Elasticsearch 8.9 Bulk批量给索引增加数据源码

下面的图来自ElasticSearch——刷盘原理流程，这篇文章主要讲的是客户端发送bulk命令到保存到ES缓冲区和translog的过程源码，不涉及到把数据刷到磁盘的逻辑，也不讲解存储在Lucene的数据结构
通过这个流程知道ES写入的数据会先暂存在哪里。主要是下面中上部分，主节点同时写入到ES缓冲区及虚线框事务日志

图片中的refresh部分，请看Elasticsearch 8.9 refresh刷Es缓冲区的数据到Lucene，更新segemnt，使数据可见
图片中的fsync部分，请看Elasticsearch 8.9 flush刷新缓存中的数据到磁盘源码

一、相关API的handler

在ActionModule.java中

//主节点处理谁分发到不同数据节点node的逻辑
 actions.register(BulkAction.INSTANCE, TransportBulkAction.class);
 //node节点接收到主节点分发的数据后的处理
 actions.register(TransportShardBulkAction.TYPE, TransportShardBulkAction.class);
 //主节点接收客户端的请求的hander
 registerHandler.accept(new RestBulkAction(settings));

二、RestBulkAction，组装bulkRequest调用TransportBulkAction

public class RestBulkAction extends BaseRestHandler {
 
    @Override
    public List<Route> routes() {
        return List.of(
            new Route(POST, "/_bulk"),
            new Route(PUT, "/_bulk"),
            new Route(POST, "/{index}/_bulk"),
            new Route(PUT, "/{index}/_bulk"),
            Route.builder(POST, "/{index}/{type}/_bulk").deprecated(TYPES_DEPRECATION_MESSAGE, RestApiVersion.V_7).build(),
            Route.builder(PUT, "/{index}/{type}/_bulk").deprecated(TYPES_DEPRECATION_MESSAGE, RestApiVersion.V_7).build()
        );
    }

    @Override
    public RestChannelConsumer prepareRequest(final RestRequest request, final NodeClient client) throws IOException {
        if (request.getRestApiVersion() == RestApiVersion.V_7 && request.hasParam("type")) {
            request.param("type");
        }
        BulkRequest bulkRequest = new BulkRequest();
        String defaultIndex = request.param("index");
        String defaultRouting = request.param("routing");
        FetchSourceContext defaultFetchSourceContext = FetchSourceContext.parseFromRestRequest(request);
        String defaultPipeline = request.param("pipeline");
        String waitForActiveShards = request.param("wait_for_active_shards");
        if (waitForActiveShards != null) {
            bulkRequest.waitForActiveShards(ActiveShardCount.parseString(waitForActiveShards));
        }
        Boolean defaultRequireAlias = request.paramAsBoolean(DocWriteRequest.REQUIRE_ALIAS, null);
        bulkRequest.timeout(request.paramAsTime("timeout", BulkShardRequest.DEFAULT_TIMEOUT));
        bulkRequest.setRefreshPolicy(request.param("refresh"));
        bulkRequest.add(
            request.requiredContent(),
            defaultIndex,
            defaultRouting,
            defaultFetchSourceContext,
            defaultPipeline,
            defaultRequireAlias,
            allowExplicitIndex,
            request.getXContentType(),
            request.getRestApiVersion()
        );
		
        return channel -> client.bulk(bulkRequest, new RestStatusToXContentListener<>(channel));
    }
}

  @Override
    public void bulk(final BulkRequest request, final ActionListener<BulkResponse> listener) {
        execute(BulkAction.INSTANCE, request, listener);
    }

其中BulkAction.INSTANCE会通过最上面的actions转到TransportBulkAction.class

三、TransportBulkAction 会把数据分发到数据该到的数据节点

public class TransportBulkAction extends HandledTransportAction<BulkRequest, BulkResponse> {

 @Override
    protected void doExecute(Task task, org.elasticsearch.action.bulk.BulkRequest bulkRequest, ActionListener<org.elasticsearch.action.bulk.BulkResponse> listener) {
        /*
         * 这是在传输过程中调用的，因此我们可以快速检查索引内存压力，但我们不想让传输线程保持繁忙。然后，一旦我们有了索引压力，
         * 我们就会分叉到其中一个写入线程池。我们这样做是因为处理批量请求可能会变得昂贵，原因如下：
         * 在将子请求分派给分片时，我们可能需要压缩它们。LZ4 速度超快，但速度足够慢，最好不要在传输线程上执行此操作，尤其是对于大型子请求。
         * 我们可以检测到这些情况，然后才分叉，但这要正确处理起来很复杂，而且分叉的开销相当低。
         */
        final int indexingOps = bulkRequest.numberOfActions();
        final long indexingBytes = bulkRequest.ramBytesUsed();
        final boolean isOnlySystem = isOnlySystem(bulkRequest, clusterService.state().metadata().getIndicesLookup(), systemIndices);
        final Releasable releasable = indexingPressure.markCoordinatingOperationStarted(indexingOps, indexingBytes, isOnlySystem);
        final ActionListener<BulkResponse> releasingListener = ActionListener.runBefore(listener, releasable::close);
        final String executorName = isOnlySystem ? Names.SYSTEM_WRITE : Names.WRITE;
        //通过线程池调用
        threadPool.executor(Names.WRITE).execute(new ActionRunnable<>(releasingListener) {
            @Override
            protected void doRun() {
                doInternalExecute(task, bulkRequest, executorName, releasingListener);
            }
        });
    }

