Elasticsearch——mget及bulk

1. mget批量查詢

1.1 批量查詢的好處

若是查詢100條數據,一條一條的查的話,就須要發送100條數據,若是進行批量查詢的話,只須要發送一次網絡請求。java

通常來講,在進行查詢的時候,若是一次性要查詢多條數據的話,那麼必定要用batch批量操做的api 儘量減小網絡開銷次數,可能能夠將性能提高數倍,甚至數十倍,很是很是之重要node

1.2 語法

一條一條的查詢json

GET test_index/test_type/1
GET test_index/test_type/2
返回
{
  "_index": "test_index",
  "_type": "test_type",
  "_id": "1",
  "_version": 1,
  "found": true,
  "_source": {
    "test_field": "create id by myself"
  }
}
複製代碼

mget批量查詢api

GET /_mget
{
  "docs": [
    {
      "_index": "test_index",
    "_type": "test_type",
    "_id": "1"
    },
    {
      "_index": "test_index",
    "_type": "test_type",
    "_id": "2"
    }
  ]
}
返回結果
{
  "docs": [
    {
      "_index": "test_index",
      "_type": "test_type",
      "_id": "1",
      "_version": 1,
      "found": true,
      "_source": {
        "test_field": "create id by myself"
      }
    },
    {
      "_index": "test_index",
      "_type": "test_type",
      "_id": "2",
      "_version": 1,
      "found": true,
      "_source": {
        "name": "Tom",
        "age": 12,
        "gender": "M"
      }
    }
  ]
}
複製代碼

若是查詢的document是一個index下的不一樣type種的話數組

GET /test_index/_mget
{
   "docs" : [
      {
         "_type" :  "test_type",
         "_id" :    1
      },
      {
         "_type" :  "test_type",
         "_id" :    2
      }
   ]
}
複製代碼

若是查詢的數據都在同一個index下的同一個type下,最簡單了性能優化

GET /test_index/test_type/_mget
{
   "ids": [1, 2]
}
複製代碼

2. bulk批量增刪改

2.1 語法

每一個操做須要兩個 json 串,語法以下:網絡

{"action": {"metadata"}}
{"data"}
複製代碼

舉例,好比你如今要建立一個文檔,放bulk裏面,看起來會是這樣子的:數據結構

{"index": {"_index": "test_index", "_type", "test_type", "_id": "1"}}
{"test_field1": "test1", "test_field2": "test2"}
複製代碼

bulk api 對 json 的語法,有嚴格的要求,每一個json串不能換行,只能放一行,同時一個json串和一個json串之間,必須有一個換行jvm

單個json串裏面有換行的話,會報錯:elasticsearch

{
  "error": {
    "root_cause": [
      {
        "type": "json_e_o_f_exception",
        "reason": "Unexpected end-of-input: expected close marker for Object (start marker at [Source: org.elasticsearch.transport.netty4.ByteBufStreamInput@79a526fa; line: 1, column: 1])\n at [Source: org.elasticsearch.transport.netty4.ByteBufStreamInput@79a526fa; line: 1, column: 3]"
      }
    ],
    "type": "json_e_o_f_exception",
    "reason": "Unexpected end-of-input: expected close marker for Object (start marker at [Source: org.elasticsearch.transport.netty4.ByteBufStreamInput@79a526fa; line: 1, column: 1])\n at [Source: org.elasticsearch.transport.netty4.ByteBufStreamInput@79a526fa; line: 1, column: 3]"
  },
  "status": 500
}
複製代碼

2.2 可執行的操做

  1. delete:刪除一個文檔,只要1個json串就能夠了
  2. create:PUT /index/type/id/_create,強制建立
  3. index:普通的put操做,能夠是建立文檔,也能夠是全量替換文檔
  4. update:執行的 partial update 操做

2.3 示例

POST /_bulk
{"delete": {"_index": "test_index", "_type": "test_type", "_id": "2"}}
{"create": {"_index": "test_index", "_type": "test_type", "_id":6}}
{"test_field": "create id 6"}
{"index": {"_index": "test_index", "_type": "test_type", "_id": 7}}
{"test_field": "put id 7"}
{"update": {"_index": "test_index", "_type": "test_type", "_id": 1}}
{"doc": {"test_field": "update id 1"}}
返回結果:
{
  "took": 62,
  "errors": false,
  "items": [
    {
      "delete": {
        "found": true,
        "_index": "test_index",
        "_type": "test_type",
        "_id": "2",
        "_version": 2,
        "result": "deleted",
        "_shards": {
          "total": 2,
          "successful": 1,
          "failed": 0
        },
        "status": 200
      }
    },
    {
      "create": {
        "_index": "test_index",
        "_type": "test_type",
        "_id": "6",
        "_version": 1,
        "result": "created",
        "_shards": {
          "total": 2,
          "successful": 1,
          "failed": 0
        },
        "created": true,
        "status": 201
      }
    },
    {
      "index": {
        "_index": "test_index",
        "_type": "test_type",
        "_id": "7",
        "_version": 1,
        "result": "created",
        "_shards": {
          "total": 2,
          "successful": 1,
          "failed": 0
        },
        "created": true,
        "status": 201
      }
    },
    {
      "update": {
        "_index": "test_index",
        "_type": "test_type",
        "_id": "1",
        "_version": 2,
        "result": "updated",
        "_shards": {
          "total": 2,
          "successful": 1,
          "failed": 0
        },
        "status": 200
      }
    }
  ]
}
複製代碼

bulk操做中,任意一個操做失敗,是不會影響其餘的操做的,可是在返回結果裏,會告訴你異常日誌

上面咱們已經create了 _id 爲6的數據,咱們再create一次,確定會報錯,執行如下語句:

