Scott Baker | e7144bc | 2019-10-01 14:16:47 -0700 | [diff] [blame^] | 1 | package sarama |
| 2 | |
| 3 | import ( |
| 4 | "encoding/binary" |
| 5 | "errors" |
| 6 | "time" |
| 7 | ) |
| 8 | |
| 9 | type partitionSet struct { |
| 10 | msgs []*ProducerMessage |
| 11 | recordsToSend Records |
| 12 | bufferBytes int |
| 13 | } |
| 14 | |
| 15 | type produceSet struct { |
| 16 | parent *asyncProducer |
| 17 | msgs map[string]map[int32]*partitionSet |
| 18 | |
| 19 | bufferBytes int |
| 20 | bufferCount int |
| 21 | } |
| 22 | |
| 23 | func newProduceSet(parent *asyncProducer) *produceSet { |
| 24 | return &produceSet{ |
| 25 | msgs: make(map[string]map[int32]*partitionSet), |
| 26 | parent: parent, |
| 27 | } |
| 28 | } |
| 29 | |
| 30 | func (ps *produceSet) add(msg *ProducerMessage) error { |
| 31 | var err error |
| 32 | var key, val []byte |
| 33 | |
| 34 | if msg.Key != nil { |
| 35 | if key, err = msg.Key.Encode(); err != nil { |
| 36 | return err |
| 37 | } |
| 38 | } |
| 39 | |
| 40 | if msg.Value != nil { |
| 41 | if val, err = msg.Value.Encode(); err != nil { |
| 42 | return err |
| 43 | } |
| 44 | } |
| 45 | |
| 46 | timestamp := msg.Timestamp |
| 47 | if msg.Timestamp.IsZero() { |
| 48 | timestamp = time.Now() |
| 49 | } |
| 50 | |
| 51 | partitions := ps.msgs[msg.Topic] |
| 52 | if partitions == nil { |
| 53 | partitions = make(map[int32]*partitionSet) |
| 54 | ps.msgs[msg.Topic] = partitions |
| 55 | } |
| 56 | |
| 57 | var size int |
| 58 | |
| 59 | set := partitions[msg.Partition] |
| 60 | if set == nil { |
| 61 | if ps.parent.conf.Version.IsAtLeast(V0_11_0_0) { |
| 62 | batch := &RecordBatch{ |
| 63 | FirstTimestamp: timestamp, |
| 64 | Version: 2, |
| 65 | Codec: ps.parent.conf.Producer.Compression, |
| 66 | CompressionLevel: ps.parent.conf.Producer.CompressionLevel, |
| 67 | ProducerID: ps.parent.txnmgr.producerID, |
| 68 | ProducerEpoch: ps.parent.txnmgr.producerEpoch, |
| 69 | } |
| 70 | if ps.parent.conf.Producer.Idempotent { |
| 71 | batch.FirstSequence = msg.sequenceNumber |
| 72 | } |
| 73 | set = &partitionSet{recordsToSend: newDefaultRecords(batch)} |
| 74 | size = recordBatchOverhead |
| 75 | } else { |
| 76 | set = &partitionSet{recordsToSend: newLegacyRecords(new(MessageSet))} |
| 77 | } |
| 78 | partitions[msg.Partition] = set |
| 79 | } |
| 80 | set.msgs = append(set.msgs, msg) |
| 81 | |
| 82 | if ps.parent.conf.Version.IsAtLeast(V0_11_0_0) { |
| 83 | if ps.parent.conf.Producer.Idempotent && msg.sequenceNumber < set.recordsToSend.RecordBatch.FirstSequence { |
| 84 | return errors.New("Assertion failed: Message out of sequence added to a batch") |
| 85 | } |
| 86 | // We are being conservative here to avoid having to prep encode the record |
| 87 | size += maximumRecordOverhead |
| 88 | rec := &Record{ |
| 89 | Key: key, |
| 90 | Value: val, |
| 91 | TimestampDelta: timestamp.Sub(set.recordsToSend.RecordBatch.FirstTimestamp), |
| 92 | } |
| 93 | size += len(key) + len(val) |
| 94 | if len(msg.Headers) > 0 { |
| 95 | rec.Headers = make([]*RecordHeader, len(msg.Headers)) |
| 96 | for i := range msg.Headers { |
| 97 | rec.Headers[i] = &msg.Headers[i] |
| 98 | size += len(rec.Headers[i].Key) + len(rec.Headers[i].Value) + 2*binary.MaxVarintLen32 |
| 99 | } |
| 100 | } |
| 101 | set.recordsToSend.RecordBatch.addRecord(rec) |
| 102 | } else { |
