sender分析之创建请求
一 Sender run方法调用流程
# 从Metadata获取集群元数据
# 调用RecordAccumulator.ready方法,根据RecordAccumulator的缓存情况,选出可以向哪些Node发送消息,返回ReadyCheckResult对象
# 如果ReadyCheckResult存在某些分区没有leader副本,则调用Metadata的requestUpdate方法,标记需要更新kafka的集群信息
# 针对ReadyCheckResult的readyNodes集合,循环调用NetworkClient的ready方法,目的是检测网络I/O方面是否符合发送条件,不符合发送条件的Node将会从readyNodes集合中删除
# 调用RecordAccumulator的drain方法获取待发送的消息集合
# 调用RecordAccumulator的abortExpiredBatchers方法处理RecordAccumulator中超时的消息
# 调用Sender的createProduceRequests方法,将发送的消息封装成ClientRequest请求
# 调用NetworkClient.send方法,将ClientRequest写入KafkaChannel的send字段
# 调用NetworkClient的poll方法,将KafkaChannel中send字段保存的ClientRequest发送出去,同时还会处理服务端发回的响应处理超时请求,调用用户自定义的函数等
void
run(long
now) {
// 从Metadata获取集群元数据
Cluster cluster
= metadata.fetch();
// 根据RecordAccumulator的缓存情况,选出可以向哪些Node发送消息,返回ReadyCheckResult对象
RecordAccumulator.ReadyCheckResultresult
= this.accumulator.ready(cluster,
now);
// 如果ReadyCheckResult存在某些分区没有leader副本,则调用Metadata的requestUpdate方法,标记需要更新kafka的集群信息
if (!result.unknownLeaderTopics.isEmpty()) {
for (String
topic : result.unknownLeaderTopics)
this.metadata.add(topic);
this.metadata.requestUpdate();
}
// 针对ReadyCheckResult的readyNodes集合,循环调用NetworkClient的ready方法,
// 目的是检测网络I/O方面是否符合发送条件,不符合发送条件的Node将会从readyNodes集合中删除
Iterator<Node>
iter = result.readyNodes.iterator();
long notReadyTimeout
= Long.MAX_VALUE;
while (iter.hasNext()) {
Node node
= iter.next();
if (!this.client.ready(node,
now)) {
iter.remove();
notReadyTimeout=
Math.min(notReadyTimeout,
this.client.connectionDelay(node,
now));
}
}
// 调用RecordAccumulator的drain方法获取待发送的消息集合
Map<Integer,
List<RecordBatch>>
batches = this.accumulator.drain(cluster,
result.readyNodes,
this.maxRequestSize,
now);
// 是否需要保证消息的顺序
if (guaranteeMessageOrder) {
// 遍历record batch
for (List<RecordBatch>
batchList : batches.values()) {
for (RecordBatch
batch : batchList)
this.accumulator.mutePartition(batch.topicPartition);
}
}
// 调用RecordAccumulator的abortExpiredBatchers方法处理RecordAccumulator中超时的消息
List<RecordBatch>
expiredBatches = this.accumulator.abortExpiredBatches(this.requestTimeout,
now);
for (RecordBatch
expiredBatch :
expiredBatches)
this.sensors.recordErrors(expiredBatch.topicPartition.topic(),
expiredBatch.recordCount);
sensors.updateProduceRequestMetrics(batches);
// 创建生产者请求
List<ClientRequest>
requests = createProduceRequests(batches,
now);
long pollTimeout
= Math.min(result.nextReadyCheckDelayMs,
notReadyTimeout);
if (result.readyNodes.size() >
0) {
log.trace("Nodes with data ready tosend: {}",
result.readyNodes);
log.trace("Created {} producerequests: {}",
requests.size(),
requests);
pollTimeout =
0;
}
// 将ClientRequest写入KafkaChannel的send字段
for (ClientRequest
request : requests)
client.send(request,
now);
// 调用NetworkClient的poll方法,将KafkaChannel中send字段保存的ClientRequest发送出去,
// 同时还会处理服务端发回的响应处理超时请求,调用用户自定义的函数等
this.client.poll(pollTimeout,
now);
}
二 创建请求
我们先分析ProduceRequest和ProduceResponse消息体格式:
api_key: API标识
api_version: API版本号
correaltion_id: 一个单调递增序号
client_id: 客户端id
acks: 确认机制,0 不需要确认,1 只需要leader确认,-1所有副本都需要确认
timeout: 超时时间
topic: topic名称
partition: partition编号
record_set: 消息
correaltion_id: 一个单调递增序号
topic: topic名称
partition: partition编号
error_code: 错误码
base_offset: 服务端为消息生成的一个offset
timestamp: 瞬间戳
throttle_time_ms: 延长时间
private List<ClientRequest> createProduceRequests(Map<Integer, List<RecordBatch>> collated, long now) {
// 保存创建的ClientRequest列表
List<ClientRequest> requests = new ArrayList<ClientRequest>(collated.size());
for (Map.Entry<Integer, List<RecordBatch>> entry : collated.entrySet())
// 将发往同一个Node的RecordBatch封装成ClientRequest
requests.add(produceRequest(now, entry.getKey(), acks, requestTimeout, entry.getValue()));
return requests;
}
private ClientRequest produceRequest(long now, int destination, short acks, int timeout, List<RecordBatch> batches) {
Map<TopicPartition, ByteBuffer> produceRecordsByPartition = new HashMap<TopicPartition, ByteBuffer>(batches.size());
final Map<TopicPartition, RecordBatch> recordsByPartition = new HashMap<TopicPartition, RecordBatch>(batches.size());
// 将RecordBatch按照partiiton分类,同时构建集合
for (RecordBatch batch : batches) {
TopicPartition tp = batch.topicPartition;
produceRecordsByPartition.put(tp, batch.records.buffer());
recordsByPartition.put(tp, batch);
}
// 创建ProduceRequest和RequestSend
ProduceRequest request = new ProduceRequest(acks, timeout, produceRecordsByPartition);
RequestSend send = new RequestSend(Integer.toString(destination),
this.client.nextRequestHeader(ApiKeys.PRODUCE),
request.toStruct());
// 创建RequestCompletionHandler作为回调对象
RequestCompletionHandler callback = new RequestCompletionHandler() {
public void onComplete(ClientResponse response) {
handleProduceResponse(response, recordsByPartition, time.milliseconds());
}
};
// 创建ClientRequest对象
return new ClientRequest(now, acks != 0, send, callback);
}