MongoDB 索引的最佳实践

前言

大部分开发者都知道加索引会快。但实际过程中，我们常碰到一些疑问&困难：

• 我们查询的字段会各种 case 都有，是不是各个涉及查询的字段都要加索引？
• 复合索引和单字段怎么选择，都加还是每一个的单个字段就好了？
• 加索引有没有副作用？
• 索引都加了，但还是不够快？怎么办？

本文尝试去解释索引的基本知识&解答上述的疑问。

一、索引究竟是什么东西？

大部分开发者接触索引，大概知道索引类似书的目录，你要找到想要的内容，通过目录找到限定的关键字，进而找到对应的章节的 pageno，再找到具体的内容。
在数据结构里面，最简单的索引实现类似 hashmap，通过关键字 key，映射到具体的位置，找到具体的内容。但除了 hash 的方式，还有多种的方式实现索引。

（一）索引的多种实现方式以及特色

hash / b-tree / b+-tree redis HSET / MongoDB&PostgreSQL / MySQL

hashmap

一图见 b-tree & b+-tree 差别：

• b+-tree 叶子存数据，非叶子存索引，不存数据，叶子间有 link
• b-tree 非叶子可存数据

算法查找复杂度上来说：

• hash 接近 O(1)
• b-tree O(1)~ O(Log(n))更快的平均查找时间，不稳定的查询时间
• b+ tree O(Log(n)) 连续数据， 查询的稳定性

至于为何 MongoDB 的实现选择 b-tree 而非 b+-tree ？
网上多篇文章有阐述，非本文重点。

（二）数据&索引的存储

index 尽量存储在内存，data 其次。
注意只保留必要的 index，内存尽量用在刀刃上。
如果 index memory 都接近占满 memory，那么就很容易读到 disk，速度就下来了。

（三）知道索引的实现&存储原理后的思考

insert/update/delete 会触发 rebalance tree，所以，增删改数据，索引会触发修改，性能会有损耗，索引不是越多越好。既然如此，选哪些字段作为索引呢？当查询用到这些条件，怎么办？
拿一个最简单的 hashmap 来讲，为什么复杂度不是 O(1)，而是所谓接近 O(1)。因为有 key 冲突 /重复，DB 去找的时候，key 冲突的数据一大堆的话，还是得轮着继续找。b-tree 看键(key)的选择也是如此。
因此一个大部分开发经常犯的错就是对没有区分度的 key 建索引。例如：很多就只有集中类别的 type/status 的 documents count 达几十万以上的 collection，通常这种索引没什么帮助。

二、复合索引

（一）复合索引不是越多越好

如果不想多建多余的索引，开发的同事在复合 & 单个字段选择上有时候挺纠结的。 根据典型碰到的场景，来做几个实验：
这里创建了个 loans collection。简化只有 100 条数据。这个是借贷的表有 _id, userId, status(借贷状态), amount(金额).

db.loans.count()100

db.loans.find({ "userId" : "59e022d33f239800129c61c7", "status" : "repayed", }).explain() { "queryPlanner" : { "plannerVersion" : 1, "namespace" : "cashLoan.loans", "indexFilterSet" : false, "parsedQuery" : { "$and" : [ { "status" : { "$eq" : "repayed" } }, { "userId" : { "$eq" : "59e022d33f239800129c61c7" } } ] }, "queryHash" : "15D5A9A1", "planCacheKey" : "15D5A9A1", "winningPlan" : { "stage" : "COLLSCAN", "filter" : { "$and" : [ { "status" : { "$eq" : "repayed" } }, { "userId" : { "$eq" : "59e022d33f239800129c61c7" } } ] }, "direction" : "forward" }, "rejectedPlans" : [ ] }, "serverInfo" : { "host" : "RMBAP", "port" : 27017, "version" : "4.1.11", "gitVersion" : "1b8a9f5dc5c3314042b55e7415a2a25045b32a94" }, "ok" : 1 } 

注意上面 COLLSCAN 全表扫描了,因为没有索引。接下来我们分别建立几个索引。
step 1 先建立 {userId:1, status:1}

db.loans.createIndex({userId:1, status:1}) { "createdCollectionAutomatically" : false, "numIndexesBefore" : 1, "numIndexesAfter" : 2, "ok" : 1 } 

