要保证数据库处于高效、稳定的状态,除了良好的硬件基础、高效高可用的数据库架构、贴合业务的数据模型之外,高效的查询语句也是不可少的。那么,如何查看并判断我们的执行计划呢?我们今天就来谈论下MongoDB的执行计划分析。
引子MongoDB 3.0之后,explain的返回与使用方法与之前版本有了不少变化,介于3.0之后的优秀特色,本文仅针对MongoDB 3.0+的explain进行讨论。现版本explain有三种模式,分别如下:queryPlannerexecutionStatsallPlansExecution由于文章字数原因,本系列将分为三个部分。本文是第一部分,主要针对queryPlanner,与executionStats进行分析。正文queryPlannerqueryPlanner是现版本explain的默认模式,queryPlanner模式下并不会去真正进行query语句查询,而是针对query语句进行执行计划分析并选出winning plan。<code>{ "queryPlanner" : { "plannerVersion" : 1,"namespace" : "game_db.game_user","indexFilterSet" : false,"parsedQuery" : { "w" : { "$eq" : 1}},"winningPlan" : { "stage" : "FETCH","inputStage" : { "stage" : "IXSCAN","keyPattern" : { "w" : 1,"n" : 1},"indexName" : "w_1_n_1","isMultiKey" : false,"direction" : "forward","indexBounds" : { "w" : ["[1.0, 1.0]"],"n" : ["[MinKey, MaxKey]"]}}},"rejectedPlans" : [{ "stage" : "FETCH","inputStage" : { "stage" : "IXSCAN","keyPattern" : { "w" : 1,"v" : 1},"indexName" : "w_1_v_1","isMultiKey" : false,"direction" : "forward","indexBounds" : { "w" : ["[1.0, 1.0]"],"v" : ["[MinKey, MaxKey]"]}}}]},</code>先来看queryPlanner模式的各个返回意义。explain.queryPlannerqueryPlanner的返回。explain.queryPlanner.namespace顾名思义,该值返回的是该query所查询的表。explain.queryPlanner.indexFilterSet针对该query是否有indexfilter(会在后文进行详细解释)。explain.queryPlanner.winningPlan查询优化器针对该query所返回的最优执行计划的详细内容。explain.queryPlanner.winningPlan.stage最优执行计划的stage,这里返回是FETCH,可以理解为通过返回的index位置去检索具体的文档(stage有数个模式,将在后文中进行详解)。explain.queryPlanner.winningPlan.inputStageexplain.queryPlanner.winningPlan.stage的child stage,此处是IXSCAN,表示进行的是index scanning。explain.queryPlanner.winningPlan.keyPattern所扫描的index内容,此处是w:1与n:1。explain.queryPlanner.winningPlan.indexNamewinning plan所选用的index。explain.queryPlanner.winningPlan.isMultiKey是否是Multikey,此处返回是false,如果索引建立在array上,此处将是true。explain.queryPlanner.winningPlan.direction此query的查询顺序,此处是forward,如果用了.sort({w:-1})将显示backward。explain.queryPlanner.winningPlan.indexBoundswinningplan所扫描的索引范围,此处查询条件是w:1,使用的index是w与n的联合索引,故w是[1.0,1.0]而n没有指定在查询条件中,故是[MinKey,MaxKey]。explain.queryPlanner.rejectedPlans其他执行计划(非最优而被查询优化器reject的)的详细返回,其中具体信息与winningPlan的返回中意义相同,故不在此赘述。executionStatsexecutionStats的返回中多了如下:<code> "executionStats" : { "executionSuccess" : true,"nReturned" : 29861,"executionTimeMillis" : 23079,"totalKeysExamined" : 29861,"totalDocsExamined" : 29861,"executionStages" : { "stage" : "FETCH","nReturned" : 29861,"executionTimeMillisEstimate" : 22685,"works" : 29862,"advanced" : 29861,"needTime" : 0,"needFetch" : 0,"saveState" : 946,"restoreState" : 946,"isEOF" : 1,"invalidates" : 0,"docsExamined" : 29861,"alreadyHasObj" : 0,"inputStage" : { "stage" : "IXSCAN","nReturned" : 29861,"executionTimeMillisEstimate" : 70,"works" : 29862,"advanced" : 29861,"needTime" : 0,"needFetch" : 0,"saveState" : 946,"restoreState" : 946,"isEOF" : 1,"invalidates" : 0,"keyPattern" : { "w" : 1,"n" : 1},"indexName" : "w_1_n_1","isMultiKey" : false,"direction" : "forward","indexBounds" : { "w" : ["[1.0, 1.0]"],"n" : ["[MinKey, MaxKey]"]},"keysExamined" : 29861,"dupsTested" : 0,"dupsDropped" : 0,"seenInvalidated" : 0,"matchTested" : 0}}},</code>executionStats模式中,我们主要需要注意的返回有如下几个executionStats.executionSuccess是否执行成功executionStats.nReturned查询的返回条数executionStats.executionTimeMillis整体执行时间executionStats.totalKeysExamined索引扫描次数executionStats.totalDocsExamineddocument扫描次数以上几个非常好理解,我们就不在这里详述,后文的案例中会有分析。executionStats.executionStages.stage这里是FETCH去扫描对于documentsexecutionStats.executionStages.nReturned由于是FETCH,所以这里该值与executionStats.nReturned一致executionStats.executionStages.docsExamined与executionStats.totalDocsExamined一致executionStats.inputStage中的与上述理解方式相同还有一些文档中没有描述的返回如:“works” : 29862,“advanced” : 29861,“isEOF” : 1,这些值都会在explan之初初始化:mongo/src/mongo/db/exec/plan_stats.h<code>struct CommonStats { CommonStats(const char* type): stageTypeStr(type),works(0),yields(0),unyields(0),invalidates(0),advanced(0),needTime(0),needYield(0),executionTimeMillis(0),isEOF(false) {}</code>以works为例,查看源码中发现,每次操作会加1,且会把执行时间记录在executionTimeMillis中。mongo/src/mongo/db/exec/fetch.cpp<code> ++_commonStats.works;// Adds the amount of time taken by work() to executionTimeMillis.ScopedTimer timer(&_commonStats.executionTimeMillis);</code>而在查询结束EOF,works又会加1,advanced不加。mongo/src/mongo/db/exec/eof.cpp<code>PlanStage::StageState EOFStage::work(WorkingSetID* out) { ++_commonStats.works;// Adds the amount of time taken by work() to executionTimeMillis.ScopedTimer timer(&_commonStats.executionTimeMillis);return PlanStage::IS_EOF;}</code>故正常的返回works会比nReturned多1,这时候isEOF为true(1):mongo/src/mongo/db/exec/eof.cpp<code>bool EOFStage::isEOF() { return true;}unique_ptr<PlanStageStats> EOFStage::getStats() { _commonStats.isEOF = isEOF();return make_unique<PlanStageStats>(_commonStats, STAGE_EOF);}</code>advanced的返回值在命中的时候+1,在skip,eof的时候不会增加如:mongo/src/mongo/db/exec/skip.cpp<code>if (PlanStage::ADVANCED == status) { // If we're still skipping results...if (_toSkip > 0) { // ...drop the result.--_toSkip;_ws->free(id);++_commonStats.needTime;return PlanStage::NEED_TIME;}*out = id;++_commonStats.advanced;return PlanStage::ADVANCED;</code>