问题
I have a collection faults in my MongoDB database which every document has these fields: rack_name, timestamp
Just for sake of testing and comparing performances, I have created these two indexes:
rack -> {'rack_name': 1}
and
time -> {'timestamp': 1}
Now I executed the following query with explain():
db.faults.find({
'rack_name': {
$in: [ 'providence1', 'helena2' ]
},
'timestamp': {
$gt: 1501548359000
}
})
.explain('allPlansExecution')
and here is the result:
{
"queryPlanner" : {
"plannerVersion" : 1,
"namespace" : "quicktester_clone.faults",
"indexFilterSet" : false,
"parsedQuery" : {
"$and" : [
{
"timestamp" : {
"$gt" : 1501548359000.0
}
},
{
"rack_name" : {
"$in" : [
"helena2",
"providence1"
]
}
}
]
},
"winningPlan" : {
"stage" : "FETCH",
"filter" : {
"timestamp" : {
"$gt" : 1501548359000.0
}
},
"inputStage" : {
"stage" : "IXSCAN",
"keyPattern" : {
"rack_name" : 1
},
"indexName" : "rack",
"isMultiKey" : false,
"multiKeyPaths" : {
"rack_name" : []
},
"isUnique" : false,
"isSparse" : false,
"isPartial" : false,
"indexVersion" : 2,
"direction" : "forward",
"indexBounds" : {
"rack_name" : [
"[\"helena2\", \"helena2\"]",
"[\"providence1\", \"providence1\"]"
]
}
}
},
"rejectedPlans" : [
{
"stage" : "FETCH",
"filter" : {
"rack_name" : {
"$in" : [
"helena2",
"providence1"
]
}
},
"inputStage" : {
"stage" : "IXSCAN",
"keyPattern" : {
"timestamp" : 1
},
"indexName" : "time",
"isMultiKey" : false,
"multiKeyPaths" : {
"timestamp" : []
},
"isUnique" : false,
"isSparse" : false,
"isPartial" : false,
"indexVersion" : 2,
"direction" : "forward",
"indexBounds" : {
"timestamp" : [
"(1501548359000.0, inf.0]"
]
}
}
}
]
},
"executionStats" : {
"executionSuccess" : true,
"nReturned" : 43,
"executionTimeMillis" : 1512,
"totalKeysExamined" : 221,
"totalDocsExamined" : 219,
"executionStages" : {
"stage" : "FETCH",
"filter" : {
"timestamp" : {
"$gt" : 1501548359000.0
}
},
"nReturned" : 43,
"executionTimeMillisEstimate" : 1431,
"works" : 222,
"advanced" : 43,
"needTime" : 177,
"needYield" : 0,
"saveState" : 64,
"restoreState" : 64,
"isEOF" : 1,
"invalidates" : 0,
"docsExamined" : 219,
"alreadyHasObj" : 0,
"inputStage" : {
"stage" : "IXSCAN",
"nReturned" : 219,
"executionTimeMillisEstimate" : 71,
"works" : 221,
"advanced" : 219,
"needTime" : 1,
"needYield" : 0,
"saveState" : 64,
"restoreState" : 64,
"isEOF" : 1,
"invalidates" : 0,
"keyPattern" : {
"rack_name" : 1
},
"indexName" : "rack",
"isMultiKey" : false,
"multiKeyPaths" : {
"rack_name" : []
},
"isUnique" : false,
"isSparse" : false,
"isPartial" : false,
"indexVersion" : 2,
"direction" : "forward",
"indexBounds" : {
"rack_name" : [
"[\"helena2\", \"helena2\"]",
"[\"providence1\", \"providence1\"]"
]
},
"keysExamined" : 221,
"seeks" : 2,
"dupsTested" : 0,
"dupsDropped" : 0,
"seenInvalidated" : 0
}
},
"allPlansExecution" : [
{
"nReturned" : 2,
"executionTimeMillisEstimate" : 31,
"totalKeysExamined" : 221,
"totalDocsExamined" : 221,
"executionStages" : {
"stage" : "FETCH",
"filter" : {
"rack_name" : {
"$in" : [
"helena2",
"providence1"
]
}
},
"nReturned" : 2,
"executionTimeMillisEstimate" : 31,
"works" : 221,
"advanced" : 2,
"needTime" : 219,
"needYield" : 0,
"saveState" : 64,
"restoreState" : 64,
"isEOF" : 0,
"invalidates" : 0,
"docsExamined" : 221,
"alreadyHasObj" : 0,
"inputStage" : {
"stage" : "IXSCAN",
"nReturned" : 221,
"executionTimeMillisEstimate" : 10,
"works" : 221,
"advanced" : 221,
"needTime" : 0,
"needYield" : 0,
"saveState" : 64,
"restoreState" : 64,
"isEOF" : 0,
"invalidates" : 0,
"keyPattern" : {
"timestamp" : 1
},
"indexName" : "time",
"isMultiKey" : false,
"multiKeyPaths" : {
"timestamp" : []
},
"isUnique" : false,
"isSparse" : false,
"isPartial" : false,
"indexVersion" : 2,
"direction" : "forward",
"indexBounds" : {
"timestamp" : [
"(1501548359000.0, inf.0]"
]
},
"keysExamined" : 221,
"seeks" : 1,
"dupsTested" : 0,
"dupsDropped" : 0,
"seenInvalidated" : 0
}
}
},
{
"nReturned" : 43,
"executionTimeMillisEstimate" : 1431,
"totalKeysExamined" : 221,
"totalDocsExamined" : 219,
"executionStages" : {
"stage" : "FETCH",
"filter" : {
"timestamp" : {
"$gt" : 1501548359000.0
}
},
"nReturned" : 43,
"executionTimeMillisEstimate" : 1431,
"works" : 221,
"advanced" : 43,
"needTime" : 177,
"needYield" : 0,
"saveState" : 64,
"restoreState" : 64,
"isEOF" : 1,
"invalidates" : 0,
"docsExamined" : 219,