    protected void doInternalExecute(Task task, org.elasticsearch.action.bulk.BulkRequest bulkRequest, String executorName, ActionListener<BulkResponse> listener) {
       //省略代码
        //在开始之前，尝试创建我们在批量处理期间需要的所有索引。
        // Step 1: 收集请求中的所有索引
        final Map<String, Boolean> indices = bulkRequest.requests.stream()  
            //  删除请求不应尝试创建索引（如果索引不存在），除非使用外部版本控制
            .filter(
                request -> request.opType() != DocWriteRequest.OpType.DELETE
                    || request.versionType() == VersionType.EXTERNAL
                    || request.versionType() == VersionType.EXTERNAL_GTE
            )
            .collect(Collectors.toMap(DocWriteRequest::index, DocWriteRequest::isRequireAlias, (v1, v2) -> v1 || v2));

        // Step 2: 筛选索引列表以查找当前不存在的索引。
        final Map<String, IndexNotFoundException> indicesThatCannotBeCreated = new HashMap<>();
        Set<String> autoCreateIndices = new HashSet<>();
        ClusterState state = clusterService.state();
        for (Map.Entry<String, Boolean> indexAndFlag : indices.entrySet()) {
            final String index = indexAndFlag.getKey();
            boolean shouldAutoCreate = indexNameExpressionResolver.hasIndexAbstraction(index, state) == false;
            //只有当我们不要求它是别名时，我们才应该自动创建
            if (shouldAutoCreate && (indexAndFlag.getValue() == false)) {
                autoCreateIndices.add(index);
            }
        }

        // Step 3: 创建所有缺失的索引（如果有任何缺失）。在所有创建返回后启动批量
        if (autoCreateIndices.isEmpty()) {
            executeBulk(task, bulkRequest, startTime, listener, executorName, responses, indicesThatCannotBeCreated);
        } else {     
			//省略代码
			 for (String index : autoCreateIndices) {
				//省略代码，遍历创建索引	
			}
		}

	}
	 void executeBulk(
        Task task,
        BulkRequest bulkRequest,
        long startTimeNanos,
        ActionListener<BulkResponse> listener,
        String executorName,
        AtomicArray<BulkItemResponse> responses,
        Map<String, IndexNotFoundException> indicesThatCannotBeCreated
    ) {
        //创建一个BulkOperation对象，执行doRun方法
        new BulkOperation(task, bulkRequest, listener, executorName, responses, startTimeNanos, indicesThatCannotBeCreated).run();
    }

}

1、把数据按分片分组，按分片分组数据再发送到指定的数据节点

 private final class BulkOperation extends ActionRunnable<BulkResponse> {
	@Override
        protected void doRun() {
           //省略代码
            Metadata metadata = clusterState.metadata();
        
            //按 ShardId -> Operations 映射对请求进行分组
            Map<ShardId, List<BulkItemRequest>> requestsByShard = new HashMap<>();
            //遍历请求的每一条数据
            for (int i = 0; i < bulkRequest.requests.size(); i++) {
                DocWriteRequest<?> docWriteRequest = bulkRequest.requests.get(i);
         		//省略代码
               
