chan item, 1), + c: make(chan recvMsg, 1), } return b } -func (b *recvBuffer) put(r item) { +func (b *recvBuffer) put(r recvMsg) { b.mu.Lock() - defer b.mu.Unlock() 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() - defer b.mu.Unlock() 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 an item in the buffer. +// get returns the channel that receives a recvMsg in the buffer. // -// Upon receipt of an item, the caller should call load to send another -// item onto the channel if there is any. -func (b *recvBuffer) get() <-chan item { +// 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 } @@ -125,7 +108,7 @@ type recvBufferReader struct { ctx context.Context goAway chan struct{} recv *recvBuffer - last *bytes.Reader // Stores the remaining data in the previous calls. + last []byte // Stores the remaining data in the previous calls. err error } @@ -136,27 +119,87 @@ func (r *recvBufferReader) Read(p []byte) (n int, err error) { if r.err != nil { return 0, r.err } - defer func() { r.err = err }() - if r.last != nil && r.last.Len() > 0 { + n, r.err = r.read(p) + return n, r.err +} + +func (r *recvBufferReader) read(p []byte) (n int, err error) { + if r.last != nil && len(r.last) > 0 { // Read remaining data left in last call. - return r.last.Read(p) + copied := copy(p, r.last) + r.last = r.last[copied:] + return copied, nil } select { case <-r.ctx.Done(): return 0, ContextErr(r.ctx.Err()) case <-r.goAway: return 0, ErrStreamDrain - case i := <-r.recv.get(): + case m := <-r.recv.get(): r.recv.load() - m := i.(*recvMsg) if m.err != nil { return 0, m.err } - r.last = bytes.NewReader(m.data) - return r.last.Read(p) + copied := copy(p, m.data) + r.last = m.data[copied:] + return copied, nil } } +// All items in an out of a controlBuffer should be the same type. +type item interface { + item() +} + +// controlBuffer is an unbounded channel of item. +type controlBuffer struct { + c chan item + mu sync.Mutex + backlog []item +} + +func newControlBuffer() *controlBuffer { + b := &controlBuffer{ + c: make(chan item, 1), + } + return b +} + +func (b *controlBuffer) put(r item) { + b.mu.Lock() + 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 *controlBuffer) load() { + b.mu.Lock() + if len(b.backlog) > 0 { + select { + case b.c <- b.backlog[0]: + b.backlog[0] = nil + b.backlog = b.backlog[1:] + default: + } + } + b.mu.Unlock() +} + +// get returns the channel that receives an item in the buffer. +// +// Upon receipt of an item, the caller should call load to send another +// item onto the channel if there is any. +func (b *controlBuffer) get() <-chan item { + return b.c +} + type streamState uint8 const ( @@ -171,11 +214,6 @@ type Stream struct { id uint32 // nil for client side Stream. st ServerTransport - // clientStatsCtx keeps the user context for stats handling. - // It's only valid on client side. Server side stats context is same as s.ctx. - // All client side stats collection should use the clientStatsCtx (instead of the stream context) - // so that all the generated stats for a particular RPC can be associated in the processing phase. - clientStatsCtx context.Context // ctx is the associated context of the stream. ctx context.Context // cancel is always nil for client side Stream. @@ -189,16 +227,20 @@ type Stream struct { recvCompress string sendCompress string buf *recvBuffer - dec io.Reader + trReader io.Reader fc *inFlow recvQuota uint32 + + // TODO: Remote this unused variable. // The accumulated inbound quota pending for window update. updateQuota uint32 - // The handler to control the window update procedure for both this - // particular stream and the associated transport. - windowHandler func(int) - sendQuotaPool *quotaPool + // Callback to state application's intentions to read data. This + // is used to adjust flow control, if need be. + requestRead func(int) + + sendQuotaPool *quotaPool + localSendQuota *quotaPool // Close headerChan to indicate the end of reception of header metadata. headerChan chan struct{} // header caches the received header metadata. @@ -220,6 +262,10 @@ type Stream struct { rstStream bool // rstError is the error that needs to be sent along with the RST_STREAM frame. rstError http2.ErrCode + // bytesSent and bytesReceived indicates whether any bytes have been sent or + // received on this stream. + bytesSent bool + bytesReceived