在查询时,您关心两件事情-准确性和性能。 出于这个目的,我在MongoDB v3.0.14中测试了几种不同的方法。
简而言之,db.doc.find({ nums: { $gt: -Infinity }})
是最快且最可靠的方法(至少在我测试的MongoDB版本中)。
编辑:这在MongoDB v3.6中已不再起作用!请参见本帖子下面的评论以获取潜在解决方案。
设置
我插入了1000个没有列表字段的文档,1000个带有空列表的文档以及5个带有非空列表的文档。
for (var i = 0; i < 1000; i++) { db.doc.insert({}); }
for (var i = 0; i < 1000; i++) { db.doc.insert({ nums: [] }); }
for (var i = 0; i < 5; i++) { db.doc.insert({ nums: [1, 2, 3] }); }
db.doc.createIndex({ nums: 1 });
我知道这个规模不足以像下面的测试一样认真对待性能,但已足以展示所选查询计划的正确性和各种查询行为。
测试
db.doc.find({'nums': {'$exists': true}})
返回错误结果(与我们想要实现的目标不符)。
MacBook-Pro(mongod-3.0.14) test> db.doc.find({'nums': {'$exists': true}}).count()
1005
db.doc.find({'nums.0': {'$exists': true}})
返回正确结果,但也很慢,需要使用完整的集合扫描(注意解释中的 COLLSCAN
阶段)。
MacBook-Pro(mongod-3.0.14) test> db.doc.find({'nums.0': {'$exists': true}}).count()
5
MacBook-Pro(mongod-3.0.14) test> db.doc.find({'nums.0': {'$exists': true}}).explain()
{
"queryPlanner": {
"plannerVersion": 1,
"namespace": "test.doc",
"indexFilterSet": false,
"parsedQuery": {
"nums.0": {
"$exists": true
}
},
"winningPlan": {
"stage": "COLLSCAN",
"filter": {
"nums.0": {
"$exists": true
}
},
"direction": "forward"
},
"rejectedPlans": [ ]
},
"serverInfo": {
"host": "MacBook-Pro",
"port": 27017,
"version": "3.0.14",
"gitVersion": "08352afcca24bfc145240a0fac9d28b978ab77f3"
},
"ok": 1
}
db.doc.find({'nums': { $exists: true, $gt: { '$size': 0 }}})
的返回结果有误。这是因为索引扫描无效,导致不会返回任何文档。如果没有索引,查询可能会正确但是缓慢。
MacBook-Pro(mongod-3.0.14) test> db.doc.find({'nums': { $exists: true, $gt: { '$size': 0 }}}).count()
0
MacBook-Pro(mongod-3.0.14) test> db.doc.find({'nums': { $exists: true, $gt: { '$size': 0 }}}).explain('executionStats').executionStats.executionStages
{
"stage": "KEEP_MUTATIONS",
"nReturned": 0,
"executionTimeMillisEstimate": 0,
"works": 2,
"advanced": 0,
"needTime": 0,
"needFetch": 0,
"saveState": 0,
"restoreState": 0,
"isEOF": 1,
"invalidates": 0,
"inputStage": {
"stage": "FETCH",
"filter": {
"$and": [
{
"nums": {
"$gt": {
"$size": 0
}
}
},
{
"nums": {
"$exists": true
}
}
]
},
"nReturned": 0,
"executionTimeMillisEstimate": 0,
"works": 1,
"advanced": 0,
"needTime": 0,
"needFetch": 0,
"saveState": 0,
"restoreState": 0,
"isEOF": 1,
"invalidates": 0,
"docsExamined": 0,
"alreadyHasObj": 0,
"inputStage": {
"stage": "IXSCAN",
"nReturned": 0,
"executionTimeMillisEstimate": 0,
"works": 1,
"advanced": 0,
"needTime": 0,
"needFetch": 0,
"saveState": 0,
"restoreState": 0,
"isEOF": 1,
"invalidates": 0,
"keyPattern": {
"nums": 1
},
"indexName": "nums_1",
"isMultiKey": true,
"direction": "forward",
"indexBounds": {
"nums": [
"({ $size: 0.0 }, [])"
]
},
"keysExamined": 0,
"dupsTested": 0,
"dupsDropped": 0,
"seenInvalidated": 0,
"matchTested": 0
}
}
}
db.doc.find({'nums': { $exists: true, $not: { '$size': 0 }}})
返回正确的结果,但性能较差。它在技术上执行了索引扫描,但然后仍然需要遍历所有文档并对它们进行过滤。
MacBook-Pro(mongod-3.0.14) test> db.doc.find({'nums': { $exists: true, $not: { '$size': 0 }}}).count()