                IndexAbstraction ia = null;
                //请求是要把文档加入到索引
                boolean includeDataStreams = docWriteRequest.opType() == DocWriteRequest.OpType.CREATE;
                try {
                    //给定的请求解析索引
                    ia = concreteIndices.resolveIfAbsent(docWriteRequest);
                    //获取具体的写入索引
                    final Index concreteIndex = docWriteRequest.getConcreteWriteIndex(ia, metadata);
                    //判断索引是否关闭
                    if (addFailureIfIndexIsClosed(docWriteRequest, concreteIndex, i, metadata)) {
                        continue;
                    }
                    //获取索引的路由信息，其中返回的indexRouting是new Unpartitioned
                    IndexRouting indexRouting = concreteIndices.routing(concreteIndex);
                    //这里如果文档没有带id，则会给文档生成一个id
                    docWriteRequest.process(indexRouting);
                    //获取分片ID 里面IdAndRoutingOnly调用的是Unpartitioned获取分片id
                    int shardId = docWriteRequest.route(indexRouting);
                    //请求和分片ID封装为BulkItemRequest对象，computeIfAbsent是如果不存在则新建
                    List<BulkItemRequest> shardRequests = requestsByShard.computeIfAbsent(
                        new ShardId(concreteIndex, shardId),
                        shard -> new ArrayList<>()
                    );
                    //并将其添加到requestsByShard中对应的分片请求列表中。
                    shardRequests.add(new BulkItemRequest(i, docWriteRequest));
                } catch (ElasticsearchParseException | IllegalArgumentException | RoutingMissingException | ResourceNotFoundException e) {
                    String name = ia != null ? ia.getName() : docWriteRequest.index();
                    BulkItemResponse.Failure failure = new BulkItemResponse.Failure(name, docWriteRequest.id(), e);
                    BulkItemResponse bulkItemResponse = BulkItemResponse.failure(i, docWriteRequest.opType(), failure);
                    responses.set(i, bulkItemResponse);
                    // make sure the request gets never processed again
                    bulkRequest.requests.set(i, null);
                }
            }
            //没有要添加的数据，直接返回了
            if (requestsByShard.isEmpty()) {
                listener.onResponse(
                    new BulkResponse(responses.toArray(new BulkItemResponse[responses.length()]), buildTookInMillis(startTimeNanos))
                );
                return;
            }
            //下面就知道是按照分片ID分别分发请求
            final AtomicInteger counter = new AtomicInteger(requestsByShard.size());
            String nodeId = clusterService.localNode().getId();
            for (Map.Entry<ShardId, List<BulkItemRequest>> entry : requestsByShard.entrySet()) {
                final ShardId shardId = entry.getKey();
                final List<BulkItemRequest> requests = entry.getValue();
                BulkShardRequest bulkShardRequest = new BulkShardRequest(
                    shardId,
                    bulkRequest.getRefreshPolicy(),
                    requests.toArray(new BulkItemRequest[requests.size()])
                );
                bulkShardRequest.waitForActiveShards(bulkRequest.waitForActiveShards());
                bulkShardRequest.timeout(bulkRequest.timeout());
                bulkShardRequest.routedBasedOnClusterVersion(clusterState.version());
                if (task != null) {
                    bulkShardRequest.setParentTask(nodeId, task.getId());
                }
                client.executeLocally(TransportShardBulkAction.TYPE, bulkShardRequest, new ActionListener<>() {
                	//成功后的响应处理
                    @Override
                    public void onResponse(BulkShardResponse bulkShardResponse) {
                        for (BulkItemResponse bulkItemResponse : bulkShardResponse.getResponses()) {    
                            if (bulkItemResponse.getResponse() != null) {
                                bulkItemResponse.getResponse().setShardInfo(bulkShardResponse.getShardInfo());
                            }
                            responses.set(bulkItemResponse.getItemId(), bulkItemResponse);
                        }
                      //所有的分片请求都完成后，则完成组装响应报文
                        if (counter.decrementAndGet() == 0) {
                            finishHim();
                        }
                    }
					//失败的处理逻辑
                    @Override
                    public void onFailure(Exception e) {
                        // create failures for all relevant requests
                        for (BulkItemRequest request : requests) {
                            final String indexName = request.index();
                            DocWriteRequest<?> docWriteRequest = request.request();
                            BulkItemResponse.Failure failure = new BulkItemResponse.Failure(indexName, docWriteRequest.id(), e);
                            responses.set(request.id(), BulkItemResponse.failure(request.id(), docWriteRequest.opType(), failure));
                        }
                        //所有的分片请求都完成后，则完成组装响应报文
                        if (counter.decrementAndGet() == 0) {
                            finishHim();
                        }
                    }

                    private void finishHim() {
                        listener.onResponse(
                            new BulkResponse(responses.toArray(new BulkItemResponse[responses.length()]), buildTookInMillis(startTimeNanos))
                        );
                    }
                });
            }
            bulkRequest = null; // allow memory for bulk request items to be reclaimed before all items have been completed允许在所有项目完成之前回收批量请求项的内存
        }
}

(1) 计算此文档发往哪个分片

其中上面的关键代码块是下面

 //获取索引的路由信息，其中返回的indexRouting是new Unpartitioned
 IndexRouting indexRouting = concreteIndices.routing(concreteIndex);
 //这里会针对文档的id做一些处理，比如会判断是否存在，不存在是抛异常还是创建一个新的
 docWriteRequest.process(indexRouting);
 //获取分片ID 里面IdAndRoutingOnly调用的是Unpartitioned获取分片id
 int shardId = docWriteRequest.route(indexRouting);

首先indexRouting的对象下面的Partitioned或者Unpartitioned

1)根据索引是否是分区索引，返回不同的索引路由对象

public static IndexRouting fromIndexMetadata(IndexMetadata metadata) {
        if (false == metadata.getRoutingPaths().isEmpty()) {
            return new ExtractFromSource(metadata);
        }
        //代码检查索引元数据是否是分区索引
        //如果是，则创建一个分区索引路由对象（Partitioned）并返回
        if (metadata.isRoutingPartitionedIndex()) {
            return new Partitioned(metadata);
        }
        //以上条件都不满足，则创建一个非分区索引路由对象（Unpartitioned）并返回
        return new Unpartitioned(metadata);
    }

2) 文档没有id会自动给文档创建id

 @Override
    public void process(IndexRouting indexRouting) {
        indexRouting.process(this);
    }

不管Partitioned还是Unpartitioned都继承自IdAndRoutingOnly

private abstract static class IdAndRoutingOnly extends IndexRouting {
 		@Override
        public void process(IndexRequest indexRequest) {
        	//往索引添加文档的id不能为空字符串，但是可以为null，后续会自动创建id
            if ("".equals(indexRequest.id())) {
                throw new IllegalArgumentException("if _id is specified it must not be empty");
            }