db.loans.find({ "userId" : "59e022d33f239800129c61c7", "status" : "repayed", }).explain() { "queryPlanner" : { "plannerVersion" : 1, "namespace" : "cashLoan.loans", "indexFilterSet" : false, "parsedQuery" : { "$and" : [ { "status" : { "$eq" : "repayed" } }, { "userId" : { "$eq" : "59e022d33f239800129c61c7" } } ] }, "queryHash" : "15D5A9A1", "planCacheKey" : "BB87F2BA", "winningPlan" : { "stage" : "FETCH", "inputStage" : { "stage" : "IXSCAN", "keyPattern" : { "userId" : 1, "status" : 1 }, "indexName" : "userId_1_status_1", "isMultiKey" : false, "multiKeyPaths" : { "userId" : [ ], "status" : [ ] }, "isUnique" : false, "isSparse" : false, "isPartial" : false, "indexVersion" : 2, "direction" : "forward", "indexBounds" : { "userId" : [ "["59e022d33f239800129c61c7", "59e022d33f239800129c61c7"]" ], "status" : [ "["repayed", "repayed"]" ] } } }, "rejectedPlans" : [ ] }, "serverInfo" : { "host" : "RMBAP", "port" : 27017, "version" : "4.1.11", "gitVersion" : "1b8a9f5dc5c3314042b55e7415a2a25045b32a94" }, "ok" : 1 } 

结果：如愿命中 {userId:1, status:1} 作为 winning plan。

step2：再建立个典型的索引 userId

db.loans.createIndex({userId:1}) { "createdCollectionAutomatically" : false, "numIndexesBefore" : 2, "numIndexesAfter" : 3, "ok" : 1 } 

db.loans.find({ "userId" : "59e022d33f239800129c61c7", "status" : "repayed", }).explain() { "queryPlanner" : { "plannerVersion" : 1, "namespace" : "cashLoan.loans", "indexFilterSet" : false, "parsedQuery" : { "$and" : [ { "status" : { "$eq" : "repayed" } }, { "userId" : { "$eq" : "59e022d33f239800129c61c7" } } ] }, "queryHash" : "15D5A9A1", "planCacheKey" : "1B1A4861", "winningPlan" : { "stage" : "FETCH", "inputStage" : { "stage" : "IXSCAN", "keyPattern" : { "userId" : 1, "status" : 1 }, "indexName" : "userId_1_status_1", "isMultiKey" : false, "multiKeyPaths" : { "userId" : [ ], "status" : [ ] }, "isUnique" : false, "isSparse" : false, "isPartial" : false, "indexVersion" : 2, "direction" : "forward", "indexBounds" : { "userId" : [ "[\"59e022d33f239800129c61c7\", \"59e022d33f239800129c61c7\"]" ], "status" : [ "[\"repayed\", \"repayed\"]" ] } } }, "rejectedPlans" : [ { "stage" : "FETCH", "filter" : { "status" : { "$eq" : "repayed" } }, "inputStage" : { "stage" : "IXSCAN", "keyPattern" : { "userId" : 1 }, "indexName" : "userId_1", "isMultiKey" : false, "multiKeyPaths" : { "userId" : [ ] }, "isUnique" : false, "isSparse" : false, "isPartial" : false, "indexVersion" : 2, "direction" : "forward", "indexBounds" : { "userId" : [ "["59e022d33f239800129c61c7", "59e022d33f239800129c61c7"]" ] } } } ] }, "serverInfo" : { "host" : "RMBAP", "port" : 27017, "version" : "4.1.11", "gitVersion" : "1b8a9f5dc5c3314042b55e7415a2a25045b32a94" }, "ok" : 1 } 

留意到 DB 检测到 {userId:1, status:1} 为更优执行的方案.

db.loans.find({ "userId" : "59e022d33f239800129c61c7" }).explain() { "queryPlanner" : { "plannerVersion" : 1, "namespace" : "cashLoan.loans", "indexFilterSet" : false, "parsedQuery" : { "userId" : { "$eq" : "59e022d33f239800129c61c7" } }, "queryHash" : "B1777DBA", "planCacheKey" : "1F09D68E", "winningPlan" : { "stage" : "FETCH", "inputStage" : { "stage" : "IXSCAN", "keyPattern" : { "userId" : 1 }, "indexName" : "userId_1", "isMultiKey" : false, "multiKeyPaths" : { "userId" : [ ] }, "isUnique" : false, "isSparse" : false, "isPartial" : false, "indexVersion" : 2, "direction" : "forward", "indexBounds" : { "userId" : [ "["59e022d33f239800129c61c7", "59e022d33f239800129c61c7"]" ] } } }, "rejectedPlns" : [ { "stage" : "FETCH", "inputStage" : { "stage" : "IXSCAN", "keyPattern" : { "userId" : 1, "status" : 1 }, "indexName" : "userId_1_status_1", "isMultiKey" : false, "multiKeyPaths" : { "userId" : [ ], "status" : [ ] }, "isUnique" : false, "isSparse" : false, "isPartial" : false, "indexVersion" : 2, "direction" : "forward", "indexBounds" : { "userId" : [ "["59e022d33f239800129c61c7", "59e022d33f239800129c61c7"]" ], "status" : [ "[MinKey, MaxKey]" ] } } } ] }, "serverInfo" : { "host" : "RMBAP", "port" : 27017, "version" : "4.1.11", "gitVersion" : "1b8a9f5dc5c3314042b55e7415a2a25045b32a94" }, "ok" : 1 } 

