和MySQL使用 LIMIT 关键字返回只有一页的结果一样，Elasticsearch接受 from 和 size 参数：

size: 结果数，默认10 
from: 跳过开始的结果数，默认0

如果想每页显示5个结果，页码从1到3，那请求如下：

GET /_search?size=5 
GET /_search?size=5&from=5 
GET /_search?size=5&from=10

注意：应该多注意分页太深或者一次请求太多的结果。结果在返回前会被排序。但是记住一个搜索请求常常涉及多个分片。每个分片生成自己排好序的结果，它们接着需要集中起来排序以确保整体排序正确。

当前数据库中的数据如下：

GET 

#响应结果
{
    "took": 3,
    "timed_out": false,
    "_shards": {
        "total": 1,
        "successful": 1,
        "skipped": 0,
        "failed": 0
    },
    "hits": {
        "total": {
            "value": 4,
            "relation": "eq"
        },
        "max_score": 1.0,
        "hits": [
            {
                "_index": "test",
                "_type": "user",
                "_id": "QHi1UoIBpyNh4YQ4T1Sq",
                "_score": 1.0,
                "_source": {
                    "id": 1001,
                    "name": "张三",
                    "age": 20,
                    "sex": "男"
                }
            }
        ]
    }
}

GET 

# 响应数据
{
    "took": 1,
    "timed_out": false,
    "_shards": {
        "total": 1,
        "successful": 1,
        "skipped": 0,
        "failed": 0
    },
    "hits": {
        "total": {
            "value": 4,
            "relation": "eq"
        },
        "max_score": 1.0,
        "hits": [
            {
                "_index": "test",
                "_type": "user",
                "_id": "1002",
                "_score": 1.0,
                "_source": {
                    "id": 1002,
                    "name": "李四",
                    "age": 23,
                    "sex": "女"
                }
            },
            {
                "_index": "test",
                "_type": "user",
                "_id": "1003",
                "_score": 1.0,
                "_source": {
                    "id": 1003,
                    "name": "王五",
                    "age": 27,
                    "sex": "男"
                }
            }
        ]
    }
}

# 当数据量不够时
GET 

#响应数据
{
    "took": 1,
    "timed_out": false,
    "_shards": {
        "total": 1,
        "successful": 1,
        "skipped": 0,
        "failed": 0
    },
    "hits": {
        "total": {
            "value": 4,
            "relation": "eq"
        },
        "max_score": 1.0,
        "hits": [
            {
                "_index": "test",
                "_type": "user",
                "_id": "1002",
                "_score": 1.0,
                "_source": {
                    "id": 1002,
                    "name": "李四",
                    "age": 23,
                    "sex": "女"
                }
            },
            {
                "_index": "test",
                "_type": "user",
                "_id": "1003",
                "_score": 1.0,
                "_source": {
                    "id": 1003,
                    "name": "王五",
                    "age": 27,
                    "sex": "男"
                }
            },
            {
                "_index": "test",
                "_type": "user",
                "_id": "1004",
                "_score": 1.0,
                "_source": {
                    "id": 1004,
                    "name": "赵六",
                    "age": 29,
                    "sex": "女"
                }
            }
        ]
    }
}

在集群系统中深度分页

        为了理解为什么深度分页是有问题的，让我们假设在一个有5个主分片的索引中搜索。当请求结果的第一 页（结果1到10）时，每个分片产生自己最顶端10个结果然后返回它们给请求节点(requesting node)，它再排序这所有的50个结果以选出顶端的10个结果。

        现在假设我们请求第1000页——结果10001到10010。工作方式都相同，不同的是每个分片都必须产生顶端的 10010个结果。然后请求节点排序这50050个结果并丢弃50040个！

        可以看到在分布式系统中，排序结果的花费随着分页的深入而成倍增长。这也是为什么网络搜索引擎中任何语句不能返回多于1000个结果的原因。如微博前端页面返回的页数只有50页，用工具是可以拿到50页以后的数据的。