            // generate id if not already provided
            if (indexRequest.id() == null) {
                indexRequest.autoGenerateId();
            }
        }
}

public void autoGenerateId() {
        assert id == null;
        assert autoGeneratedTimestamp == UNSET_AUTO_GENERATED_TIMESTAMP : "timestamp has already been generated!";
        assert ifSeqNo == UNASSIGNED_SEQ_NO;
        assert ifPrimaryTerm == UNASSIGNED_PRIMARY_TERM;
        autoGeneratedTimestamp = Math.max(0, System.currentTimeMillis());
        String uid = UUIDs.base64UUID();
        id(uid);
    }

3)根据不同的索引路由对象，id和routing决定此文档发往哪个分片

其中route的接口如下

@Override
    public int route(IndexRouting indexRouting) {
        return indexRouting.indexShard(id, routing, contentType, source);
    }

    private abstract static class IdAndRoutingOnly extends IndexRouting {
	   protected abstract int shardId(String id, @Nullable String routing);
	    @Override
        public int indexShard(String id, @Nullable String routing, XContentType sourceType, BytesReference source) {
            if (id == null) {
                throw new IllegalStateException("id is required and should have been set by process");
            }
            checkRoutingRequired(id, routing);
            return shardId(id, routing);
        }
}

其中shardId有两种实现，分别是Partitioned还是Unpartitioned

 private static class Unpartitioned extends IdAndRoutingOnly {
        Unpartitioned(IndexMetadata metadata) {
            super(metadata);
        }
		//优先routing，如果没有则用id
        @Override
        protected int shardId(String id, @Nullable String routing) {
            return hashToShardId(effectiveRoutingToHash(routing == null ? id : routing));
        }

        @Override
        public void collectSearchShards(String routing, IntConsumer consumer) {
            consumer.accept(hashToShardId(effectiveRoutingToHash(routing)));
        }
    }

    private static class Partitioned extends IdAndRoutingOnly {
        private final int routingPartitionSize;

        Partitioned(IndexMetadata metadata) {
            super(metadata);
            this.routingPartitionSize = metadata.getRoutingPartitionSize();
        }
		//其中routing不能为null
        @Override
        protected int shardId(String id, @Nullable String routing) {
            if (routing == null) {
                throw new IllegalArgumentException("A routing value is required for gets from a partitioned index");
            }
            int offset = Math.floorMod(effectiveRoutingToHash(id), routingPartitionSize);
            return hashToShardId(effectiveRoutingToHash(routing) + offset);
        }

        @Override
        public void collectSearchShards(String routing, IntConsumer consumer) {
            int hash = effectiveRoutingToHash(routing);
            for (int i = 0; i < routingPartitionSize; i++) {
                consumer.accept(hashToShardId(hash + i));
            }
        }
    }

下面只看Unpartitioned

  /**
     * Convert a routing value into a hash.
     * 将路由值转换为哈希值。
     */
    private static int effectiveRoutingToHash(String effectiveRouting) {
        return Murmur3HashFunction.hash(effectiveRouting);
    }

/**
 * Hash function based on the Murmur3 algorithm, which is the default as of Elasticsearch 2.0.
 * 基于 Murmur3 算法的哈希函数，这是 Elasticsearch 2.0 的默认算法。
 */
public final class Murmur3HashFunction {

    private Murmur3HashFunction() {
        // no instance
    }

    public static int hash(String routing) {
        final byte[] bytesToHash = new byte[routing.length() * 2];
        for (int i = 0; i < routing.length(); ++i) {
            final char c = routing.charAt(i);
            final byte b1 = (byte) c, b2 = (byte) (c >>> 8);
            assert ((b1 & 0xFF) | ((b2 & 0xFF) << 8)) == c; // no information loss
            bytesToHash[i * 2] = b1;
            bytesToHash[i * 2 + 1] = b2;
        }
        return hash(bytesToHash, 0, bytesToHash.length);
    }

    public static int hash(byte[] bytes, int offset, int length) {
        return StringHelper.murmurhash3_x86_32(bytes, offset, length, 0);
    }
}

 /**
     * Convert a hash generated from an {@code (id, routing}) pair into a
     * shard id. 将从 {@code （id， routing}） 对生成的哈希转换为分片 ID。
     */
    protected final int hashToShardId(int hash) {
        return Math.floorMod(hash, routingNumShards) / routingFactor;
    }

这样就指定了文档的分片id

(2)、通过taskManager注册Task执行action.execute发送到数据节点

 client.executeLocally(TransportShardBulkAction.TYPE, bulkShardRequest, new ActionListener<>()

 public <Request extends ActionRequest, Response extends ActionResponse> Task executeLocally(
        ActionType<Response> action,
        Request request,
        ActionListener<Response> listener
    ) {
        return taskManager.registerAndExecute(
            "transport",
            transportAction(action),
            request,
            localConnection,
            new SafelyWrappedActionListener<>(listener)
        );
    }

