Optimistic Concurrency Control

乐观并发控制，即乐观锁。乐观锁是一种轻量级的实现并发控制的思想，乐观锁的实现一般依靠version版本号。比如在mysql中，我们在建表的时候常常额外新增一个version字段，在更新某行数据时对version字段值进行比对，以此来判断在此期间是否有他人更新该行数据。

代码语言：javascript复制

UPDATE table_name
SET    field = ${value},
       version = version   1
WHERE  id = ${id}
AND    version = ${version}

在6.7版本之前，Elasticsearch也是借助于version版本号来实现乐观并发控制的，但由于Elasticsearch的分布式特性，version版本号并不能唯一标识主本分片和其副本分片中同一文档的版本。具体地，索引中的文档都是根据路由规则分散存储到主本分片中的，而每个主本分片又会有若干副本分片，每个副本分片只会对应一个主本分片；当针对一个文档进行index、update或delete操作时，主本分片会将该文档异步复制到其所有副本分片中；正常情况下，主本分片与副本分片中该文本版本是一致的；但如果主本分片意外地挂掉了，主副本复制流程没有完全完成，且主本分片所在节点在某个时刻又重新返回集群（此时它已经不再是主本分片了，因为在挂掉时，Elasticsearch会选择其中一个副本分片，将其提升为主本分片），那么此时该文档在主副本分片中的版本就是不一致的了。

从6.7版本开始，Elasticsearch使用seq_num和primary_term参数来实现乐观并发控制。在每个索引被创建时，seq_num初始值为0，当面向该索引针对文档进行index、update或delete操作时，seq_num会不断递增；primary_term用于区分新主本分片和旧主本分片，其初始值为1，同样是不断递增的，当主本分片所在节点挂掉时，Elasticsearch会在其众多副本分片中选择一个然后将其提升为主本分片，而primary_term递增的时机就在此刻，

如何获取`seq_num`和`primary_term`参数值

Get API

代码语言：javascript复制

GET /products/_doc/1
-----------------------
{
    "_index": "products",
    "_type": "_doc",
    "_id": "1",
    "_version": 1,
    "_seq_no": 0,
    "_primary_term": 1,
    "found": true,
    "_source": {
        "product": "r2d2",
        "details": "A resourceful astromech droid"
    }
}

Search API

代码语言：javascript复制

GET /products/_search
{
    "seq_no_primary_term": true,
    "query": {
        "match_all": {}
    }
}
-----------------------
{
    "took": 766,
    "timed_out": false,
    "_shards": {
        "total": 1,
        "successful": 1,
        "skipped": 0,
        "failed": 0
    },
    "hits": {
        "total": {
            "value": 1,
            "relation": "eq"
        },
        "max_score": 1.0,
        "hits": [
            {
                "_index": "products",
                "_type": "_doc",
                "_id": "1",
                "_seq_no": 1,
                "_primary_term": 1,
                "_score": 1.0,
                "_source": {
                    "product": "r2d2",
                    "details": "A resourceful astromech droid",
                    "tags": [
                        "droid"
                    ]
                }
            }
        ]
    }
}

使用if_seq_no和if_primary_term请求参数进行更新

代码语言：javascript复制

PUT /products/_doc/1?if_seq_no=0&if_primary_term=2
{
    "product": "r2d2",
    "details": "A resourceful astromech droid",
    "tags": [
        "droid"
    ]
}
-----------------------
{
    "error": {
        "root_cause": [
            {
                "type": "version_conflict_engine_exception",
                "reason": "[1]: version conflict, required seqNo [0], primary term [2]. current document has seqNo [0] and primary term [1]",
                "index_uuid": "jwzozHXXSYuw88SJ2SpxdQ",
                "shard": "0",
                "index": "products"
            }
        ],
        "type": "version_conflict_engine_exception",
        "reason": "[1]: version conflict, required seqNo [0], primary term [2]. current document has seqNo [0] and primary term [1]",
        "index_uuid": "jwzozHXXSYuw88SJ2SpxdQ",
        "shard": "0",
        "index": "products"
    },
    "status": 409
}

从上面响应来看，更新失败了，这是因为传入的if_primary_term值与当前primary_term值并不一致，只要将其调整为1就可以成功了。

代码语言：javascript复制

PUT /products/_doc/1?if_seq_no=0&if_primary_term=2
{
    "product": "r2d2",
    "details": "A resourceful astromech droid",
    "tags": [
        "droid"
    ]
}
-----------------------
{
    "_index": "products",
    "_type": "_doc",
    "_id": "1",
    "_version": 2,
    "result": "updated",
    "_shards": {
        "total": 2,
        "successful": 1,
        "failed": 0
    },
    "_seq_no": 1,
    "_primary_term": 1
}

参考文档

https://www.elastic.co/guide/en/elasticsearch/reference/current/optimistic-concurrency-control.html
https://www.elastic.co/cn/blog/elasticsearch-sequence-ids-6-0

https api ElasticsearchService 网络安全

0 人点赞

Optimistic Concurrency Control