5
MacBook-Pro(mongod-3.0.14) test> db.doc.find({'nums': { $exists: true, $not: { '$size': 0 }}}).explain('executionStats').executionStats.executionStages
{
"stage": "KEEP_MUTATIONS",
"nReturned": 5,
"executionTimeMillisEstimate": 0,
"works": 2016,
"advanced": 5,
"needTime": 2010,
"needFetch": 0,
"saveState": 15,
"restoreState": 15,
"isEOF": 1,
"invalidates": 0,
"inputStage": {
"stage": "FETCH",
"filter": {
"$and": [
{
"nums": {
"$exists": true
}
},
{
"$not": {
"nums": {
"$size": 0
}
}
}
]
},
"nReturned": 5,
"executionTimeMillisEstimate": 0,
"works": 2016,
"advanced": 5,
"needTime": 2010,
"needFetch": 0,
"saveState": 15,
"restoreState": 15,
"isEOF": 1,
"invalidates": 0,
"docsExamined": 2005,
"alreadyHasObj": 0,
"inputStage": {
"stage": "IXSCAN",
"nReturned": 2005,
"executionTimeMillisEstimate": 0,
"works": 2015,
"advanced": 2005,
"needTime": 10,
"needFetch": 0,
"saveState": 15,
"restoreState": 15,
"isEOF": 1,
"invalidates": 0,
"keyPattern": {
"nums": 1
},
"indexName": "nums_1",
"isMultiKey": true,
"direction": "forward",
"indexBounds": {
"nums": [
"[MinKey, MaxKey]"
]
},
"keysExamined": 2015,
"dupsTested": 2015,
"dupsDropped": 10,
"seenInvalidated": 0,
"matchTested": 0
}
}
}
db.doc.find({'nums': { $exists: true, $ne: [] }})
返回正确结果,速度略快,但性能仍不理想。它使用IXSCAN只推进具有现有列表字段的文档,但然后必须逐个过滤空列表。
MacBook-Pro(mongod-3.0.14) test> db.doc.find({'nums': { $exists: true, $ne: [] }}).count()
5
MacBook-Pro(mongod-3.0.14) test> db.doc.find({'nums': { $exists: true, $ne: [] }}).explain('executionStats').executionStats.executionStages
{
"stage": "KEEP_MUTATIONS",
"nReturned": 5,
"executionTimeMillisEstimate": 0,
"works": 1018,
"advanced": 5,
"needTime": 1011,
"needFetch": 0,
"saveState": 15,
"restoreState": 15,
"isEOF": 1,
"invalidates": 0,
"inputStage": {
"stage": "FETCH",
"filter": {
"$and": [
{
"$not": {
"nums": {
"$eq": [ ]
}
}
},
{
"nums": {
"$exists": true
}
}
]
},
"nReturned": 5,
"executionTimeMillisEstimate": 0,
"works": 1017,
"advanced": 5,
"needTime": 1011,
"needFetch": 0,
"saveState": 15,
"restoreState": 15,
"isEOF": 1,
"invalidates": 0,
"docsExamined": 1005,
"alreadyHasObj": 0,
"inputStage": {
"stage": "IXSCAN",
"nReturned": 1005,
"executionTimeMillisEstimate": 0,
"works": 1016,
"advanced": 1005,
"needTime": 11,
"needFetch": 0,
"saveState": 15,
"restoreState": 15,
"isEOF": 1,
"invalidates": 0,
"keyPattern": {
"nums": 1
},
"indexName": "nums_1",
"isMultiKey": true,
"direction": "forward",
"indexBounds": {
"nums": [
"[MinKey, undefined)",
"(undefined, [])",
"([], MaxKey]"
]
},
"keysExamined": 1016,
"dupsTested": 1015,
"dupsDropped": 10,
"seenInvalidated": 0,
"matchTested": 0
}
}
}
db.doc.find({'nums': { $gt: [] }})
很危险,因为它取决于使用的索引,可能会给出意外结果。这是由于无效的索引扫描导致未推进任何文档。
MacBook-Pro(mongod-3.0.14) test> db.doc.find({'nums': { $gt: [] }}).count()
0
MacBook-Pro(mongod-3.0.14) test> db.doc.find({'nums': { $gt: [] }}).hint({ nums: 1 }).count()
0