后面的逻辑就不梳理了，直接看TransportShardBulkAction.TYPE

四、数据节点(TransportShardBulkAction)处理处理来自主节点的数据

public class TransportShardBulkAction extends TransportWriteAction<BulkShardRequest, BulkShardRequest, BulkShardResponse> {
 //这里处理属于属于当前节点分片的数据，请求是从主节点上过来的
    @Override
    protected void dispatchedShardOperationOnPrimary(
        BulkShardRequest request,
        IndexShard primary,
        ActionListener<PrimaryResult<BulkShardRequest, BulkShardResponse>> listener
    ) {
        ClusterStateObserver observer = new ClusterStateObserver(clusterService, request.timeout(), logger, threadPool.getThreadContext());
        performOnPrimary(request, primary, updateHelper, threadPool::absoluteTimeInMillis, (update, shardId, mappingListener) -> {
            assert update != null;
            assert shardId != null;
            mappingUpdatedAction.updateMappingOnMaster(shardId.getIndex(), update, mappingListener);
        }, mappingUpdateListener -> observer.waitForNextChange(new ClusterStateObserver.Listener() {
           //省略代码
            }
        }), listener, threadPool, executor(primary), postWriteRefresh, postWriteAction);
    }


}

performOnPrimary 直接看这个

  public static void performOnPrimary(
        org.elasticsearch.action.bulk.BulkShardRequest request,
        IndexShard primary,
        UpdateHelper updateHelper,
        LongSupplier nowInMillisSupplier,
        MappingUpdatePerformer mappingUpdater,
        Consumer<ActionListener<Void>> waitForMappingUpdate,
        ActionListener<PrimaryResult<BulkShardRequest, BulkShardResponse>> listener,
        ThreadPool threadPool,
        String executorName,
        @Nullable PostWriteRefresh postWriteRefresh,
        @Nullable Consumer<Runnable> postWriteAction
    ) {
        new ActionRunnable<>(listener) {

            private final Executor executor = threadPool.executor(executorName);

            private final BulkPrimaryExecutionContext context = new BulkPrimaryExecutionContext(request, primary);

            final long startBulkTime = System.nanoTime();

            @Override
            protected void doRun() throws Exception {
                //只要所有的请求没有执行完
                while (context.hasMoreOperationsToExecute()) {
                    if (executeBulkItemRequest(
                        context,
                        updateHelper,
                        nowInMillisSupplier,
                        mappingUpdater,
                        waitForMappingUpdate,
                        ActionListener.wrap(v -> executor.execute(this), this::onRejection)
                    ) == false) {
                        //我们正在等待另一个线程上的映射更新，一旦完成，它将再次调用此操作，因此我们在这里爆发。
                        return;
                    }
                    assert context.isInitial(); // either completed and moved to next or reset 要么完成并移至下一个，要么重置
                }
                primary.getBulkOperationListener().afterBulk(request.totalSizeInBytes(), System.nanoTime() - startBulkTime); 
                finishRequest();
            }

            @Override
            public void onRejection(Exception e) {
               //省略代码
            }

            private void finishRequest() {
                //省略代码
            }
        }.run();
    }

1、针对此节点上索引分片进行操作

static boolean executeBulkItemRequest(
       BulkPrimaryExecutionContext context,
        UpdateHelper updateHelper,
        LongSupplier nowInMillisSupplier,
        MappingUpdatePerformer mappingUpdater,
        Consumer<ActionListener<Void>> waitForMappingUpdate,
        ActionListener<Void> itemDoneListener
    ) throws Exception {
            //，则获取IndexRequest对象，并创建SourceToParse对象，将相应参数传递给primary的applyIndexOperationOnPrimary方法进行索引操作
            final IndexRequest request = context.getRequestToExecute();
            final SourceToParse sourceToParse = new SourceToParse(
                request.id(),
                request.source(),
                request.getContentType(),
                request.routing(),
                request.getDynamicTemplates()
            );
            //把文档数据保存到分片，返回结果保存结果
            result = primary.applyIndexOperationOnPrimary(
                version,
                request.versionType(),
                sourceToParse,
                request.ifSeqNo(),
                request.ifPrimaryTerm(),
                request.getAutoGeneratedTimestamp(),
                request.isRetry()
            );
      
        //从结果中得到，需要更新索引Mapper的映射，则更新索引的mapper
        if (result.getResultType() == Engine.Result.Type.MAPPING_UPDATE_REQUIRED) {
			//省略代码
        } 
        return true;
    }