bool } // RecvCompress returns the compression algorithm applied to the inbound @@ -247,16 +293,24 @@ func (s *Stream) GoAway() <-chan struct{} { // Header 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 cancelled/expired. +// header metadata or iii) the stream is canceled/expired. func (s *Stream) Header() (metadata.MD, error) { + var err error select { case <-s.ctx.Done(): - return nil, ContextErr(s.ctx.Err()) + err = ContextErr(s.ctx.Err()) case <-s.goAway: - return nil, ErrStreamDrain + err = ErrStreamDrain + case <-s.headerChan: + return s.header.Copy(), nil + } + // Even if the stream is closed, header is returned if available. + select { case <-s.headerChan: return s.header.Copy(), nil + default: } + return nil, err } // Trailer returns the cached trailer metedata. Note that if it is not called @@ -264,8 +318,9 @@ func (s *Stream) Header() (metadata.MD, error) { // side only. func (s *Stream) Trailer() metadata.MD { s.mu.RLock() - defer s.mu.RUnlock() - return s.trailer.Copy() + c := s.trailer.Copy() + s.mu.RUnlock() + return c } // ServerTransport returns the underlying ServerTransport for the stream. @@ -293,14 +348,16 @@ func (s *Stream) Status() *status.Status { // Server side only. func (s *Stream) SetHeader(md metadata.MD) error { s.mu.Lock() - defer s.mu.Unlock() if s.headerOk || s.state == streamDone { + s.mu.Unlock() return ErrIllegalHeaderWrite } if md.Len() == 0 { + s.mu.Unlock() return nil } s.header = metadata.Join(s.header, md) + s.mu.Unlock() return nil } @@ -311,25 +368,44 @@ func (s *Stream) SetTrailer(md metadata.MD) error { return nil } s.mu.Lock() - defer s.mu.Unlock() s.trailer = metadata.Join(s.trailer, md) + s.mu.Unlock() return nil } func (s *Stream) write(m recvMsg) { - s.buf.put(&m) + s.buf.put(m) } -// Read reads all the data available for this Stream from the transport and +// 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. -func (s *Stream) Read(p []byte) (n int, err error) { - n, err = s.dec.Read(p) +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 } - s.windowHandler(n) + t.windowHandler(n) return } @@ -341,6 +417,22 @@ func (s *Stream) finish(st *status.Status) { close(s.done) } +// BytesSent indicates whether any bytes have been sent on this stream. +func (s *Stream) BytesSent() bool { + s.mu.Lock() + bs := s.bytesSent + s.mu.Unlock() + return bs +} + +// BytesReceived indicates whether any bytes have been received on this stream. +func (s *Stream) BytesReceived() bool { + s.mu.Lock() + br := s.bytesReceived + s.mu.Unlock() + return br +} + // GoString is implemented by Stream so context.String() won't // race when printing %#v. func (s *Stream) GoString() string { @@ -367,19 +459,22 @@ type transportState int const ( reachable transportState = iota - unreachable 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 + 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 } // NewServerTransport creates a ServerTransport with conn or non-nil error @@ -407,6 +502,14 @@ type ConnectOptions struct { 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 } // TargetInfo contains the information of the target such as network address and metadata. @@ -417,8 +520,8 @@ type TargetInfo struct { // NewClientTransport establishes the transport with the required ConnectOptions // and returns it to the caller. -func NewClientTransport(ctx context.Context, target TargetInfo, opts ConnectOptions) (ClientTransport, error) { - return newHTTP2Client(ctx, target, opts) +func NewClientTransport(ctx context.Context, target TargetInfo, opts ConnectOptions, timeout time.Duration) (ClientTransport, error) { + return newHTTP2Client(ctx, target, opts, timeout) } // Options provides additional hints and information for message @@ -430,7 +533,7 @@ type Options struct { // Delay is a hint to the transport implementation for whether // the data could be buffered for a batching write. The - // Transport implementation may ignore the hint. + // transport implementation may ignore the hint. Delay bool } @@ -450,10 +553,15 @@ type CallHdr struct { // outbound message. SendCompress string + // Creds specifies credentials.PerRPCCredentials for a call. + Creds credentials.PerRPCCredentials + // Flush indicates whether a new stream command should be sent // to the peer without waiting for the first data. This is - // only a hint. The transport may