由正则表示式匹配($regex)引起的一次mongo数据库cpu占用率高的问题

某一天，监控到mongo数据库cpu使用率高了很多，查了一下，发现是下面这种语句引起的：

db.example_collection.find({ "idField" : { "$regex" : "123456789012345678"} , "dateField" : { "$regex" : "2019/10/10"}})

通常，遇到这种情况，我第一反应是缺少相关字段的索引，导致每执行一次这种语句都会全表扫描一次。

但是我用explain( )语句分析了下，发现上面所涉及的两个字段idField、dateField是有索引的，并且该语句也是有使用到索引的。如下为explain( )的结果：

mgset-11111111:PRIMARY> db.example_collection.find({ "idField" : { "$regex" : "123456789012345678"} , "dateField" : { "$regex" : "2019/10/10"}}).explain("queryPlanner") {         "queryPlanner" : {                 "plannerVersion" : 1,                 "namespace" : "example_db.example_collection",                 "indexFilterSet" : false,                 "parsedQuery" : {                         "$and" : [                                 {                                         "idField" : {                                                 "$regex" : "123456789012345678"                                         }                                 },                                 {                                         "dateField" : {                                                 "$regex" : "2019/10/10"                                         }                                 }                         ]                 },                 "winningPlan" : {                         "stage" : "FETCH",                         "inputStage" : {                                 "stage" : "IXSCAN",                                 "filter" : {                                         "$and" : [                                                 {                                                         "idField" : {                                                                 "$regex" : "123456789012345678"                                                         }                                                 },                                                 {                                                         "dateField" : {                                                                 "$regex" : "2019/10/10"                                                         }                                                 }                                         ]                                 },                                 "keyPattern" : {                                         "idField" : 1,                                         "dateField" : 1                                 },                                 "indexName" : "idField_1_dateField_1",                                 "isMultiKey" : false,                                 "multiKeyPaths" : {                                         "idField" : [ ],                                         "dateField" : [ ]                                 },                                 "isUnique" : false,                                 "isSparse" : false,                                 "isPartial" : false,                                 "indexVersion" : 2,                                 "direction" : "forward",                                 "indexBounds" : {                                         "idField" : [                                                 "[\"\", {})",                                                 "[/123456789012345678/, /123456789012345678/]"                                         ],                                         "dateField" : [                                                 "[\"\", {})",                                                 "[/2019/10/10/, /2019/10/10/]"                                         ]                                 }                         }                 },                 "rejectedPlans" : [ ]         },         "ok" : 1 }

查看mongo的日志发现，这种语句执行一次就要800~900ms，的确是比较慢。除非数据库cpu核数很多，要不然只要这种语句每秒并发稍微高一点，cpu很快就被占满了。

之后搜索了下，发现有可能是正则表达式的问题。原来，虽然该语句的确是使用了索引，但是explain( )语句的输出中还有一个字段"indexBounds"，表示执行该语句时所需扫描的索引范围。说实话，上面那个输出中，我始终没看明白它那个索引范围。上面的语句对idField、dateField这两个字段都进行了普通的正则表达式匹配，我猜测它应该是扫描了整个索引树，所以导致索引并未实际提升该语句的查询效率。

我看了下数据库里面的数据，发现idField、dateField这两个字段完全没有必要进行正则匹配，进行普通的文本匹配就行。将正则匹配操作$regex去掉之后，再分析一下，结果是这样的：

mgset-11111111:PRIMARY> db.example_collection.find({ "idField" : "123456789012345678", "dateField" : "2019/10/10"}).explain("queryPlanner") {         "queryPlanner" : {                 "plannerVersion" : 1,                 "namespace" : "example_db.example_collection",                 "indexFilterSet" : false,                 "parsedQuery" : {                         "$and" : [                                 {                                         "idField" : {                                                 "$eq" : "123456789012345678"                                         }                                 },                                 {                                         "dateField" : {                                                 "$eq" : "2019/10/10"                                         }                                 }                         ]                 },                 "winningPlan" : {                         "stage" : "FETCH",                         "inputStage" : {                                 "stage" : "IXSCAN",                                 "keyPattern" : {                                         "idField" : 1,                                         "dateField" : 1                                 },                                 "indexName" : "idField_1_dateField_1",                                 "isMultiKey" : false,                                 "multiKeyPaths" : {                                         "idField" : [ ],                                         "dateField" : [ ]                                 },                                 "isUnique" : false,                                 "isSparse" : false,                                 "isPartial" : false,                                 "indexVersion" : 2,                                 "direction" : "forward",                                 "indexBounds" : {                                         "idField" : [                                                 "[\"123456789012345678\", \"123456789012345678\"]"                                         ],                                         "dateField" : [                                                 "[\"2019/10/10\", \"2019/10/10\"]"                                         ]                                 }                         }                 },                 "rejectedPlans" : [ ]         },         "ok" : 1 }

可以看到，仍然使用到了索引，并且索引扫描范围是仅限于一个值的。

后来跟开发人员确认了下，该语句确实没必要使用正则匹配，就让他把正则匹配去掉了。之后就没有再出现问题了，mongo慢日志中也未再出现该语句。