(1) 组装Engine.Index

public Engine.IndexResult applyIndexOperationOnPrimary(
        long version,
        VersionType versionType,
        SourceToParse sourceToParse,
        long ifSeqNo,
        long ifPrimaryTerm,
        long autoGeneratedTimestamp,
        boolean isRetry
    ) throws IOException {
        assert versionType.validateVersionForWrites(version);
        //针对索引的操作，包括更新TransLog
        return applyIndexOperation(
            getEngine(),
            UNASSIGNED_SEQ_NO,
            getOperationPrimaryTerm(),
            version,
            versionType,
            ifSeqNo,
            ifPrimaryTerm,
            autoGeneratedTimestamp,
            isRetry,
            Engine.Operation.Origin.PRIMARY,
            sourceToParse
        );
    }

private Engine.IndexResult applyIndexOperation(
        Engine engine,
        long seqNo,
        long opPrimaryTerm,
        long version,
        @Nullable VersionType versionType,
        long ifSeqNo,
        long ifPrimaryTerm,
        long autoGeneratedTimeStamp,
        boolean isRetry,
        Engine.Operation.Origin origin,
        SourceToParse sourceToParse
    ) throws IOException {
        assert opPrimaryTerm <= getOperationPrimaryTerm()
            : "op term [ " + opPrimaryTerm + " ] > shard term [" + getOperationPrimaryTerm() + "]";
        ensureWriteAllowed(origin);
        Engine.Index operation;
        try {
            //组装index
            operation = prepareIndex(
                mapperService,
                sourceToParse,
                seqNo,
                opPrimaryTerm,
                version,
                versionType,
                origin,
                autoGeneratedTimeStamp,
                isRetry,
                ifSeqNo,
                ifPrimaryTerm,
                getRelativeTimeInNanos()
            );
            Mapping update = operation.parsedDoc().dynamicMappingsUpdate();
            if (update != null) {
                return new Engine.IndexResult(update, operation.parsedDoc().id());
            }
        } catch (Exception e) {
           //省略代码
        }

        return index(engine, operation);
    }

 private Engine.IndexResult index(Engine engine, Engine.Index index) throws IOException {
        active.set(true);
        final Engine.IndexResult result;
        final Engine.Index preIndex = indexingOperationListeners.preIndex(shardId, index);
        try {
           	//省略代码
            //InternalEngine.index 逐条写入doc
            // Engine 封装了Lucene和translog的调用,对外提供读写接口.
            result = engine.index(preIndex);
           //省略代码
        } catch (Exception e) {
          	//省略代码
            indexingOperationListeners.postIndex(shardId, preIndex, e);
            throw e;
        }
        indexingOperationListeners.postIndex(shardId, preIndex, result);
        return result;
    }

其中engine.index的子类是InternalEngine.index方法

2 把index数据写入到内存缓冲区和translog

@Override
    public IndexResult index(Index index) throws IOException {
        assert Objects.equals(index.uid().field(), IdFieldMapper.NAME) : index.uid().field();
        final boolean doThrottle = index.origin().isRecovery() == false;
        try (ReleasableLock releasableLock = readLock.acquire()) {
        	 //获取此次的索引策略
              final IndexingStrategy plan = indexingStrategyForOperation(index);
          		//省略代码		 
                    //如果是主分片上的操作，则生成新的Index对象
                    if (index.origin() == Operation.Origin.PRIMARY) {
                        index = new Index(
                            index.uid(),
                            index.parsedDoc(),
                            generateSeqNoForOperationOnPrimary(index),
                            index.primaryTerm(),
                            index.version(),
                            index.versionType(),
                            index.origin(),
                            index.startTime(),
                            index.getAutoGeneratedIdTimestamp(),
                            index.isRetry(),
                            index.getIfSeqNo(),
                            index.getIfPrimaryTerm()
                        );

                        final boolean toAppend = plan.indexIntoLucene && plan.useLuceneUpdateDocument == false;
                        if (toAppend == false) {
                            advanceMaxSeqNoOfUpdatesOnPrimary(index.seqNo());
                        }
                    } else {
                        //其他分片就标记为已见
                        markSeqNoAsSeen(index.seqNo());
                    }

                    if (plan.indexIntoLucene || plan.addStaleOpToLucene) {
                        //把数据更新到Lucene中
                        indexResult = indexIntoLucene(index, plan);
                    } else {
                        indexResult = new IndexResult(
                            plan.versionForIndexing,
                            index.primaryTerm(),
                            index.seqNo(),
                            plan.currentNotFoundOrDeleted,
                            index.id()
                        );
                    }
                }
                if (index.origin().isFromTranslog() == false) {
                    final Translog.Location location;
                    //如果更新Lucene成功，则把索引数据放入到translog中
                    if (indexResult.getResultType() == Result.Type.SUCCESS) {
                        location = translog.add(new Translog.Index(index, indexResult));
                    } 
                    //省略代码
                    indexResult.setTranslogLocation(location);
                }
               //省略代码
                indexResult.setTook(relativeTimeInNanosSupplier.getAsLong() - index.startTime());
                indexResult.freeze();
                return indexResult;
         
       
    }

(1) 获得此次的索引数据处理策略

protected IndexingStrategy indexingStrategyForOperation(final Index index) throws IOException {
        //首先判断索引的来源是否是主要的（PRIMARY）
        if (index.origin() == Operation.Origin.PRIMARY) {
            //planIndexingAsPrimary方法来计划主要索引策略
            return planIndexingAsPrimary(index);
        } else {
            // non-primary mode (i.e., replica or recovery)
            //调用planIndexingAsNonPrimary方法来计划非主要索引策略
            return planIndexingAsNonPrimary(index);
        }
    }