留意到 DB 检测到 {userId:1} 为更优执行的方案，嗯~，如我们所料.

db.loans.find({ "status" : "repayed" }).explain() { "queryPlanner" : { "plannerVersion" : 1, "namespace" : "cashLoan.loans", "indexFilterSet" : false, "parsedQuery" : { "status" : { "$eq" : "repayed" } }, "queryHash" : "E6304EB6", "planCacheKey" : "7A94191B", "winningPlan" : { "stage" : "COLLSCAN", "filter" : { "status" : { "$eq" : "repayed" } }, "direction" : "forward" }, "rejectedPlans" : [ ] }, "serverInfo" : { "host" : "RMBAP", "port" : 27017, "version" : "4.1.11", "gitVersion" : "1b8a9f5dc5c3314042b55e7415a2a25045b32a94" }, "ok" : 1 } 

有趣的部分：status 不命中索引，全表扫描
接下来的步骤，加个 sort :

 db.loans.find({ "userId" : "59e022d33f239800129c61c7" }).sort({status:1}).explain() { "queryPlanner" : { "plannerVersion" : 1, "namespace" : "cashLoan.loans", "indexFilterSet" : false, "parsedQuery" : { "userId" : { "$eq" : "59e022d33f239800129c61c7" } }, "queryHash" : "F5ABB1AA", "planCacheKey" : "764CBAA8", "winningPlan" : { "stage" : "FETCH", "inputStage" : { "stage" : "IXSCAN", "keyPattern" : { "userId" : 1, "status" : 1 }, "indexName" : "userId_1_status_1", "isMultiKey" : false, "multiKeyPaths" : { "userId" : [ ], "status" : [ ] }, "isUnique" : false, "isSparse" : false, "isPartial" : false, "indexVersion" : 2, "direction" : "forward", "indexBounds" : { "userId" : [ "["59e022d33f239800129c61c7", "59e022d33f239800129c61c7"]" ], "status" : [ "[MinKey, MaxKey]" ] } } }, "rejectedPlans" : [ { "stage" : "SORT", "sortPattern" : { "status" : 1 }, "inputStage" : { "stage" : "SORT_KEY_GENERATOR", "inputStage" : { "stage" : "FETCH", "inputStage" : { "stage" : "IXSCAN", "keyPattern" : { "userId" : 1 }, "indexName" : "userId_1", "isMultiKey" : false, "multiKeyPaths" : { "userId" : [ ] }, "isUnique" : false, "isSparse" : false, "isPartial" : false, "indexVersion" : 2, "direction" : "forward", "indexBounds" : { "userId" : [ "["59e022d33f239800129c61c7", "59e022d33f239800129c61c7"]" ] } } } } } ] }, "serverInfo" : { "host" : "RMBAP", "port" : 27017, "version" : "4.1.11", "gitVersion" : "1b8a9f5dc5c3314042b55e7415a2a25045b32a94" }, "ok" : 1 } 

（二）其他尝试

有趣的部分：status 不命中索引

db.loans.find({ "status" : "repayed","userId" : "59e022d33f239800129c61c7", }).explain() { "queryPlanner" : { "plannerVersion" : 1, "namespace" : "cashLoan.loans", "indexFilterSet" : false, "parsedQuery" : { "$and" : [ { "status" : { "$eq" : "repayed" } }, { "userId" : { "$eq" : "59e022d33f239800129c61c7" } } ] }, "queryHash" : "15D5A9A1", "planCacheKey" : "1B1A4861", "winningPlan" : { "stage" : "FETCH", "inputStage" : { "stage" : "IXSCAN", "keyPattern" : { "userId" : 1, "status" : 1 }, "indexName" : "userId_1_status_1", "isMultiKey" : false, "multiKeyPaths" : { "userId" : [ ], "status" : [ ] }, "isUnique" : false, "isSparse" : false, "isPartial" : false, "indexVersion" : 2, "direction" : "forward", "indexBounds" : { "userId" : [ "[\"59e022d33f239800129c61c7\", \"59e022d33f239800129c61c7\"]" ], "status" : [ "[\"repayed\", \"repayed\"]" ] } } }, "rejectedPlans" : [ { "stage" : "FETCH", "filter" : { "status" : { "$eq" : "repayed" } }, "inputStage" : { "stage" : "IXSCAN", "keyPattern" : { "userId" : 1 }, "indexName" : "userId_1", "isMultiKey" : false, "multiKeyPaths" : { "userId" : [ ] }, "isUnique" : false, "isSparse" : false, "isPartial" : false, "indexVersion" : 2, "direction" : "forward", "indexBounds" : { "userId" : [ "["59e022d33f239800129c61c7", "59e022d33f239800129c61c7"]" ] } } } ] }, "serverInfo" : { "host" : "RMBAP", "port" : 27017, "version" : "4.1.11", "gitVersion" : "1b8a9f5dc5c3314042b55e7415a2a25045b32a94" }, "ok" : 1 } 