"alreadyHasObj" : 0,
"inputStage" : {
"stage" : "IXSCAN",
"nReturned" : 219,
"executionTimeMillisEstimate" : 71,
"works" : 221,
"advanced" : 219,
"needTime" : 1,
"needYield" : 0,
"saveState" : 64,
"restoreState" : 64,
"isEOF" : 1,
"invalidates" : 0,
"keyPattern" : {
"rack_name" : 1
},
"indexName" : "rack",
"isMultiKey" : false,
"multiKeyPaths" : {
"rack_name" : []
},
"isUnique" : false,
"isSparse" : false,
"isPartial" : false,
"indexVersion" : 2,
"direction" : "forward",
"indexBounds" : {
"rack_name" : [
"[\"helena2\", \"helena2\"]",
"[\"providence1\", \"providence1\"]"
]
},
"keysExamined" : 221,
"seeks" : 2,
"dupsTested" : 0,
"dupsDropped" : 0,
"seenInvalidated" : 0
}
}
}
]
},
"serverInfo" : {
"host" : "dtauto-sna01.mascorp.com",
"port" : 27017,
"version" : "3.4.4",
"gitVersion" : "888390515874a9debd1b6c5d36559ca86b44babd"
},
"ok" : 1.0
}
Two things I do not understand:
When you look at
AllPlansExecutionthenReturnedkey has different value on each plan. Second plan (index: rack) is actually the winner plan and returns 43 results which is the actual return result of the whole query, but what are those 2nReturnedresult from the firstMore challenge comes to this point that why the first plan (index: time which is the rejected plan) has been reported with less
executionTimeMillisvalue 31 than the winner planexecutionTimeMillis1431?
What's going on?
回答1:
This was explained in the allPlansExecution Mode documentation page. To paraphrase:
MongoDB runs the query optimizer to choose the winning plan and executes the winning plan to completion. In "allPlansExecution" mode, MongoDB returns statistics describing the execution of the winning plan as well as statistics for the other candidate plans captured during plan selection.
During plan selection, if there are more than one index that can satisfy a query, MongoDB will run a trial using all the valid plans to determine which one performed to be the best. See Query Plans for details regarding this process.
As of MongoDB 3.4.6, the plan selection involves running candidate plans in parallel in a "race", and see which candidate plan returns 101 results first. In your example above, by the time the winning plan returned 101 results in the race, the losing plan managed only 2 results. The winning plan then gets executed to completion. This is the reason why the losing plan shows only nReturned: 2 in the stats.
This "race" is performed since if there are two identical-looking plans, MongoDB doesn't know which plan is the best for a particular query due to the flexibility of JSON documents (unlike e.g. SQL where the structure of the tables are known). Of course, it is entirely possible that MongoDB guesses wrong, and end up with a less than performant plan, since it is an empirical process. For this reason, it's best to create indexes that supports your queries so MongoDB doesn't have to guess. Otherwise, you can use hint() to tell MongoDB which index to use for a certain query.
Hence:
- The statistics for the winning plan is the actual query's result statistics.
- The statistics for the losing plan(s) only shows the statistics for the query planning trial run.
- The plan selection involves running a "race" to 101 results. This race is only performed when there are multiple indexes that can satisfy the query.
Note 1: Neither of the two plans you saw were great. The winning plan shows "nReturned" : 43, "totalKeysExamined" : 221, and "totalDocsExamined" : 219. This means that MongoDB needs to examine 219 documents only to return 43 of them: only 20% efficiency. Ideally, you want to have the nReturned numbers equal to totalDocsExamined.
Note 2: Try creating the compound index {'rack_name': 1, 'timestamp': 1}. With the same query, you should get a better efficiency number.
Note 3: Note that since allPlansExecution was specified, all statistics is duly returned to you by MongoDB for thoroughness, while it has no bearing whatsoever to the final nReturned result. It was a rejected plan, and the nReturned: 2 number can be confusing. You won't see this statistics if you use executionStats setting. Primarily, the allPlansExecution is used for fine-tuning and determining why some plans are rejected.
来源：https://stackoverflow.com/questions/45497433/why-mongodb-different-query-plans-show-different-nreturned-value