elasticsearch-ThreadPool源码

1.ES的线程池类型

  ThreadPool的内部类Names定义了ES线程池的所有类型，如下所示。利用静态类定义线程名称，类似于枚举

  public static class Names {
        public static final String SAME = "same";
        public static final String GENERIC = "generic";
        public static final String LISTENER = "listener";
        public static final String GET = "get";
        public static final String INDEX = "index";
        public static final String BULK = "bulk";
        public static final String SEARCH = "search";
        public static final String SUGGEST = "suggest";
        public static final String PERCOLATE = "percolate";
        public static final String MANAGEMENT = "management";
        public static final String FLUSH = "flush";
        public static final String MERGE = "merge";
        public static final String REFRESH = "refresh";
        public static final String WARMER = "warmer";
        public static final String SNAPSHOT = "snapshot";
        public static final String OPTIMIZE = "optimize";
    }

2.ES的线程池设置

     线程池的大小默认设置都是根据CPU的核数计算的，核数可以设置也可以自己获取；

     另外设置了线程池的queue的大小的默认值。

     官方的推荐不修改ES线程池的配置，但是实际的情况还是有些不满足实际需求，需要动态的调整。

 int availableProcessors = EsExecutors.boundedNumberOfProcessors(settings);

        defaultExecutorTypeSettings = ImmutableMap.<String, Settings>builder()
                .put(Names.GENERIC, settingsBuilder().put("type", "cached").put("keep_alive", "30s").build())
                .put(Names.INDEX, settingsBuilder().put("type", "fixed").put("size", availableProcessors).put("queue_size", 200).build())
                .put(Names.BULK, settingsBuilder().put("type", "fixed").put("size", availableProcessors).put("queue_size", 50).build())
                .put(Names.GET, settingsBuilder().put("type", "fixed").put("size", availableProcessors).put("queue_size", 1000).build())
                .put(Names.SEARCH, settingsBuilder().put("type", "fixed").put("size", availableProcessors * 3).put("queue_size", 1000).build())
                .put(Names.SUGGEST, settingsBuilder().put("type", "fixed").put("size", availableProcessors).put("queue_size", 1000).build())
                .put(Names.PERCOLATE, settingsBuilder().put("type", "fixed").put("size", availableProcessors).put("queue_size", 1000).build())
                .put(Names.MANAGEMENT, settingsBuilder().put("type", "scaling").put("keep_alive", "5m").put("size", 5).build())
                        // no queue as this means clients will need to handle rejections on listener queue even if the operation succeeded
                        // the assumption here is that the listeners should be very lightweight on the listeners side
                .put(Names.LISTENER, settingsBuilder().put("type", "fixed").put("size", halfProcMaxAt10).build())
                .put(Names.FLUSH, settingsBuilder().put("type", "scaling").put("keep_alive", "5m").put("size", halfProcMaxAt5).build())
                .put(Names.MERGE, settingsBuilder().put("type", "scaling").put("keep_alive", "5m").put("size", halfProcMaxAt5).build())
                .put(Names.REFRESH, settingsBuilder().put("type", "scaling").put("keep_alive", "5m").put("size", halfProcMaxAt10).build())
                .put(Names.WARMER, settingsBuilder().put("type", "scaling").put("keep_alive", "5m").put("size", halfProcMaxAt5).build())
                .put(Names.SNAPSHOT, settingsBuilder().put("type", "scaling").put("keep_alive", "5m").put("size", halfProcMaxAt5).build())
                .put(Names.OPTIMIZE, settingsBuilder().put("type", "fixed").put("size", 1).build())
                .build();