Bulk注册
在启动类BootStrap的start()方法中，启动了node.start()方法。在Node初始化的过程中，加载了一系列的模块和插件，其中包含ActionModel。
ActionModule actionModule = new ActionModule(false, settings, clusterModule.getIndexNameExpressionResolver(),
settingsModule.getIndexScopedSettings(), settingsModule.getClusterSettings(), settingsModule.getSettingsFilter(),
threadPool, pluginsService.filterPlugins(ActionPlugin.class), client, circuitBreakerService);
modules.add(actionModule);在ActionModel中，注册了我们常用的一些操作action，比如说我们这次解析的BulkAction：
actions.register(UpdateAction.INSTANCE, TransportUpdateAction.class);
actions.register(MultiGetAction.INSTANCE, TransportMultiGetAction.class,TransportShardMultiGetAction.class);
actions.register(BulkAction.INSTANCE, TransportBulkAction.class,TransportShardBulkAction.class);
并且初始化RestHandler：
registerHandler.accept(new RestMultiTermVectorsAction(settings, restController));
registerHandler.accept(new RestBulkAction(settings, restController));
registerHandler.accept(new RestUpdateAction(settings, restController));
在RestBulkAction中规定了我们的查询方式：
controller.registerHandler(POST, "/_bulk", this);
controller.registerHandler(PUT, "/_bulk", this);
controller.registerHandler(POST, "/{index}/_bulk", this);
controller.registerHandler(PUT, "/{index}/_bulk", this);
controller.registerHandler(POST, "/{index}/{type}/_bulk", this);
controller.registerHandler(PUT, "/{index}/{type}/_bulk", this);
接收到请求
RestBulkAction在prepareRequest方法中将我们普通的RestRequest转化为BulkReqest，并通过NodeClient调用：
channel -> client.bulk(bulkRequest, new RestStatusToXContentListener<>(channel));
而在NodeClient的bulk中则是调用了NodeClient的doExecute()方法。
doExecute(Action<Request, Response, RequestBuilder> action, Request request, ActionListener<Response> listener)
传入的Action是BulkAction.Instance，request就是上一步封装的BulkRequest，listener则是监听器。
在doExecute方法中，首先将普通的action转化为tansportAction，然后用转化后的tansportAction执行该请求：
transportAction(action).execute(request, listener);
bulkAction转化后变为TransportBulkAction，而TransportBulkAction的execute方法则是调用本身的doExecute()方法。在doExecut()方法中首先将存在和不存在的索引分类：
1）Step 1: collect all the indices in the request
2）Step 2: filter that to indices that don't exist and we can create. At the same time build a map of indices we can't create that we'll use when we try to run the requests.
3）Step 3: create all the indices that are missing, if there are any missing. start the bulk after all the creates come back.
然后执行executeBulk()方法，接着在executeBulk中创建一个BulkOperation，并开始执行该BulkOperation：
void executeBulk(Task task, final BulkRequest bulkRequest, final long startTimeNanos, final ActionListener<BulkResponse> listener,
final AtomicArray<BulkItemResponse> responses, Map<String, IndexNotFoundException> indicesThatCannotBeCreated) {
new BulkOperation(task, bulkRequest, listener, responses, startTimeNanos, indicesThatCannotBeCreated).run();
}
在BulkOperation中存在两次遍历Bulk中所有的请求，次遍历则将给该请求设置Routing，Mapping等等，如果允许产生ID，则自动生成ID。第二次遍历则是根据shardID将请求分类。ES官网有说到批量处理时让用bulk，原因是bulk处理请求时做了一些底层的优化。这就是一个优化点，将同一个shard的请求集合在一起直接发送到节点对应的shard，避免请求在节点间传递，影响效率。
for (int i = ; i < bulkRequest.requests.size(); i++) {
....
switch (docWriteRequest.opType()) {
case CREATE:
case INDEX:
IndexRequest indexRequest = (IndexRequest) docWriteRequest;
MappingMetaData mappingMd = null;
final IndexMetaData indexMetaData = metaData.index(concreteIndex);
if (indexMetaData != null) {
mappingMd = indexMetaData.mappingOrDefault(indexRequest.type());
}
indexRequest.resolveRouting(metaData);
indexRequest.process(mappingMd, allowIdGeneration, concreteIndex.getName());
....
}
....
}
....
