ElasticSearch(三十)基于scoll+bulk+索引别名实现零停机重建索引
1.为什么要重建索引?
总结,一个type下的mapping中的filed不能被修改,所以如果需要修改,则需要重建索引
2.怎么zero time重建索引?
一个field的设置是不能被修改的,如果要修改一个Field,那么应该重新按照新的mapping,建立一个index,然后将数据批量查询出来,重新用bulk api写入index中
批量查询的时候,建议采用scroll api,并且采用多线程并发的方式来reindex数据,每次scoll就查询指定日期的一段数据,交给一个线程即可。
过程:
(1)一开始,依靠dynamic mapping,插入数据,但是不小心有些数据是2017-01-01这种日期格式的,所以title这种field被自动映射为了date类型,实际上它应该是string类型的
PUT /my_index/my_type/3 { "title": "2017-01-03" }
查看mapping:
{ "my_index": { "mappings": { "my_type": { "properties": { "title": { "type": "date" } } } } } }
(2)当后期向索引中加入string类型的title值的时候,就会报错
PUT /my_index/my_type/4 { "title": "my first article" }
{ "error": { "root_cause": [ { "type": "mapper_parsing_exception", "reason": "failed to parse [title]" } ], "type": "mapper_parsing_exception", "reason": "failed to parse [title]", "caused_by": { "type": "illegal_argument_exception", "reason": "Invalid format: \"my first article\"" } }, "status": 400 }
(3)如果此时想修改title的类型,是不可能的
PUT /my_index/_mapping/my_type { "properties": { "title": { "type": "text" } } } { "error": { "root_cause": [ { "type": "illegal_argument_exception", "reason": "mapper [title] of different type, current_type [date], merged_type [text]" } ], "type": "illegal_argument_exception", "reason": "mapper [title] of different type, current_type [date], merged_type [text]" }, "status": 400 }
(4)此时,唯一的办法,就是进行reindex,也就是说,重新建立一个索引,将旧索引的数据查询出来,再导入新索引
(5)如果说旧索引的名字,是old_index,新索引的名字是new_index,终端java应用,已经在使用old_index在操作了,难道还要去停止java应用,修改使用的index为new_index,才重新启动java应用吗?这个过程中,就会导致java应用停机,可用性降低。
(6)所以说,给java应用一个别名,这个别名是指向旧索引的,java应用先用着,java应用先用goods_index alias来操作,此时实际指向的是旧的my_index
PUT /my_index/_alias/goods_index
(7)新建一个index,调整其title的类型为string
PUT /my_index_new { "mappings": { "my_type": { "properties": { "title": { "type": "text" } } } } }
(8)使用scroll api将数据批量查询出来
GET /my_index/_search?scroll=1m { "query": { "match_all": {} }, "sort": ["_doc"], "size": 1 }
{ "_scroll_id": "DnF1ZXJ5VGhlbkZldGNoBQAAAAAAADpAFjRvbnNUWVZaVGpHdklqOV9zcFd6MncAAAAAAAA6QRY0b25zVFlWWlRqR3ZJajlfc3BXejJ3AAAAAAAAOkIWNG9uc1RZVlpUakd2SWo5X3NwV3oydwAAAAAAADpDFjRvbnNUWVZaVGpHdklqOV9zcFd6MncAAAAAAAA6RBY0b25zVFlWWlRqR3ZJajlfc3BXejJ3", "took": 1, "timed_out": false, "_shards": { "total": 5, "successful": 5, "failed": 0 }, "hits": { "total": 3, "max_score": null, "hits": [ { "_index": "my_index", "_type": "my_type", "_id": "2", "_score": null, "_source": { "title": "2017-01-02" }, "sort": [ 0 ] } ] } }
(9)采用bulk api将scoll查出来的一批数据,批量写入新索引
POST /_bulk { "index": { "_index": "my_index_new", "_type": "my_type", "_id": "2" }} { "title": "2017-01-02" }
(10)反复循环8~9,查询一批又一批的数据出来,采取bulk api将每一批数据批量写入新索引
(11)将goods_index alias切换到my_index_new上去,java应用会直接通过index别名使用新的索引中的数据,java应用程序不需要停机,零停机,高可用
POST /_aliases { "actions": [ { "remove": { "index": "my_index", "alias": "goods_index" }}, { "add": { "index": "my_index_new", "alias": "goods_index" }} ] }
(12)直接通过goods_index别名来查询,是否ok
GET /goods_index/my_type/_search
3、基于alias对client透明切换index
PUT /my_index_v1/_alias/my_index
client对my_index进行操作
reindex操作,完成之后,切换v1到v2
POST /_aliases { "actions": [ { "remove": { "index": "my_index_v1", "alias": "my_index" }}, { "add": { "index": "my_index_v2", "alias": "my_index" }} ] }