Matthias Andreas Benkard | 832a54e | 2019-01-29 09:27:38 +0100 | [diff] [blame] | 1 | /* |
| 2 | * |
| 3 | * Copyright 2014 gRPC authors. |
| 4 | * |
| 5 | * Licensed under the Apache License, Version 2.0 (the "License"); |
| 6 | * you may not use this file except in compliance with the License. |
| 7 | * You may obtain a copy of the License at |
| 8 | * |
| 9 | * http://www.apache.org/licenses/LICENSE-2.0 |
| 10 | * |
| 11 | * Unless required by applicable law or agreed to in writing, software |
| 12 | * distributed under the License is distributed on an "AS IS" BASIS, |
| 13 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 14 | * See the License for the specific language governing permissions and |
| 15 | * limitations under the License. |
| 16 | * |
| 17 | */ |
| 18 | |
| 19 | package transport |
| 20 | |
| 21 | import ( |
| 22 | "fmt" |
| 23 | "math" |
| 24 | "sync" |
| 25 | "sync/atomic" |
| 26 | "time" |
| 27 | ) |
| 28 | |
| 29 | const ( |
| 30 | // The default value of flow control window size in HTTP2 spec. |
| 31 | defaultWindowSize = 65535 |
| 32 | // The initial window size for flow control. |
| 33 | initialWindowSize = defaultWindowSize // for an RPC |
| 34 | infinity = time.Duration(math.MaxInt64) |
| 35 | defaultClientKeepaliveTime = infinity |
| 36 | defaultClientKeepaliveTimeout = 20 * time.Second |
| 37 | defaultMaxStreamsClient = 100 |
| 38 | defaultMaxConnectionIdle = infinity |
| 39 | defaultMaxConnectionAge = infinity |
| 40 | defaultMaxConnectionAgeGrace = infinity |
| 41 | defaultServerKeepaliveTime = 2 * time.Hour |
| 42 | defaultServerKeepaliveTimeout = 20 * time.Second |
| 43 | defaultKeepalivePolicyMinTime = 5 * time.Minute |
| 44 | // max window limit set by HTTP2 Specs. |
| 45 | maxWindowSize = math.MaxInt32 |
| 46 | // defaultWriteQuota is the default value for number of data |
| 47 | // bytes that each stream can schedule before some of it being |
| 48 | // flushed out. |
| 49 | defaultWriteQuota = 64 * 1024 |
| 50 | ) |
| 51 | |
| 52 | // writeQuota is a soft limit on the amount of data a stream can |
| 53 | // schedule before some of it is written out. |
| 54 | type writeQuota struct { |
| 55 | quota int32 |
| 56 | // get waits on read from when quota goes less than or equal to zero. |
| 57 | // replenish writes on it when quota goes positive again. |
| 58 | ch chan struct{} |
| 59 | // done is triggered in error case. |
| 60 | done <-chan struct{} |
| 61 | // replenish is called by loopyWriter to give quota back to. |
| 62 | // It is implemented as a field so that it can be updated |
| 63 | // by tests. |
| 64 | replenish func(n int) |
| 65 | } |
| 66 | |
| 67 | func newWriteQuota(sz int32, done <-chan struct{}) *writeQuota { |
| 68 | w := &writeQuota{ |
| 69 | quota: sz, |
| 70 | ch: make(chan struct{}, 1), |
| 71 | done: done, |
| 72 | } |
| 73 | w.replenish = w.realReplenish |
| 74 | return w |
| 75 | } |
| 76 | |
| 77 | func (w *writeQuota) get(sz int32) error { |
| 78 | for { |
| 79 | if atomic.LoadInt32(&w.quota) > 0 { |
| 80 | atomic.AddInt32(&w.quota, -sz) |
| 81 | return nil |
| 82 | } |
| 83 | select { |
| 84 | case <-w.ch: |
| 85 | continue |
| 86 | case <-w.done: |
| 87 | return errStreamDone |
| 88 | } |
| 89 | } |
| 90 | } |
| 91 | |
| 92 | func (w *writeQuota) realReplenish(n int) { |
| 93 | sz := int32(n) |
| 94 | a := atomic.AddInt32(&w.quota, sz) |
| 95 | b := a - sz |
| 96 | if b <= 0 && a > 0 { |
| 97 | select { |
| 98 | case w.ch <- struct{}{}: |
| 99 | default: |
| 100 | } |
| 101 | } |
| 102 | } |
| 103 | |
| 104 | type trInFlow struct { |
| 105 | limit uint32 |
| 106 | unacked uint32 |
| 107 | effectiveWindowSize uint32 |
| 108 | } |
| 109 | |
| 110 | func (f *trInFlow) newLimit(n uint32) uint32 { |
| 111 | d := n - f.limit |
| 112 | f.limit = n |
| 113 | f.updateEffectiveWindowSize() |
| 114 | return d |
| 115 | } |
| 116 | |
| 117 | func (f *trInFlow) onData(n uint32) uint32 { |
| 118 | f.unacked += n |
| 119 | if f.unacked >= f.limit/4 { |
| 120 | w := f.unacked |
| 121 | f.unacked = 0 |
| 122 | f.updateEffectiveWindowSize() |
| 123 | return w |
| 124 | } |
| 125 | f.updateEffectiveWindowSize() |
| 126 | return 0 |
| 127 | } |
| 128 | |
| 129 | func (f *trInFlow) reset() uint32 { |
| 130 | w := f.unacked |
| 131 | f.unacked = 0 |
| 132 | f.updateEffectiveWindowSize() |
| 133 | return w |
| 134 | } |