for (int i = ; i < bulkRequest.requests.size(); i++) {
DocWriteRequest request = bulkRequest.requests.get(i);
if (request == null) {
continue;
}
String concreteIndex = concreteIndices.getConcreteIndex(request.index()).getName();
ShardId shardId = clusterService.operationRouting().indexShards(clusterState, concreteIndex, request.id(), request.routing()).shardId();
List<BulkItemRequest> shardRequests = requestsByShard.computeIfAbsent(shardId, shard -> new ArrayList<>());
shardRequests.add(new BulkItemRequest(i, request));
}
然后针对不同的shardRequest，分别用shardBulkAction处理：
shardBulkAction.execute(bulkShardRequest, new ActionListener<BulkShardResponse>() {}
每个shard的处理流程
接下来就是复杂的类继承关系了：
TransportShardBulkAction>TransportWriteAction >TransportReplicationAction>TransportAction
上一步的shardBulkAction.execute()方法则是执行的TransportAction的execute方法。我看得源码版本是5.6版本的，与5.0版本相比，ES增加了一个
TransportWriteAction类，而且在TransportReplicationAction不是直接运行run方法，而是通过transportService的RPC接口在实现功能。具体的流程如下：
1）TransportAction.execute()方法会调用TransportReplicationAction的doExecute()方法
2）在TransportReplicationAction的doExecute()方法中执行ReroutePhase的run方法，run方法中根据请求的shardID获取到primary shardID，同时得到primary shard的NodeID，如果当前节点包含primary shard，则执行performLocalAction方法，否则执行performRemoteAction。
3）performLocalAction和performRemoteAction终都将执行performAction方法，在performAction中我们可以看到，transportService发送请求：
transportService.sendRequest(node, action, requestToPerform, transportOptions, new TransportResponseHandler<Response>() {}
4）transportService接收到请求后用的PrimaryOperationTransportHandler处理，至于PrimaryOperationTransportHandler是在TransportReplicationAction中注册的：
transportService.registerRequestHandler(transportPrimaryAction, () -> new ConcreteShardRequest<>(request), executor,
new PrimaryOperationTransportHandler());
5）PrimaryOperationTransportHandler则是一个primary操作的处理类，在这个类接收到信息之后调用AsyncPrimaryAction处理：
@Override
public void messageReceived(ConcreteShardRequest<Request> request, TransportChannel channel, Task task) {
new AsyncPrimaryAction(request.request, request.targetAllocationID, channel, (ReplicationTask) task).run();
}
6）在AsyncPrimaryAction中首先获取shard锁，如果成功的获取到锁则调用自身的onresponse()方法，否则将获取操作加入线程池：
synchronized (this) {
releasable = tryAcquire();
if (releasable == null) {
// blockOperations is executing, this operation will be retried by blockOperations once it finishes
if (delayedOperations == null) {
delayedOperations = new ArrayList<>();
}
final Supplier<StoredContext> contextSupplier = threadPool.getThreadContext().newRestorableContext(false);
if (executorOnDelay != null) {
delayedOperations.add(
new ThreadedActionListener<>(logger, threadPool, executorOnDelay,
new ContextPreservingActionListener<>(contextSupplier, onAcquired), forceExecution));
} else {
delayedOperations.add(new ContextPreservingActionListener<>(contextSupplier, onAcquired));
}
return;
}
}
7）在onresponse中，如果该primaryShardReference已经被移动了，则获取到正确的primary shard和nodeID重新发送请求。否则就用primaryShardReference直接处理：
@Override
public void onResponse(PrimaryShardReference primaryShardReference) {
try {
if (primaryShardReference.isRelocated()) {
primaryShardReference.close(); // release shard operation lock as soon as possible
setPhase(replicationTask, "primary_delegation");
// delegate primary phase to relocation target
// it is safe to execute primary phase on relocation target as there are no more in-flight operations where primary
// phase is executed on local shard and all subsequent operations are executed on relocation target as primary phase.
final ShardRouting primary = primaryShardReference.routingEntry();
assert primary.relocating() : "indexShard is marked as relocated but routing isn't" + primary;
DiscoveryNode relocatingNode = clusterService.state().nodes().get(primary.relocatingNodeId());
transportService.sendRequest(relocatingNode, transportPrimaryAction,
new ConcreteShardRequest<>(request, primary.allocationId().getRelocationId()),
transportOptions,
new TransportChannelResponseHandler<Response>(logger, channel, "rerouting indexing to target primary " + primary,
TransportReplicationAction.this::newResponseInstance) {
@Override
public void handleResponse(Response response) {
setPhase(replicationTask, "finished");
super.handleResponse(response);
}
@Override
public void handleException(TransportException exp) {
setPhase(replicationTask, "finished");
super.handleException(exp);
}
});
} else {
setPhase(replicationTask, "primary");
final IndexMetaData indexMetaData = clusterService.state().getMetaData().index(request.shardId().getIndex());
final boolean executeOnReplicas = (indexMetaData == null) || shouldExecuteReplication(indexMetaData);
final ActionListener<Response> listener = createResponseListener(primaryShardReference);
createReplicatedOperation(request,
ActionListener.wrap(result -> result.respond(listener), listener::onFailure),
primaryShardReference, executeOnReplicas)
.execute();
}
} catch (Exception e) {
Releasables.closeWhileHandlingException(primaryShardReference); // release shard operation lock before responding to caller
onFailure(e);
}
}
8）createReplicatedOperation看名字还以为直接就是副本处理了，点进去看了之后才发现是先执行primary，后执行replia。
primaryResult = primary.perform(request);
...
performOnReplicas(replicaRequest, shards);
主分片处理
主分片的处理调用的是PrimaryShardReference.perform()方法，在该方法中则是调用shardOperationOnPrimary()进行主分片的处理。
shardOperationOnPrimary()方法则是由TransportShardBulkAction来实现的，具体执行的步骤如下：
1）获取节点中所有的索引元数据
2）获取版本号
3）更新mapping
4）调用Engin底层的代码。比如说primary.delete(delete)，primary.index(operation)等等。
5）写到tanslog中
副本分片和主分片类似，这里就不做过多解释。
作者：YG_9013
链接：https://www.jianshu.com/p/e5c90a348931
来源：简书
著作权归作者所有。商业转载请联系作者获得授权，非商业转载请注明出处。