modify the flush decision + // only a hint. + // If it's true, the transport may modify the flush decision // for performance purposes. + // If it's false, new stream will never be flushed. Flush bool } @@ -471,7 +579,7 @@ type ClientTransport interface { // 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, data []byte, opts *Options) error + 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) @@ -489,10 +597,13 @@ type ClientTransport interface { // once the transport is initiated. Error() <-chan struct{} - // GoAway returns a channel that is closed when ClientTranspor + // 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 } // ServerTransport is the common interface for all gRPC server-side transport @@ -510,7 +621,7 @@ type ServerTransport interface { // Write sends the data for the given stream. // Write may not be called on all streams. - Write(s *Stream, data []byte, opts *Options) error + 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. @@ -592,41 +703,80 @@ func (e StreamError) Error() string { return fmt.Sprintf("stream error: code = %s desc = %q", e.Code, e.Desc) } +// wait blocks until it can receive from one of the provided contexts or +// channels. ctx is the context of the RPC, tctx is the context of the +// transport, done is a channel closed to indicate the end of the RPC, goAway +// is a channel closed to indicate a GOAWAY was received, and proceed is a +// quota channel, whose received value is returned from this function if none +// of the other signals occur first. +func wait(ctx, tctx context.Context, done, goAway <-chan struct{}, proceed <-chan int) (int, error) { + select { + case <-ctx.Done(): + return 0, ContextErr(ctx.Err()) + case <-done: + return 0, io.EOF + case <-goAway: + return 0, ErrStreamDrain + case <-tctx.Done(): + return 0, ErrConnClosing + case i := <-proceed: + return i, nil + } +} + // ContextErr converts the error from context package into a StreamError. func ContextErr(err error) StreamError { switch err { - case context.DeadlineExceeded: + case context.DeadlineExceeded, stdctx.DeadlineExceeded: return streamErrorf(codes.DeadlineExceeded, "%v", err) - case context.Canceled: + case context.Canceled, stdctx.Canceled: return streamErrorf(codes.Canceled, "%v", err) } - panic(fmt.Sprintf("Unexpected error from context packet: %v", err)) + return streamErrorf(codes.Internal, "Unexpected error from context packet: %v", err) } -// wait blocks until it can receive from ctx.Done, closing, or proceed. -// If it receives from ctx.Done, it returns 0, the StreamError for ctx.Err. -// If it receives from done, it returns 0, io.EOF if ctx is not done; otherwise -// it return the StreamError for ctx.Err. -// If it receives from goAway, it returns 0, ErrStreamDrain. -// If it receives from closing, it returns 0, ErrConnClosing. -// If it receives from proceed, it returns the received integer, nil. -func wait(ctx context.Context, done, goAway, closing <-chan struct{}, proceed <-chan int) (int, error) { - select { - case <-ctx.Done(): - return 0, ContextErr(ctx.Err()) - case <-done: - // User cancellation has precedence. +// GoAwayReason contains the reason for the GoAway frame received. +type GoAwayReason uint8 + +const ( + // Invalid indicates that no GoAway frame is received. + Invalid GoAwayReason = 0 + // NoReason is the default value when GoAway frame is received. + NoReason GoAwayReason = 1 + // TooManyPings indicates that a GoAway frame with ErrCodeEnhanceYourCalm + // was received and that the debug data said "too_many_pings". + TooManyPings GoAwayReason = 2 +) + +// loopyWriter is run in a separate go routine. It is the single code path that will +// write data on wire. +func loopyWriter(ctx context.Context, cbuf *controlBuffer, handler func(item) error) { + for { select { + case i := <-cbuf.get(): + cbuf.load() + if err := handler(i); err != nil { + return + } case <-ctx.Done(): - return 0, ContextErr(ctx.Err()) - default: + return + } + hasData: + for { + select { + case i := <-cbuf.get(): + cbuf.load() + if err := handler(i); err != nil { + return + } + case <-ctx.Done(): + return + default: + if err := handler(&flushIO{}); err != nil { + return + } + break hasData + } } - return 0, io.EOF - case <-goAway: - return 0, ErrStreamDrain - case <-closing: - return 0, ErrConnClosing - case i := <-proceed: - return i, nil } } |