虽然设置了索引，但简单的MongoDB查询非常慢

Question

I've got a MongoDB collection that holds about 100M documents.

The documents basically look like this:

_id             : ObjectId("asd1234567890")
_reference_1_id : ObjectId("fgh4567890123")
_reference_2_id : ObjectId("jkl7890123456")
name            : "Test1"
id              : "4815162342"
created_time    : Date( 1331882436000 )
_contexts       : ["context1", "context2"]
...

There are some indexes set, here's the output of db.mycoll.getIndexes();

[
{
    "v" : 1,
    "key" : {
        "_id" : 1
    },
    "ns" : "mydb.mycoll",
    "name" : "_id_"
},
{
    "v" : 1,
    "key" : {
        "_reference_1_id" : 1,
        "_reference_2_id" : 1,
        "id" : 1
    },
    "unique" : true,
    "ns" : "mydb.mycoll",
    "name" : "_reference_1_id_1__reference_2_id_1_id_1"
},
{
    "v" : 1,
    "key" : {
        "_reference_1_id" : 1,
        "_reference_2_id" : 1,
        "_contexts" : 1,
        "created_time" : 1
    },
    "ns" : "mydb.mycoll",
    "name" : "_reference_1_id_1__reference_2_id_1__contexts_1_created_time_1"
}
]

When I execute a query like

db.mycoll.find({"_reference_2_id" : ObjectId("jkl7890123456")})

it takes over an hour (!) until it's finished, no matter if there are results or not. Any ideas?

Update: Here's what the output of

db.mycoll.find({"_reference_2_id" : ObjectId("jkl7890123456")}).explain();

looks like:

{
"cursor" : "BasicCursor",
"nscanned" : 99209163,
"nscannedObjects" : 99209163,
"n" : 5007,
"millis" : 5705175,
"nYields" : 17389,
"nChunkSkips" : 0,
"isMultiKey" : false,
"indexOnly" : false,
"indexBounds" : {

}
}

Answer 1

You don't have any index that mongo will automatically use for that, so it's doing a full table scan.

As mentioned in the docs

If the first key [of the index] is not present in the query, the index will only be used if hinted explicitly.

Why

If you have an index on a,b - and you search by a alone - an index will automatically be used. This is because it's the start of the index (which is fast to do), the db can just ignore the rest of the index value.

An index on a,b is inefficient when searching by b alone simply because it doesn't give the possibility to use the index searching with "starts with thisfixedstring".

So, either:

• Include _reference_1_id in the query (probably irrelevant)
• OR add an index on _reference_2_id (if you query by the field often)
• OR use a hint

Hint

Probably your lowest-cost option right now.

Add a query hint to force using your _reference_1_id_1__reference_2_id_1_id_1 index. Which is likely to be a lot faster than a full table scan, but still a lot slower than an index which starts with the field you are using in the query.

ie

db.mycoll
    .find({"_reference_2_id" : ObjectId("jkl7890123456")})
    .hint("_reference_1_id_1__reference_2_id_1_id_1");

Answer 2

Hye, I've quiet the same problem on an equivalent amount of datas. In the documentation, it's written that queries with index must fit in ram. I think this is not the case, the query must be doing a lot of disk access to first retrieve the index and then get the value. In your case, a direct collection read will be faster.

EV.

Answer 3

我会尝试在_reference_2_id上设置一个非唯一索引，因为目前我怀疑你将完全相当于全表扫描，即使索引包含_reference_2_id ，它们也不会被使用（见这里 ）。