下面就是根据索引的一些信息生成不同的IndexingStrategy逻辑

private IndexingStrategy planIndexingAsPrimary(Index index) throws IOException {
        assert index.origin() == Operation.Origin.PRIMARY : "planing as primary but origin isn't. got " + index.origin();
        final int reservingDocs = index.parsedDoc().docs().size();
        final IndexingStrategy plan;
        //判断是否可以对添加文档的操作进行优化
        final boolean canOptimizeAddDocument = canOptimizeAddDocument(index);
        //如果可以优化添加文档操作且索引之前未被索引过
        if (canOptimizeAddDocument && mayHaveBeenIndexedBefore(index) == false) {
            //则尝试获取正在进行中的文档数，并根据结果确定计划策略
            final Exception reserveError = tryAcquireInFlightDocs(index, reservingDocs);
            //如果获取正在进行中的文档数遇到错误，则使用失败策略，将错误信息和索引ID作为参数构造 IndexingStrategy 对象。
            if (reserveError != null) {
                plan = IndexingStrategy.failAsTooManyDocs(reserveError, index.id());
            } else {
             //否则，使用优化的追加模式构造 IndexingStrategy 对象
                plan = IndexingStrategy.optimizedAppendOnly(1L, reservingDocs);
            }
        } else {
           	//省略代码
             //尝试获取正在进行中的文档，并检查是否有错误发生。
             final Exception reserveError = tryAcquireInFlightDocs(index, reservingDocs);
             if (reserveError != null) {//如果有错误发生，则创建相应的计划策略。
                   plan = IndexingStrategy.failAsTooManyDocs(reserveError, index.id());
             } else {
                //根据当前未找到或已删除状态、是否可以优化添加文档、当前版本和索引的版本信息，创建相应的计划策略
                plan = IndexingStrategy.processNormally(
                     currentNotFoundOrDeleted,
                     canOptimizeAddDocument ? 1L : index.versionType().updateVersion(currentVersion, index.version()),
                     reservingDocs
                  );
              
            }
        return plan;
    }

下面是IndexingStrategy的构造方法

 private IndexingStrategy(
            boolean currentNotFoundOrDeleted,
            boolean useLuceneUpdateDocument,
            boolean indexIntoLucene,
            boolean addStaleOpToLucene,
            long versionForIndexing,
            int reservedDocs,
            IndexResult earlyResultOnPreFlightError
        ) {
           //省略代码
        }

其中注意下面两种生成

 static IndexingStrategy optimizedAppendOnly(long versionForIndexing, int reservedDocs) {
            return new IndexingStrategy(true, false, true, false, versionForIndexing, reservedDocs, null);
 }
  static IndexingStrategy processNormally(boolean currentNotFoundOrDeleted, long versionForIndexing, int reservedDocs) {
            return new IndexingStrategy(
                currentNotFoundOrDeleted,
                currentNotFoundOrDeleted == false,
                true,
                false,
                versionForIndexing,
                reservedDocs,
                null
            );
 }

其中indexIntoLucene和addStaleOpToLucene字段决定是否执行写入下面第二步方法的关键

 if (plan.indexIntoLucene || plan.addStaleOpToLucene) {
     //把数据更新到Lucene中
    indexResult = indexIntoLucene(index, plan);
 } 

(2) 根据策略选择不同的API写入Es缓冲区(indexWriter)

在ES缓冲区中的数据，是不能被检索的，需要_refresh把缓冲区的数据刷到Lucene 才可以被检索

import org.apache.lucene.index.IndexWriter;

 private IndexResult indexIntoLucene(Index index, IndexingStrategy plan) throws IOException {
        try {
            if (plan.addStaleOpToLucene) { //添加
                addStaleDocs(index.docs(), indexWriter);
            } else if (plan.useLuceneUpdateDocument) { //更新
                assert assertMaxSeqNoOfUpdatesIsAdvanced(index.uid(), index.seqNo(), true, true);
                updateDocs(index.uid(), index.docs(), indexWriter);
            } else {
                //文档不存在，我们可以优化创建，但请仔细检查断言是否正在运行
                assert assertDocDoesNotExist(index, canOptimizeAddDocument(index) == false);
                addDocs(index.docs(), indexWriter);
            }
            return new IndexResult(plan.versionForIndexing, index.primaryTerm(), index.seqNo(), plan.currentNotFoundOrDeleted, index.id());
        } catch (Exception ex) {
          //省略代码
        }
    }
	//这是第一种，添加文档，一般是副本分片添加文档用到的
     private void addStaleDocs(final List<LuceneDocument> docs, final IndexWriter indexWriter) throws IOException {
        for (LuceneDocument doc : docs) {
            doc.add(softDeletesField); // 在添加到 Lucene 之前软删除了每个文档的标志
        }
        if (docs.size() > 1) {
            indexWriter.addDocuments(docs);
        } else {
            indexWriter.addDocument(docs.get(0));
        }
    }
    //这是第二种，修改文档
     private void updateDocs(final Term uid, final List<LuceneDocument> docs, final IndexWriter indexWriter) throws IOException {
        if (docs.size() > 1) {
            indexWriter.softUpdateDocuments(uid, docs, softDeletesField);
        } else {
            indexWriter.softUpdateDocument(uid, docs.get(0), softDeletesField);
        }
        numDocUpdates.inc(docs.size());
    }
    //这是第三种 添加文档
    private void addDocs(final List<LuceneDocument> docs, final IndexWriter indexWriter) throws IOException {
        if (docs.size() > 1) {
            indexWriter.addDocuments(docs);
        } else {
            indexWriter.addDocument(docs.get(0));
        }
        numDocAppends.inc(docs.size());
    }