| 103 | msgToSend := &Message{Codec: CompressionNone, Key: key, Value: val} |
| 104 | if ps.parent.conf.Version.IsAtLeast(V0_10_0_0) { |
| 105 | msgToSend.Timestamp = timestamp |
| 106 | msgToSend.Version = 1 |
| 107 | } |
| 108 | set.recordsToSend.MsgSet.addMessage(msgToSend) |
| 109 | size = producerMessageOverhead + len(key) + len(val) |
| 110 | } |
| 111 | |
| 112 | set.bufferBytes += size |
| 113 | ps.bufferBytes += size |
| 114 | ps.bufferCount++ |
| 115 | |
| 116 | return nil |
| 117 | } |
| 118 | |
| 119 | func (ps *produceSet) buildRequest() *ProduceRequest { |
| 120 | req := &ProduceRequest{ |
| 121 | RequiredAcks: ps.parent.conf.Producer.RequiredAcks, |
| 122 | Timeout: int32(ps.parent.conf.Producer.Timeout / time.Millisecond), |
| 123 | } |
| 124 | if ps.parent.conf.Version.IsAtLeast(V0_10_0_0) { |
| 125 | req.Version = 2 |
| 126 | } |
| 127 | if ps.parent.conf.Version.IsAtLeast(V0_11_0_0) { |
| 128 | req.Version = 3 |
| 129 | } |
| 130 | |
| 131 | for topic, partitionSets := range ps.msgs { |
| 132 | for partition, set := range partitionSets { |
| 133 | if req.Version >= 3 { |
| 134 | // If the API version we're hitting is 3 or greater, we need to calculate |
| 135 | // offsets for each record in the batch relative to FirstOffset. |
| 136 | // Additionally, we must set LastOffsetDelta to the value of the last offset |
| 137 | // in the batch. Since the OffsetDelta of the first record is 0, we know that the |
| 138 | // final record of any batch will have an offset of (# of records in batch) - 1. |
| 139 | // (See https://cwiki.apache.org/confluence/display/KAFKA/A+Guide+To+The+Kafka+Protocol#AGuideToTheKafkaProtocol-Messagesets |
| 140 | // under the RecordBatch section for details.) |
| 141 | rb := set.recordsToSend.RecordBatch |
| 142 | if len(rb.Records) > 0 { |
| 143 | rb.LastOffsetDelta = int32(len(rb.Records) - 1) |
| 144 | for i, record := range rb.Records { |
| 145 | record.OffsetDelta = int64(i) |
| 146 | } |
| 147 | } |
| 148 | req.AddBatch(topic, partition, rb) |
| 149 | continue |
| 150 | } |
| 151 | if ps.parent.conf.Producer.Compression == CompressionNone { |
| 152 | req.AddSet(topic, partition, set.recordsToSend.MsgSet) |
| 153 | } else { |
| 154 | // When compression is enabled, the entire set for each partition is compressed |
| 155 | // and sent as the payload of a single fake "message" with the appropriate codec |
| 156 | // set and no key. When the server sees a message with a compression codec, it |
| 157 | // decompresses the payload and treats the result as its message set. |
| 158 | |
| 159 | if ps.parent.conf.Version.IsAtLeast(V0_10_0_0) { |
| 160 | // If our version is 0.10 or later, assign relative offsets |
| 161 | // to the inner messages. This lets the broker avoid |
| 162 | // recompressing the message set. |
| 163 | // (See https://cwiki.apache.org/confluence/display/KAFKA/KIP-31+-+Move+to+relative+offsets+in+compressed+message+sets |
| 164 | // for details on relative offsets.) |
| 165 | for i, msg := range set.recordsToSend.MsgSet.Messages { |
| 166 | msg.Offset = int64(i) |
| 167 | } |
| 168 | } |
| 169 | payload, err := encode(set.recordsToSend.MsgSet, ps.parent.conf.MetricRegistry) |
| 170 | if err != nil { |
| 171 | Logger.Println(err) // if this happens, it's basically our fault. |
| 172 | panic(err) |
| 173 | } |
| 174 | compMsg := &Message{ |
| 175 | Codec: ps.parent.conf.Producer.Compression, |