| 135 | |
| 136 | func (f *trInFlow) updateEffectiveWindowSize() { |
| 137 | atomic.StoreUint32(&f.effectiveWindowSize, f.limit-f.unacked) |
| 138 | } |
| 139 | |
| 140 | func (f *trInFlow) getSize() uint32 { |
| 141 | return atomic.LoadUint32(&f.effectiveWindowSize) |
| 142 | } |
| 143 | |
| 144 | // TODO(mmukhi): Simplify this code. |
| 145 | // inFlow deals with inbound flow control |
| 146 | type inFlow struct { |
| 147 | mu sync.Mutex |
| 148 | // The inbound flow control limit for pending data. |
| 149 | limit uint32 |
| 150 | // pendingData is the overall data which have been received but not been |
| 151 | // consumed by applications. |
| 152 | pendingData uint32 |
| 153 | // The amount of data the application has consumed but grpc has not sent |
| 154 | // window update for them. Used to reduce window update frequency. |
| 155 | pendingUpdate uint32 |
| 156 | // delta is the extra window update given by receiver when an application |
| 157 | // is reading data bigger in size than the inFlow limit. |
| 158 | delta uint32 |
| 159 | } |
| 160 | |
| 161 | // newLimit updates the inflow window to a new value n. |
| 162 | // It assumes that n is always greater than the old limit. |
| 163 | func (f *inFlow) newLimit(n uint32) uint32 { |
| 164 | f.mu.Lock() |
| 165 | d := n - f.limit |
| 166 | f.limit = n |
| 167 | f.mu.Unlock() |
| 168 | return d |
| 169 | } |
| 170 | |
| 171 | func (f *inFlow) maybeAdjust(n uint32) uint32 { |
| 172 | if n > uint32(math.MaxInt32) { |
| 173 | n = uint32(math.MaxInt32) |
| 174 | } |
| 175 | f.mu.Lock() |
| 176 | // estSenderQuota is the receiver's view of the maximum number of bytes the sender |
| 177 | // can send without a window update. |
| 178 | estSenderQuota := int32(f.limit - (f.pendingData + f.pendingUpdate)) |
| 179 | // estUntransmittedData is the maximum number of bytes the sends might not have put |
| 180 | // on the wire yet. A value of 0 or less means that we have already received all or |
| 181 | // more bytes than the application is requesting to read. |
| 182 | estUntransmittedData := int32(n - f.pendingData) // Casting into int32 since it could be negative. |
| 183 | // This implies that unless we send a window update, the sender won't be able to send all the bytes |
| 184 | // for this message. Therefore we must send an update over the limit since there's an active read |
| 185 | // request from the application. |
| 186 | if estUntransmittedData > estSenderQuota { |
| 187 | // Sender's window shouldn't go more than 2^31 - 1 as specified in the HTTP spec. |
| 188 | if f.limit+n > maxWindowSize { |
| 189 | f.delta = maxWindowSize - f.limit |
| 190 | } else { |
| 191 | // Send a window update for the whole message and not just the difference between |
| 192 | // estUntransmittedData and estSenderQuota. This will be helpful in case the message |
| 193 | // is padded; We will fallback on the current available window(at least a 1/4th of the limit). |
| 194 | f.delta = n |
| 195 | } |
| 196 | f.mu.Unlock() |
| 197 | return f.delta |
| 198 | } |
| 199 | f.mu.Unlock() |
| 200 | return 0 |
| 201 | } |
| 202 | |
| 203 | // onData is invoked when some data frame is received. It updates pendingData. |
| 204 | func (f *inFlow) onData(n uint32) error { |
| 205 | f.mu.Lock() |
| 206 | f.pendingData += n |
| 207 | if f.pendingData+f.pendingUpdate > f.limit+f.delta { |
| 208 | limit := f.limit |
| 209 | rcvd := f.pendingData + f.pendingUpdate |
| 210 | f.mu.Unlock() |
| 211 | return fmt.Errorf("received %d-bytes data exceeding the limit %d bytes", rcvd, limit) |
| 212 | } |
| 213 | f.mu.Unlock() |
| 214 | return nil |
| 215 | } |
| 216 | |
| 217 | // onRead is invoked when the application reads the data. It returns the window size |
| 218 | // to be sent to the peer. |
| 219 | func (f *inFlow) onRead(n uint32) uint32 { |
| 220 | f.mu.Lock() |
| 221 | if f.pendingData == 0 { |
| 222 | f.mu.Unlock() |
| 223 | return 0 |
| 224 | } |
| 225 | f.pendingData -= n |
| 226 | if n > f.delta { |
| 227 | n -= f.delta |
| 228 | f.delta = 0 |
| 229 | } else { |
| 230 | f.delta -= n |
| 231 | n = 0 |
| 232 | } |
| 233 | f.pendingUpdate += n |
| 234 | if f.pendingUpdate >= f.limit/4 { |
| 235 | wu := f.pendingUpdate |
| 236 | f.pendingUpdate = 0 |
| 237 | f.mu.Unlock() |
| 238 | return wu |
| 239 | } |
| 240 | f.mu.Unlock() |
| 241 | return 0 |
| 242 | } |