命中索引，跟 query 的各个字段顺序不相关，如我们猜测。
有趣部分再来， 我们删掉索引{userId:1}

db.loans.dropIndex({"userId":1}) { "nIndexesWas" : 3, "ok" : 1 } db.loans.find({"userId" : "59e022d33f239800129c61c7", }).explain() { "queryPlanner" : { "plannerVersion" : 1, "namespace" : "cashLoan.loans", "indexFilterSet" : false, "parsedQuery" : { "userId" : { "$eq" : "59e022d33f239800129c61c7" } }, "queryHash" : "B1777DBA", "planCacheKey" : "5776AB9C", "winningPlan" : { "stage" : "FETCH", "inputStage" : { "stage" : "IXSCAN", "keyPattern" : { "userId" : 1, "status" : 1 }, "indexName" : "userId_1_status_1", "isMultiKey" : false, "multiKeyPaths" : { "userId" : [ ], "status" : [ ] }, "isUnique" : false, "isSparse" : false, "isPartial" : false, "indexVersion" : 2, "direction" : "forward", "indexBounds" : { "userId" : [ "["59e022d33f239800129c61c7", "59e022d33f239800129c61c7"]" ], "status" : [ "[MinKey, MaxKey]" ] } } }, "rejectedPlans" : [ ] }, "serverInfo" : { "host" : "RMBAP", "port" : 27017, "version" : "4.1.11", "gitVersion" : "1b8a9f5dc5c3314042b55e7415a2a25045b32a94" }, "ok" : 1 } 

DB 执行分析器觉得索引{userId:1, status:1} 能更优,没有命中复合索引，这个是因为 status 不是 leading field。

db.loans.find({ "status" : "repayed" }).explain() { "queryPlanner" : { "plannerVersion" : 1, "namespace" : "cashLoan.loans", "indexFilterSet" : false, "parsedQuery" : { "status" : { "$eq" : "repayed" } }, "queryHash" : "E6304EB6", "planCacheKey" : "7A94191B", "winningPlan" : { "stage" : "COLLSCAN", "filter" : { "status" : { "$eq" : "repayed" } }, "direction" : "forward" }, "rejectedPlans" : [ ] }, "serverInfo" : { "host" : "RMBAP", "port" : 27017, "version" : "4.1.11", "gitVersion" : "1b8a9f5dc5c3314042b55e7415a2a25045b32a94" }, "ok" : 1 } 

再换个角度 sort 一遍， 与前面 query & sort 互换，之前是：

db.loans.find({userId:1}).sort({ "status" : "repayed" }) 

看看有啥不一样？

db.loans.find({ "status" : "repayed" }).sort({userId:1}).explain() { "queryPlanner" : { "plannerVersion" : 1, "namespace" : "cashLoan.loans", "indexFilterSet" : false, "parsedQuery" : { "status" : { "$eq" : "repayed" } }, "queryHash" : "56EA6313", "planCacheKey" : "2CFCDA7F", "winningPlan" : { "stage" : "FETCH", "filter" : { "status" : { "$eq" : "repayed" } }, "inputStage" : { "stage" : "IXSCAN", "keyPattern" : { "userId" : 1, "status" : 1 }, "indexName" : "userId_1_status_1", "isMultiKey" : false, "multiKeyPaths" : { "userId" : [ ], "status" : [ ] }, "isUnique" : false, "isSparse" : false, "isPartial" : false, "indexVersion" : 2, "direction" : "forward", "indexBounds" : { "userId" : [ "[MinKey, MaxKey]" ], "status" : [ "[MinKey, MaxKey]" ] } } }, "rejectedPlans" : [ ] }, "serverInfo" : { "host" : "RMBAP", "port" : 27017, "version" : "4.1.11", "gitVersion" : "1b8a9f5dc5c3314042b55e7415a2a25045b32a94" }, "ok" : 1 } 