POST /_bulk
{"delete": {"_index": "test_index", "_type": "test_type", "_id": "2"}}
{"create": {"_index": "test_index", "_type": "test_type", "_id":6}}
{"test_field": "create id 6"}
{"index": {"_index": "test_index", "_type": "test_type", "_id": 9}}
{"test_field": "put id 9"}
{"update": {"_index": "test_index", "_type": "test_type", "_id": 1}}
{"doc": {"test_field": "update id 1"}}
返回結果:
{
  "took": 10,
  "errors": true,
  "items": [
    {
      "delete": {
        "found": false,
        "_index": "test_index",
        "_type": "test_type",
        "_id": "2",
        "_version": 1,
        "result": "not_found",
        "_shards": {
          "total": 2,
          "successful": 1,
          "failed": 0
        },
        "status": 404
      }
    },
    {
      "create": {
        "_index": "test_index",
        "_type": "test_type",
        "_id": "6",
        "status": 409,
        "error": {
          "type": "version_conflict_engine_exception",
          "reason": "[test_type][6]: version conflict, document already exists (current version [1])",
          "index_uuid": "rsiZYqiwSCC2XdR8N2bJow",
          "shard": "2",
          "index": "test_index"
        }
      }
    },
    {
      "index": {
        "_index": "test_index",
        "_type": "test_type",
        "_id": "9",
        "_version": 2,
        "result": "updated",
        "_shards": {
          "total": 2,
          "successful": 1,
          "failed": 0
        },
        "created": false,
        "status": 200
      }
    },
    {
      "update": {
        "_index": "test_index",
        "_type": "test_type",
        "_id": "1",
        "_version": 2,
        "result": "noop",
        "_shards": {
          "total": 2,
          "successful": 1,
          "failed": 0
        },
        "status": 200
      }
    }
  ]
}
複製代碼

能夠看到返回結果中 create 報錯

若是修改的是同一個index, 同一個index和同一個type,下面的語法也能夠:

POST /test_index/_bulk
{ "delete": { "_type": "test_type", "_id": "3" }} 
{ "create": { "_type": "test_type", "_id": "12" }}
{ "test_field":    "test12" }
{ "index":  { "_type": "test_type" }}
{ "test_field":    "auto-generate id test" }
{ "index":  { "_type": "test_type", "_id": "2" }}
{ "test_field":    "replaced test2" }
{ "update": { "_type": "test_type", "_id": "1", "_retry_on_conflict" : 3} }
{ "doc" : {"test_field2" : "bulk test1"} }

POST /test_index/test_type/_bulk
{ "delete": { "_id": "3" }} 
{ "create": { "_id": "12" }}
{ "test_field":    "test12" }
{ "index":  { }}
{ "test_field":    "auto-generate id test" }
{ "index":  { "_id": "2" }}
{ "test_field":    "replaced test2" }
{ "update": { "_id": "1", "_retry_on_conflict" : 3} }
{ "doc" : {"test_field2" : "bulk test1"} }
複製代碼

2.4 bulk size 最佳大小

bulk request會加載到內存裏,若是太大的話,性能反而會降低,所以須要反覆嘗試一個最佳的bulk size。通常從1000~5000條數據開始,嘗試逐漸增長。另外,若是看大小的話,最好是在5~15MB之間。

2.5 _bulk api的奇特json格式與底層性能優化關係

bulk api奇特的json格式

{"action": {"meta"}}
{"data"}
{"action": {"meta"}}
{"data"}
複製代碼

爲何不是下面這種格式

[{
  "action": {
 
  },
  "data": {

  }
}]
複製代碼
  1. bulk中的每一個操做均可能要轉發到不一樣的node的shard去執行

  2. 若是採用比較良好的json數組格式

    容許任意的換行,整個可讀性很是棒,讀起來很爽,es拿到那種標準格式的json串之後,要按照下述流程去進行處理

    • 將json數組解析爲JSONArray對象,這個時候,整個數據,就會在內存中出現一份如出一轍的拷貝,一份數據是json文本,一份數據是JSONArray對象
    • 解析json數組裏的每一個json,對每一個請求中的document進行路由
    • 爲路由到同一個shard上的多個請求,建立一個請求數組
    • 將這個請求數組序列化
    • 將序列化後的請求數組發送到對應的節點上去
  3. 耗費更多內存,更多的jvm gc開銷

    咱們以前提到過bulk size最佳大小的那個問題,通常建議說在幾千條那樣,而後大小在10MB左右,因此說,可怕的事情來了。假設說如今100個bulk請求發送到了一個節點上去,而後每一個請求是10MB,100個請求,就是1000MB = 1GB,而後每一個請求的json都copy一份爲jsonarray對象,此時內存中的佔用就會翻倍,就會佔用2GB的內存,甚至還不止。由於弄成jsonarray以後,還可能會多搞一些其餘的數據結構,2GB+的內存佔用。

    佔用更多的內存可能就會積壓其餘請求的內存使用量,好比說最重要的搜索請求,分析請求,等等,此時就可能會致使其餘請求的性能急速降低 另外的話,佔用內存更多,就會致使java虛擬機的垃圾回收次數更多,跟頻繁,每次要回收的垃圾對象更多,耗費的時間更多,致使es的java虛擬機中止工做線程的時間更多

  4. 如今的奇特格式

    {"action": {"meta"}}
    {"data"}
    {"action": {"meta"}}
    {"data"}
    複製代碼
    • 不用將其轉換爲json對象,不會出現內存中的相同數據的拷貝,直接按照換行符切割json
    • 對每兩個一組的json,讀取meta,進行document路由
    • 直接將對應的json發送到node上去
  5. 最大的優點在於,不須要將json數組解析爲一個JSONArray對象,造成一份大數據的拷貝,浪費內存空間,儘量地保證性能

相關文章
相關標籤/搜索