(3)上诉第二步成功后，再写入translog

在写入到transLog日志中，会先转成new Translog.Index 再添加到translog

    public Location add(final Operation operation) throws IOException {
         final ReleasableBytesStreamOutput out = new ReleasableBytesStreamOutput(bigArrays);
		try {
            writeOperationWithSize(out, operation);
            final BytesReference bytes = out.bytes();
            try (ReleasableLock ignored = readLock.acquire()) {
                ensureOpen();
               //省略代码
                return current.add(bytes, operation.seqNo());
            }
        }
}

    private ReleasableBytesStreamOutput buffer;

 /**
     *将给定的字节添加到具有指定序列号的转录日志中;返回字节写入到的位置。
     * @param data  the bytes to write 要写入的字节数
     * @param seqNo the sequence number associated with the operation 与操作关联的序列号
     * @return the location the bytes were written to 字节写入到的位置
     * @throws IOException if writing to the translog resulted in an I/O exception
     */
    public Translog.Location add(final BytesReference data, final long seqNo) throws IOException {
        //首先检查缓冲的字节数是否超过了forceWriteThreshold阈值，如果超过了，则调用writeBufferedOps方法将缓冲的操作写入。
        long bufferedBytesBeforeAdd = this.bufferedBytes;
        if (bufferedBytesBeforeAdd >= forceWriteThreshold) {
            writeBufferedOps(Long.MAX_VALUE, bufferedBytesBeforeAdd >= forceWriteThreshold * 4);
        }

        final Translog.Location location;
        synchronized (this) {
            ensureOpen();
            //代码确保buffer不为null，
            if (buffer == null) {
                buffer = new ReleasableBytesStreamOutput(bigArrays);
            }
            //数据写入缓冲区。然后更新minSeqNo和maxSeqNo的值
            assert bufferedBytes == buffer.size();
            final long offset = totalOffset;
            totalOffset += data.length();
            data.writeTo(buffer);

            assert minSeqNo != SequenceNumbers.NO_OPS_PERFORMED || operationCounter == 0;
            assert maxSeqNo != SequenceNumbers.NO_OPS_PERFORMED || operationCounter == 0;

            minSeqNo = SequenceNumbers.min(minSeqNo, seqNo);
            maxSeqNo = SequenceNumbers.max(maxSeqNo, seqNo);
            //并将seqNo添加到nonFsyncedSequenceNumbers中。操作计数器递增
            nonFsyncedSequenceNumbers.add(seqNo);

            operationCounter++;

            assert assertNoSeqNumberConflict(seqNo, data);
            //然后使用generation、offset和数据长度创建一个Translog.Location对象。
            location = new Translog.Location(generation, offset, data.length());
            //调用operationListener.operationAdded方法通知操作监听器有新的操作添加，并更新bufferedBytes的值。
            operationListener.operationAdded(data, seqNo, location);
            bufferedBytes = buffer.size();
        }

        return location;
    }

介绍一下Translog类

/**Translog 是每个索引的分片组件，它以持久的方式记录所有未提交的索引操作。
在 Elasticsearch 中，每个 {@link org.elasticsearch.index.engine.InternalEngine} 都有一个 Translog 实例。
此外，从 Elasticsearch 2.0 开始，引擎还会在每次提交时记录一个 {@link *TRANSLOG_UUID_KEY}，以确保 lucene 索引与事务日志文件之间的强关联。
此 UUID 用于防止从属于其他引擎的事务日志中意外恢复。

每个 Translog 只有一个 translog 文件打开，供 translog 生成 ID 随时引用的写入。
此 ID 将写入 {@code translog.ckp} 文件，该文件旨在适合单个磁盘块，因此文件的写入是原子的。
检查点文件在 translog 的每个 fsync 操作上写入，并记录写入的操作数、当前 translog 的文件生成、其 fsync 偏移量（以字节为单位）以及其他重要统计信息。

当当前转录文件达到特定大小 （{@link IndexSettingsINDEX_TRANSLOG_GENERATION_THRESHOLD_SIZE_SETTING}） 时，或者当新旧操作之间明确分离时（在主要术语更改时），
将重新打开当前文件进行只读，并创建一个新的只写文件。
任何非最新的、只读的 translog 文件总是有一个与之关联的 {@code translog-{gen}.ckp}，它是其上一个 {@code translog.ckp} 的同步副本，因此在灾难恢复中，最后一个 fsync 偏移量、操作数等仍会保留。
**/
public class Translog extends AbstractIndexShardComponent implements IndexShardComponent, Closeable {
}