Scott Baker | e7144bc | 2019-10-01 14:16:47 -0700 | [diff] [blame^] | 1 | package cluster |
| 2 | |
| 3 | import ( |
| 4 | "sort" |
| 5 | "sync" |
| 6 | "time" |
| 7 | |
| 8 | "github.com/Shopify/sarama" |
| 9 | ) |
| 10 | |
| 11 | // PartitionConsumer allows code to consume individual partitions from the cluster. |
| 12 | // |
| 13 | // See docs for Consumer.Partitions() for more on how to implement this. |
| 14 | type PartitionConsumer interface { |
| 15 | sarama.PartitionConsumer |
| 16 | |
| 17 | // Topic returns the consumed topic name |
| 18 | Topic() string |
| 19 | |
| 20 | // Partition returns the consumed partition |
| 21 | Partition() int32 |
| 22 | |
| 23 | // InitialOffset returns the offset used for creating the PartitionConsumer instance. |
| 24 | // The returned offset can be a literal offset, or OffsetNewest, or OffsetOldest |
| 25 | InitialOffset() int64 |
| 26 | |
| 27 | // MarkOffset marks the offset of a message as preocessed. |
| 28 | MarkOffset(offset int64, metadata string) |
| 29 | |
| 30 | // ResetOffset resets the offset to a previously processed message. |
| 31 | ResetOffset(offset int64, metadata string) |
| 32 | } |
| 33 | |
| 34 | type partitionConsumer struct { |
| 35 | sarama.PartitionConsumer |
| 36 | |
| 37 | state partitionState |
| 38 | mu sync.Mutex |
| 39 | |
| 40 | topic string |
| 41 | partition int32 |
| 42 | initialOffset int64 |
| 43 | |
| 44 | closeOnce sync.Once |
| 45 | closeErr error |
| 46 | |
| 47 | dying, dead chan none |
| 48 | } |
| 49 | |
| 50 | func newPartitionConsumer(manager sarama.Consumer, topic string, partition int32, info offsetInfo, defaultOffset int64) (*partitionConsumer, error) { |
| 51 | offset := info.NextOffset(defaultOffset) |
| 52 | pcm, err := manager.ConsumePartition(topic, partition, offset) |
| 53 | |
| 54 | // Resume from default offset, if requested offset is out-of-range |
| 55 | if err == sarama.ErrOffsetOutOfRange { |
| 56 | info.Offset = -1 |
| 57 | offset = defaultOffset |
| 58 | pcm, err = manager.ConsumePartition(topic, partition, offset) |
| 59 | } |
| 60 | if err != nil { |
| 61 | return nil, err |
| 62 | } |
| 63 | |
| 64 | return &partitionConsumer{ |
| 65 | PartitionConsumer: pcm, |
| 66 | state: partitionState{Info: info}, |
| 67 | |
| 68 | topic: topic, |
| 69 | partition: partition, |
| 70 | initialOffset: offset, |
| 71 | |
| 72 | dying: make(chan none), |
| 73 | dead: make(chan none), |
| 74 | }, nil |
| 75 | } |
| 76 | |
| 77 | // Topic implements PartitionConsumer |
| 78 | func (c *partitionConsumer) Topic() string { return c.topic } |
| 79 | |
| 80 | // Partition implements PartitionConsumer |
| 81 | func (c *partitionConsumer) Partition() int32 { return c.partition } |
| 82 | |
| 83 | // InitialOffset implements PartitionConsumer |
| 84 | func (c *partitionConsumer) InitialOffset() int64 { return c.initialOffset } |
| 85 | |
| 86 | // AsyncClose implements PartitionConsumer |
| 87 | func (c *partitionConsumer) AsyncClose() { |
| 88 | c.closeOnce.Do(func() { |
| 89 | c.closeErr = c.PartitionConsumer.Close() |
| 90 | close(c.dying) |
| 91 | }) |
| 92 | } |
| 93 | |
| 94 | // Close implements PartitionConsumer |
| 95 | func (c *partitionConsumer) Close() error { |
| 96 | c.AsyncClose() |
| 97 | <-c.dead |
| 98 | return c.closeErr |
| 99 | } |
| 100 | |
| 101 | func (c *partitionConsumer) waitFor(stopper <-chan none, errors chan<- error) { |
| 102 | defer close(c.dead) |
| 103 | |
| 104 | for { |
| 105 | select { |
| 106 | case err, ok := <-c.Errors(): |
| 107 | if !ok { |
| 108 | return |
| 109 | } |
| 110 | select { |
| 111 | case errors <- err: |
| 112 | case <-stopper: |
| 113 | return |
| 114 | case <-c.dying: |
| 115 | return |
| 116 | } |
| 117 | case <-stopper: |
| 118 | return |
| 119 | case <-c.dying: |
| 120 | return |
| 121 | } |
| 122 | } |
| 123 | } |
| 124 | |
| 125 | func (c *partitionConsumer) multiplex(stopper <-chan none, messages chan<- *sarama.ConsumerMessage, errors chan<- error) { |
| 126 | defer close(c.dead) |
| 127 | |
| 128 | for { |
| 129 | select { |
| 130 | case msg, ok := <-c.Messages(): |
| 131 | if !ok { |
| 132 | return |
| 133 | } |
| 134 | select { |
| 135 | case messages <- msg: |
| 136 | case <-stopper: |
| 137 | return |
| 138 | case <-c.dying: |
