Unacceptably slow MongoDB $lte + $gte query with index - mongodb

Question

I am running Community MongoDB 3.4.9 on my laptop with 64 GB RAM. I have a collection with 12+ million documents. Each document has at least from and to fields of type Int64. The from-to are unique ranges. There are no documents with overlapping ranges. There is an index on the collection as follows:
{
"v" : NumberInt(1),
"unique" : true,
"key" : {
"from" : NumberInt(1),
"to" : NumberInt(1)
},
"name" : "range",
"ns" : "db.location",
"background" : true
}
The server/database is idle. There are no clients. I run the query below over and over and I get a constant execution time of roughly 21 seconds.
db.location.find({from:{$lte:NumberLong(3682093364)},to:{$gte:NumberLong(3682093364)}}).limit(1)
Reversal of and conditions does not make a difference with respect to execution time. The explain command shows the following.
{
"queryPlanner" : {
"plannerVersion" : 1.0,
"namespace" : "db.location",
"indexFilterSet" : false,
"parsedQuery" : {
"$and" : [
{
"from" : {
"$lte" : NumberLong(3682093364)
}
},
{
"to" : {
"$gte" : NumberLong(3682093364)
}
}
]
},
"winningPlan" : {
"stage" : "LIMIT",
"limitAmount" : 1.0,
"inputStage" : {
"stage" : "FETCH",
"inputStage" : {
"stage" : "IXSCAN",
"keyPattern" : {
"from" : 1.0,
"to" : 1.0
},
"indexName" : "range",
"isMultiKey" : false,
"multiKeyPaths" : {
"from" : [
],
"to" : [
]
},
"isUnique" : true,
"isSparse" : false,
"isPartial" : false,
"indexVersion" : 1.0,
"direction" : "forward",
"indexBounds" : {
"from" : [
"[-inf.0, 3682093364]"
],
"to" : [
"[3682093364, inf.0]"
]
}
}
}
},
"rejectedPlans" : [
]
},
"executionStats" : {
"executionSuccess" : true,
"nReturned" : 1.0,
"executionTimeMillis" : 21526.0,
"totalKeysExamined" : 12284007.0,
"totalDocsExamined" : 1.0,
"executionStages" : {
"stage" : "LIMIT",
"nReturned" : 1.0,
"executionTimeMillisEstimate" : 20945.0,
"works" : 12284008.0,
"advanced" : 1.0,
"needTime" : 12284006.0,
"needYield" : 0.0,
"saveState" : 96299.0,
"restoreState" : 96299.0,
"isEOF" : 1.0,
"invalidates" : 0.0,
"limitAmount" : 1.0,
"inputStage" : {
"stage" : "FETCH",
"nReturned" : 1.0,
"executionTimeMillisEstimate" : 20714.0,
"works" : 12284007.0,
"advanced" : 1.0,
"needTime" : 12284006.0,
"needYield" : 0.0,
"saveState" : 96299.0,
"restoreState" : 96299.0,
"isEOF" : 0.0,
"invalidates" : 0.0,
"docsExamined" : 1.0,
"alreadyHasObj" : 0.0,
"inputStage" : {
"stage" : "IXSCAN",
"nReturned" : 1.0,
"executionTimeMillisEstimate" : 20357.0,
"works" : 12284007.0,
"advanced" : 1.0,
"needTime" : 12284006.0,
"needYield" : 0.0,
"saveState" : 96299.0,
"restoreState" : 96299.0,
"isEOF" : 0.0,
"invalidates" : 0.0,
"keyPattern" : {
"from" : 1.0,
"to" : 1.0
},
"indexName" : "range",
"isMultiKey" : false,
"multiKeyPaths" : {
"from" : [
],
"to" : [
]
},
"isUnique" : true,
"isSparse" : false,
"isPartial" : false,
"indexVersion" : 1.0,
"direction" : "forward",
"indexBounds" : {
"from" : [
"[-inf.0, 3682093364]"
],
"to" : [
"[3682093364, inf.0]"
]
},
"keysExamined" : 12284007.0,
"seeks" : 12284007.0,
"dupsTested" : 0.0,
"dupsDropped" : 0.0,
"seenInvalidated" : 0.0
}
}
},
"allPlansExecution" : [
]
},
"serverInfo" : {
"host" : "LAPTOP-Q96TVSN8",
"port" : 27017.0,
"version" : "3.4.9",
"gitVersion" : "876ebee8c7dd0e2d992f36a848ff4dc50ee6603e"
},
"ok" : 1.0
}
Supplying a hint does not make a difference. explain seems to indicate that the proper (and only) index is already being used but most of the execution time (20s) is spent in IXSCAN. The MongoDB log shows that many index items were scanned but only one document was ever touched and returned. It also shows a crazy number of locks and yields considering there are ZERO concurrent operations on the database. The underlying engine is wiredTiger on an SSD disk. MongoDB RAM usage is at 7 GB.
2017-10-10T10:06:14.456+0200 I COMMAND [conn33] command db.location appName: "MongoDB Shell" command: explain { explain: { find: "location", filter: { from: { $lte: 3682093364 }, to: { $gte: 3682093364 } }, limit: 1.0, singleBatch: false }, verbosity: "allPlansExecution" } numYields:96299 reslen:1944 locks:{ Global: { acquireCount: { r: 192600 } }, Database: { acquireCount: { r: 96300 } }, Collection: { acquireCount: { r: 96300 } } } protocol:op_command 21526ms
Is there a better way to structure the document so that the lookups are faster considering my ranges are never overlapping? Is there something obvious that I am doing wrong?
UPDATE:
When I drop the index, COLLSCAN is used and the document is found in consistent 8-9 seconds.

Answer 1

I hate to answer my own questions but then again I am happy for finding the solution.
Even though it makes sense to create such a composite index, considering the specifics of non-overlapping ranges it turns out that the search scope is just too broad. The higher the input number, the longer it will take to find the result as more and more index entries are found that satisfy from <= number and last result in the search scope is actually the one we are looking for (index is scanned from left to right).
The solution is to modify the index to be either { from: -1 } or { to: 1 }. The composite index is really not necessary in this scenario as the ranges are not overlapping and the very first document found by the index is the very document being returned. This is now lightning fast just as expected.
You live and learn...