MacBook-Pro(mongod-3.0.14) test> db.doc.find({'nums': { $gt: [] }}).hint({ _id: 1 }).count()
5
MacBook-Pro(mongod-3.0.14) test> db.doc.find({'nums': { $gt: [] }}).explain('executionStats').executionStats.executionStages
{
"stage": "KEEP_MUTATIONS",
"nReturned": 0,
"executionTimeMillisEstimate": 0,
"works": 1,
"advanced": 0,
"needTime": 0,
"needFetch": 0,
"saveState": 0,
"restoreState": 0,
"isEOF": 1,
"invalidates": 0,
"inputStage": {
"stage": "FETCH",
"filter": {
"nums": {
"$gt": [ ]
}
},
"nReturned": 0,
"executionTimeMillisEstimate": 0,
"works": 1,
"advanced": 0,
"needTime": 0,
"needFetch": 0,
"saveState": 0,
"restoreState": 0,
"isEOF": 1,
"invalidates": 0,
"docsExamined": 0,
"alreadyHasObj": 0,
"inputStage": {
"stage": "IXSCAN",
"nReturned": 0,
"executionTimeMillisEstimate": 0,
"works": 1,
"advanced": 0,
"needTime": 0,
"needFetch": 0,
"saveState": 0,
"restoreState": 0,
"isEOF": 1,
"invalidates": 0,
"keyPattern": {
"nums": 1
},
"indexName": "nums_1",
"isMultiKey": true,
"direction": "forward",
"indexBounds": {
"nums": [
"([], BinData(0, ))"
]
},
"keysExamined": 0,
"dupsTested": 0,
"dupsDropped": 0,
"seenInvalidated": 0,
"matchTested": 0
}
}
}
db.doc.find({'nums.0': {$gt: -Infinity}})
返回正确的结果,但性能较差(使用完整的集合扫描)。
MacBook-Pro(mongod-3.0.14) test> db.doc.find({'nums.0': { $gt: -Infinity }}).count()
5
MacBook-Pro(mongod-3.0.14) test> db.doc.find({'nums.0': { $gt: -Infinity }}).explain('executionStats').executionStats.executionStages
{
"stage": "COLLSCAN",
"filter": {
"nums.0": {
"$gt": -Infinity
}
},
"nReturned": 5,
"executionTimeMillisEstimate": 0,
"works": 2007,
"advanced": 5,
"needTime": 2001,
"needFetch": 0,
"saveState": 15,
"restoreState": 15,
"isEOF": 1,
"invalidates": 0,
"direction": "forward",
"docsExamined": 2005
}
db.doc.find({'nums': { $gt: -Infinity }})
很令人惊讶地,它非常有效! 它可以提供正确的结果,并且速度快,从索引扫描阶段提前5个文档。
MacBook-Pro(mongod-3.0.14) test> db.doc.find({'nums': { $gt: -Infinity }}).explain('executionStats').executionStats.executionStages
{
"stage": "FETCH",
"nReturned": 5,
"executionTimeMillisEstimate": 0,
"works": 16,
"advanced": 5,
"needTime": 10,
"needFetch": 0,
"saveState": 0,
"restoreState": 0,
"isEOF": 1,
"invalidates": 0,
"docsExamined": 5,
"alreadyHasObj": 0,
"inputStage": {
"stage": "IXSCAN",
"nReturned": 5,
"executionTimeMillisEstimate": 0,
"works": 15,
"advanced": 5,
"needTime": 10,
"needFetch": 0,
"saveState": 0,
"restoreState": 0,
"isEOF": 1,
"invalidates": 0,
"keyPattern": {
"nums": 1
},
"indexName": "nums_1",
"isMultiKey": true,
"direction": "forward",
"indexBounds": {
"nums": [
"(-inf.0, inf.0]"
]
},
"keysExamined": 15,
"dupsTested": 15,
"dupsDropped": 10,
"seenInvalidated": 0,
"matchTested": 0
}
}
ME.find({ pictures: { $gt: [] } })
也是危险的。如果你在列表字段上建立了索引并且该索引在查询过程中被利用,你将会得到意想不到的结果。例如:db.doc.find({'nums': { $gt: [] }}).hint({ _id: 1 }).count()
返回正确的数量,而db.doc.find({'nums': { $gt: [] }}).hint({ nums: 1 }).count()
返回0
。 - wojcikstefan