| 176 | CompressionLevel: ps.parent.conf.Producer.CompressionLevel, |
| 177 | Key: nil, |
| 178 | Value: payload, |
| 179 | Set: set.recordsToSend.MsgSet, // Provide the underlying message set for accurate metrics |
| 180 | } |
| 181 | if ps.parent.conf.Version.IsAtLeast(V0_10_0_0) { |
| 182 | compMsg.Version = 1 |
| 183 | compMsg.Timestamp = set.recordsToSend.MsgSet.Messages[0].Msg.Timestamp |
| 184 | } |
| 185 | req.AddMessage(topic, partition, compMsg) |
| 186 | } |
| 187 | } |
| 188 | } |
| 189 | |
| 190 | return req |
| 191 | } |
| 192 | |
| 193 | func (ps *produceSet) eachPartition(cb func(topic string, partition int32, pSet *partitionSet)) { |
| 194 | for topic, partitionSet := range ps.msgs { |
| 195 | for partition, set := range partitionSet { |
| 196 | cb(topic, partition, set) |
| 197 | } |
| 198 | } |
| 199 | } |
| 200 | |
| 201 | func (ps *produceSet) dropPartition(topic string, partition int32) []*ProducerMessage { |
| 202 | if ps.msgs[topic] == nil { |
| 203 | return nil |
| 204 | } |
| 205 | set := ps.msgs[topic][partition] |
| 206 | if set == nil { |
| 207 | return nil |
| 208 | } |
| 209 | ps.bufferBytes -= set.bufferBytes |
| 210 | ps.bufferCount -= len(set.msgs) |
| 211 | delete(ps.msgs[topic], partition) |
| 212 | return set.msgs |
| 213 | } |
| 214 | |
| 215 | func (ps *produceSet) wouldOverflow(msg *ProducerMessage) bool { |
| 216 | version := 1 |
| 217 | if ps.parent.conf.Version.IsAtLeast(V0_11_0_0) { |
| 218 | version = 2 |
| 219 | } |
| 220 | |
| 221 | switch { |
| 222 | // Would we overflow our maximum possible size-on-the-wire? 10KiB is arbitrary overhead for safety. |
| 223 | case ps.bufferBytes+msg.byteSize(version) >= int(MaxRequestSize-(10*1024)): |
| 224 | return true |
| 225 | // Would we overflow the size-limit of a message-batch for this partition? |
| 226 | case ps.msgs[msg.Topic] != nil && ps.msgs[msg.Topic][msg.Partition] != nil && |
| 227 | ps.msgs[msg.Topic][msg.Partition].bufferBytes+msg.byteSize(version) >= ps.parent.conf.Producer.MaxMessageBytes: |
| 228 | return true |
| 229 | // Would we overflow simply in number of messages? |
| 230 | case ps.parent.conf.Producer.Flush.MaxMessages > 0 && ps.bufferCount >= ps.parent.conf.Producer.Flush.MaxMessages: |
| 231 | return true |
| 232 | default: |
| 233 | return false |
| 234 | } |
| 235 | } |
| 236 | |
| 237 | func (ps *produceSet) readyToFlush() bool { |
| 238 | switch { |
| 239 | // If we don't have any messages, nothing else matters |
| 240 | case ps.empty(): |
| 241 | return false |
| 242 | // If all three config values are 0, we always flush as-fast-as-possible |
| 243 | case ps.parent.conf.Producer.Flush.Frequency == 0 && ps.parent.conf.Producer.Flush.Bytes == 0 && ps.parent.conf.Producer.Flush.Messages == 0: |
| 244 | return true |
| 245 | // If we've passed the message trigger-point |
| 246 | case ps.parent.conf.Producer.Flush.Messages > 0 && ps.bufferCount >= ps.parent.conf.Producer.Flush.Messages: |
| 247 | return true |
| 248 | // If we've passed the byte trigger-point |
| 249 | case ps.parent.conf.Producer.Flush.Bytes > 0 && ps.bufferBytes >= ps.parent.conf.Producer.Flush.Bytes: |
| 250 | return true |
| 251 | default: |
| 252 | return false |
| 253 | } |
| 254 | } |
| 255 | |
| 256 | func (ps *produceSet) empty() bool { |
| 257 | return ps.bufferCount == 0 |
| 258 | } |