| 139 | return |
| 140 | } |
| 141 | case err, ok := <-c.Errors(): |
| 142 | if !ok { |
| 143 | return |
| 144 | } |
| 145 | select { |
| 146 | case errors <- err: |
| 147 | case <-stopper: |
| 148 | return |
| 149 | case <-c.dying: |
| 150 | return |
| 151 | } |
| 152 | case <-stopper: |
| 153 | return |
| 154 | case <-c.dying: |
| 155 | return |
| 156 | } |
| 157 | } |
| 158 | } |
| 159 | |
| 160 | func (c *partitionConsumer) getState() partitionState { |
| 161 | c.mu.Lock() |
| 162 | state := c.state |
| 163 | c.mu.Unlock() |
| 164 | |
| 165 | return state |
| 166 | } |
| 167 | |
| 168 | func (c *partitionConsumer) markCommitted(offset int64) { |
| 169 | c.mu.Lock() |
| 170 | if offset == c.state.Info.Offset { |
| 171 | c.state.Dirty = false |
| 172 | } |
| 173 | c.mu.Unlock() |
| 174 | } |
| 175 | |
| 176 | // MarkOffset implements PartitionConsumer |
| 177 | func (c *partitionConsumer) MarkOffset(offset int64, metadata string) { |
| 178 | c.mu.Lock() |
| 179 | if next := offset + 1; next > c.state.Info.Offset { |
| 180 | c.state.Info.Offset = next |
| 181 | c.state.Info.Metadata = metadata |
| 182 | c.state.Dirty = true |
| 183 | } |
| 184 | c.mu.Unlock() |
| 185 | } |
| 186 | |
| 187 | // ResetOffset implements PartitionConsumer |
| 188 | func (c *partitionConsumer) ResetOffset(offset int64, metadata string) { |
| 189 | c.mu.Lock() |
| 190 | if next := offset + 1; next <= c.state.Info.Offset { |
| 191 | c.state.Info.Offset = next |
| 192 | c.state.Info.Metadata = metadata |
| 193 | c.state.Dirty = true |
| 194 | } |
| 195 | c.mu.Unlock() |
| 196 | } |
| 197 | |
| 198 | // -------------------------------------------------------------------- |
| 199 | |
| 200 | type partitionState struct { |
| 201 | Info offsetInfo |
| 202 | Dirty bool |
| 203 | LastCommit time.Time |
| 204 | } |
| 205 | |
| 206 | // -------------------------------------------------------------------- |
| 207 | |
| 208 | type partitionMap struct { |
| 209 | data map[topicPartition]*partitionConsumer |
| 210 | mu sync.RWMutex |
| 211 | } |
| 212 | |
| 213 | func newPartitionMap() *partitionMap { |
| 214 | return &partitionMap{ |
| 215 | data: make(map[topicPartition]*partitionConsumer), |
| 216 | } |
| 217 | } |
| 218 | |
| 219 | func (m *partitionMap) IsSubscribedTo(topic string) bool { |
| 220 | m.mu.RLock() |
| 221 | defer m.mu.RUnlock() |
| 222 | |
| 223 | for tp := range m.data { |
| 224 | if tp.Topic == topic { |
| 225 | return true |
| 226 | } |
| 227 | } |
| 228 | return false |
| 229 | } |
| 230 | |
| 231 | func (m *partitionMap) Fetch(topic string, partition int32) *partitionConsumer { |
| 232 | m.mu.RLock() |
| 233 | pc, _ := m.data[topicPartition{topic, partition}] |
| 234 | m.mu.RUnlock() |
| 235 | return pc |
| 236 | } |
| 237 | |
| 238 | func (m *partitionMap) Store(topic string, partition int32, pc *partitionConsumer) { |
| 239 | m.mu.Lock() |
| 240 | m.data[topicPartition{topic, partition}] = pc |
| 241 | m.mu.Unlock() |
| 242 | } |
| 243 | |
| 244 | func (m *partitionMap) Snapshot() map[topicPartition]partitionState { |
| 245 | m.mu.RLock() |
| 246 | defer m.mu.RUnlock() |
| 247 | |
| 248 | snap := make(map[topicPartition]partitionState, len(m.data)) |
| 249 | for tp, pc := range m.data { |
| 250 | snap[tp] = pc.getState() |
| 251 | } |
| 252 | return snap |
| 253 | } |
| 254 | |
| 255 | func (m *partitionMap) Stop() { |
| 256 | m.mu.RLock() |
| 257 | defer m.mu.RUnlock() |
| 258 | |
| 259 | var wg sync.WaitGroup |
| 260 | for tp := range m.data { |
| 261 | wg.Add(1) |
| 262 | go func(p *partitionConsumer) { |
| 263 | _ = p.Close() |
| 264 | wg.Done() |
| 265 | }(m.data[tp]) |
| 266 | } |
| 267 | wg.Wait() |
| 268 | } |
| 269 | |
| 270 | func (m *partitionMap) Clear() { |
| 271 | m.mu.Lock() |
| 272 | for tp := range m.data { |
| 273 | delete(m.data, tp) |
| 274 | } |
| 275 | m.mu.Unlock() |
| 276 | } |
| 277 | |
| 278 | func (m *partitionMap) Info() map[string][]int32 { |
| 279 | info := make(map[string][]int32) |
| 280 | m.mu.RLock() |
| 281 | for tp := range m.data { |
| 282 | info[tp.Topic] = append(info[tp.Topic], tp.Partition) |
| 283 | } |
| 284 | m.mu.RUnlock() |
| 285 | |
| 286 | for topic := range info { |
| 287 | sort.Sort(int32Slice(info[topic])) |
| 288 | } |
| 289 | return info |
| 290 | } |