-
Type:
Bug
-
Resolution: Done
-
Priority:
Major - P3
-
None
-
Affects Version/s: 2.6.0
-
Component/s: Querying
-
None
-
ALL
-
None
-
None
-
None
-
None
-
None
-
None
-
None
Reproducer is a bit convoluted, but here are the steps:
for (p = 0; p < 10; p++) {
var t = db.foo;
t.drop();
for ( i = 0; i < 10000; i++ ) {
var a = [];
for ( var j = 0; j < 10; j++ ) {
a.push( Math.floor( Math.random() * 2000 ) );
a.push( Math.floor( Math.random() * 20 ) );
}
t.insert( { x : a } );
}
t.ensureIndex( { x : 1 } );
for (j=0;j<20;j++) {
var foo = t.find( { x : { $all : [ j , j*10 ] } } ).explain();
if (foo.cursor == "Complex Plan") throw "Got it with " +j;
}
}
Assuming you got "Got it" printed, you now have j to see result for explain(true) which shows something like:
db.foo.find({x:{$all:[ j,j*10]}}).explain(true)
{
"cursor" : "Complex Plan",
"n" : 15,
"nscannedObjects" : 0,
"nscanned" : 3925,
"nscannedObjectsAllPlans" : 3925,
"nscannedAllPlans" : 7851,
"nYields" : 61,
"nChunkSkips" : 0,
"millis" : 17,
"allPlans" : [
{
"cursor" : "Complex Plan",
"n" : 15,
"nscannedObjects" : 0,
"nscanned" : 3925,
"nChunkSkips" : 0
},
{
"cursor" : "BtreeCursor x_1",
"isMultiKey" : true,
"n" : 15,
"nscannedObjects" : 3925,
"nscanned" : 3926,
"scanAndOrder" : false,
"indexOnly" : false,
"nChunkSkips" : 0,
"indexBounds" : {
"x" : [
[
5,
5
]
]
}
}
],
"server" : "Asyas-MacBook-Pro.local:27017",
"filterSet" : false,
"stats" : {
"type" : "FETCH",
"works" : 3926,
"yields" : 61,
"unyields" : 61,
"invalidates" : 0,
"advanced" : 15,
"needTime" : 3909,
"needFetch" : 0,
"isEOF" : 1,
"alreadyHasObj" : 0,
"forcedFetches" : 0,
"matchTested" : 0,
"children" : [
{
"type" : "KEEP_MUTATIONS",
"works" : 3925,
"yields" : 61,
"unyields" : 61,
"invalidates" : 0,
"advanced" : 15,
"needTime" : 3909,
"needFetch" : 0,
"isEOF" : 1,
"children" : [
{
"type" : "AND_SORTED",
"works" : 3925,
"yields" : 61,
"unyields" : 61,
"invalidates" : 0,
"advanced" : 15,
"needTime" : 3909,
"needFetch" : 0,
"isEOF" : 1,
"flagged" : 0,
"matchTested" : 0,
"failedAnd_0" : 50,
"failedAnd_1" : 35,
"children" : [
{
"type" : "IXSCAN",
"works" : 3874,
"yields" : 61,
"unyields" : 61,
"invalidates" : 0,
"advanced" : 3874,
"needTime" : 0,
"needFetch" : 0,
"isEOF" : 0,
"keyPattern" : "{ x: 1.0 }",
"boundsVerbose" : "field #0['x']: [5.0, 5.0]",
"isMultiKey" : 1,
"yieldMovedCursor" : 0,
"dupsTested" : 3874,
"dupsDropped" : 0,
"seenInvalidated" : 0,
"matchTested" : 0,
"keysExamined" : 3875,
"children" : [ ]
},
{
"type" : "IXSCAN",
"works" : 51,
"yields" : 61,
"unyields" : 61,
"invalidates" : 0,
"advanced" : 50,
"needTime" : 0,
"needFetch" : 0,
"isEOF" : 1,
"keyPattern" : "{ x: 1.0 }",
"boundsVerbose" : "field #0['x']: [50.0, 50.0]",
"isMultiKey" : 1,
"yieldMovedCursor" : 0,
"dupsTested" : 50,
"dupsDropped" : 0,
"seenInvalidated" : 0,
"matchTested" : 0,
"keysExamined" : 50,
"children" : [ ]
}
]
}
]
}
]
}
}
This is racing off index intersection against index scan for first value of $all only (j in which case). If you switch the query to $all:[j*10,j ] you will now see the index scan win against index intersection.
This is a case of an $and query where position actually makes a difference to the query plan which I would not have expected. I would expect a race against intersection, index scan for only first value and index scan for only second value. (in case of more values, then as many of them as we allow for a single race-off).