码迷,mamicode.com
首页 > 其他好文 > 详细

elasticSearch6源码分析(3)cluster模块

时间:2018-12-11 17:22:04      阅读:259      评论:0      收藏:0      [点我收藏+]

标签:parse   desc   single   routing   new   sch   objects   progress   mic   

1. cluser概述

One of the main roles of the master is to decide which shards to allocate to which nodes, and when to move shards between nodes in order to rebalance the cluster.

2. ClusterModule模块的作用

Configures classes and services that affect the entire cluster

3.重点类介绍:

  3.1 配置类Decider

  各种Decider关系如下

 

技术分享图片

以EnableAllocationDecider.java为例:

    public static final Setting<Allocation> CLUSTER_ROUTING_ALLOCATION_ENABLE_SETTING =
        new Setting<>("cluster.routing.allocation.enable", Allocation.ALL.toString(), Allocation::parse,
            Property.Dynamic, Property.NodeScope);
    public static final Setting<Allocation> INDEX_ROUTING_ALLOCATION_ENABLE_SETTING =
        new Setting<>("index.routing.allocation.enable", Allocation.ALL.toString(), Allocation::parse,
            Property.Dynamic, Property.IndexScope);

    public static final Setting<Rebalance> CLUSTER_ROUTING_REBALANCE_ENABLE_SETTING =
        new Setting<>("cluster.routing.rebalance.enable", Rebalance.ALL.toString(), Rebalance::parse,
            Property.Dynamic, Property.NodeScope);
    public static final Setting<Rebalance> INDEX_ROUTING_REBALANCE_ENABLE_SETTING =
        new Setting<>("index.routing.rebalance.enable", Rebalance.ALL.toString(), Rebalance::parse,
            Property.Dynamic, Property.IndexScope);

    private volatile Rebalance enableRebalance;
    private volatile Allocation enableAllocation;

 

Decider的使用

ClusterModule.java
 public static Collection<AllocationDecider> createAllocationDeciders(Settings settings, ClusterSettings clusterSettings,
                                                                         List<ClusterPlugin> clusterPlugins) {
        // collect deciders by class so that we can detect duplicates
        Map<Class, AllocationDecider> deciders = new LinkedHashMap<>();
        addAllocationDecider(deciders, new MaxRetryAllocationDecider(settings));
        addAllocationDecider(deciders, new ResizeAllocationDecider(settings));
        addAllocationDecider(deciders, new ReplicaAfterPrimaryActiveAllocationDecider(settings));
        addAllocationDecider(deciders, new RebalanceOnlyWhenActiveAllocationDecider(settings));
        addAllocationDecider(deciders, new ClusterRebalanceAllocationDecider(settings, clusterSettings));
        addAllocationDecider(deciders, new ConcurrentRebalanceAllocationDecider(settings, clusterSettings));
        addAllocationDecider(deciders, new EnableAllocationDecider(settings, clusterSettings));
        addAllocationDecider(deciders, new NodeVersionAllocationDecider(settings));
        addAllocationDecider(deciders, new SnapshotInProgressAllocationDecider(settings));
        addAllocationDecider(deciders, new RestoreInProgressAllocationDecider(settings));
        addAllocationDecider(deciders, new FilterAllocationDecider(settings, clusterSettings));
        addAllocationDecider(deciders, new SameShardAllocationDecider(settings, clusterSettings));
        addAllocationDecider(deciders, new DiskThresholdDecider(settings, clusterSettings));
        addAllocationDecider(deciders, new ThrottlingAllocationDecider(settings, clusterSettings));
        addAllocationDecider(deciders, new ShardsLimitAllocationDecider(settings, clusterSettings));
        addAllocationDecider(deciders, new AwarenessAllocationDecider(settings, clusterSettings));

        clusterPlugins.stream()
            .flatMap(p -> p.createAllocationDeciders(settings, clusterSettings).stream())
            .forEach(d -> addAllocationDecider(deciders, d));

        return deciders.values();
    }

3.2 配置类Allocator

实现类ShardsAllocator

使用

    private static ShardsAllocator createShardsAllocator(Settings settings, ClusterSettings clusterSettings,
                                                         List<ClusterPlugin> clusterPlugins) {
        Map<String, Supplier<ShardsAllocator>> allocators = new HashMap<>();
        allocators.put(BALANCED_ALLOCATOR, () -> new BalancedShardsAllocator(settings, clusterSettings));

        for (ClusterPlugin plugin : clusterPlugins) {
            plugin.getShardsAllocators(settings, clusterSettings).forEach((k, v) -> {
                if (allocators.put(k, v) != null) {
                    throw new IllegalArgumentException("ShardsAllocator [" + k + "] already defined");
                }
            });
        }
        String allocatorName = SHARDS_ALLOCATOR_TYPE_SETTING.get(settings);
        Supplier<ShardsAllocator> allocatorSupplier = allocators.get(allocatorName);
        if (allocatorSupplier == null) {
            throw new IllegalArgumentException("Unknown ShardsAllocator [" + allocatorName + "]");
        }
        return Objects.requireNonNull(allocatorSupplier.get(),
            "ShardsAllocator factory for [" + allocatorName + "] returned null");
    }

3.3 服务类

ClusterService.java
    @Override
    protected synchronized void doStart() {
        clusterApplierService.start();
        masterService.start();
    }

分别调用

2.3.1clusterApplierService

    @Override
    protected synchronized void doStart() {
        Objects.requireNonNull(nodeConnectionsService, "please set the node connection service before starting");
        Objects.requireNonNull(state.get(), "please set initial state before starting");
        addListener(localNodeMasterListeners);
        threadPoolExecutor = EsExecutors.newSinglePrioritizing(
                nodeName + "/" + CLUSTER_UPDATE_THREAD_NAME,
                daemonThreadFactory(settings, CLUSTER_UPDATE_THREAD_NAME),
                threadPool.getThreadContext(),
                threadPool.scheduler());
    }

3.3.2 masterService

    @Override
    protected synchronized void doStart() {
        Objects.requireNonNull(clusterStatePublisher, "please set a cluster state publisher before starting");
        Objects.requireNonNull(clusterStateSupplier, "please set a cluster state supplier before starting");
        threadPoolExecutor = EsExecutors.newSinglePrioritizing(
                nodeName + "/" + MASTER_UPDATE_THREAD_NAME,
                daemonThreadFactory(settings, MASTER_UPDATE_THREAD_NAME),
                threadPool.getThreadContext(),
                threadPool.scheduler());
        taskBatcher = new Batcher(logger, threadPoolExecutor);
    }

 

elasticSearch6源码分析(3)cluster模块

标签:parse   desc   single   routing   new   sch   objects   progress   mic   

原文地址:https://www.cnblogs.com/davidwang456/p/10102927.html

(0)
(0)
   
举报
评论 一句话评论(0
登录后才能评论!
© 2014 mamicode.com 版权所有  联系我们:gaon5@hotmail.com
迷上了代码!