如猜测，命中索引。
再来玩玩，确认下 leading filed 试验：

db.loans.dropIndex("userId_1_status_1") { "nIndexesWas" : 2, "ok" : 1 } 

db.loans.getIndexes() [ { "v" : 2, "key" : { "id" : 1 }, "name" : "id_", "ns" : "cashLoan.loans" } ] 

db.loans.createIndex({status:1, userId:1}) { "createdCollectionAutomatically" : false, "numIndexesBefore" : 1, "numIndexesAfter" : 2, "ok" : 1 } 

db.loans.getIndexes() [ { "v" : 2, "key" : { "id" : 1 }, "name" : "id_", "ns" : "cashLoan.loans" }, { "v" : 2, "key" : { "status" : 1, "userId" : 1 }, "name" : "status_1_userId_1", "ns" : "cashLoan.loans" } ] 

db.loans.find({ "status" : "repayed" }).explain() { "queryPlanner" : { "plannerVersion" : 1, "namespace" : "cashLoan.loans", "indexFilterSet" : false, "parsedQuery" : { "status" : { "$eq" : "repayed" } }, "queryHash" : "E6304EB6", "planCacheKey" : "7A94191B", "winningPlan" : { "stage" : "FETCH", "inputStage" : { "stage" : "IXSCAN", "keyPattern" : { "status" : 1, "userId" : 1 }, "indexName" : "status_1_userId_1", "isMultiKey" : false, "multiKeyPaths" : { "status" : [ ], "userId" : [ ] }, "isUnique" : false, "isSparse" : false, "isPartial" : false, "indexVersion" : 2, "direction" : "forward", "indexBounds" : { "status" : [ "["repayed", "repayed"]" ], "userId" : [ "[MinKey, MaxKey]" ] } } }, "rejectedPlans" : [ ] }, "serverInfo" : { "host" : "RMBAP", "port" : 27017, "version" : "4.1.11", "gitVersion" : "1b8a9f5dc5c3314042b55e7415a2a25045b32a94" }, "ok" : 1 } 

db.loans.getIndexes() [ { "v" : 2, "key" : { "id" : 1 }, "name" : "id_", "ns" : "cashLoan.loans" }, { "v" : 2, "key" : { "status" : 1, "userId" : 1 }, "name" : "status_1_userId_1", "ns" : "cashLoan.loans" } ] 

db.loans.find({"userId" : "59e022d33f239800129c61c7", }).explain() { "queryPlanner" : { "plannerVersion" : 1, "namespace" : "cashLoan.loans", "indexFilterSet" : false, "parsedQuery" : { "userId" : { "$eq" : "59e022d33f239800129c61c7" } }, "queryHash" : "B1777DBA", "planCacheKey" : "5776AB9C", "winningPlan" : { "stage" : "COLLSCAN", "filter" : { "userId" : { "$eq" : "59e022d33f239800129c61c7" } }, "direction" : "forward" }, "rejectedPlans" : [ ] }, "serverInfo" : { "host" : "RMBAP", "port" : 27017, "version" : "4.1.11", "gitVersion" : "1b8a9f5dc5c3314042b55e7415a2a25045b32a94" }, "ok" : 1 } 

看完这个试验，明白了 {userId:1, status:1} vs {status:1,userId:1} 的差别了吗？

PS：这个 case 里面其实 status 区分度不高，这里只是作为实例展示。

三、总结：

• 注意使用上、使用频率上、区分高的 /常用的在前面；
• 如果需要减少索引以节省 memory/提高修改数据的性能的话，可以保留区分度高，常用的，去除区分度不高，不常用的索引。
• 学会用 explain （）验证分析性能：

DB 一般都有执行器优化的分析，MySQL & MongoDB 都是 用 explain 来做分析。
语法上 MySQL :

explain your_sql

MongoDB:

yoursql.explain()

总结典型：理想的查询是结合 explain 的指标，他们通常是多个的混合:

• IXSCAN : 索引命中；
• Limit : 带 limit ；
• Projection : 相当于非 select * ；
• Docs Size less is better ；
• Docs Examined less is better ；
• nReturned=totalDocsExamined=totalKeysExamined ；
• SORT in index：sort 也是命中索引，否则，需要拿到数据后，再执行一遍排序；
• Limit Array elements： 限定数组返回的条数，数组也不应该太多数据，否则 schema 设计不合理。

彩蛋

文末，还有最开头 1 个问题没回答：如果我的索引改加的都加了，还不够快,怎么办？
留个悬念，之后再写一篇。