Indexing with MongoDB

Indexing with
Aaron Staple
aaron@10gen.com

What are indexes?
References to your documents, efficiently ordered by key
Maintained in a tree structure, allowing fast lookup
{x:1}
{y:1}
{x:0.5,y:0.5}
{x:2,y:0.5}
{x:5,y:2}
{x:-4,y:10}
{x:3,y:’f’}

Fast document lookup
db.c.findOne( {_id:2} ), using index {_id:1}
db.c.find( {x:2} ), using index {x:1}
db.c.find( {x:{$in:[2,3]}} ), using index {x:1}
db.c.find( {‘x.a’:1} ), using index {‘x.a’:1}
Matches {_id:1,x:{a:1}}
db.c.find( {x:{a:1}} ), using index {x:1}
Matches {_id:1,x:{a:1}}, but not {_id:2,x:{a:1,b:2}}
QUESTION: What about db.c.find( {$where:“this.x == this.y”} ), using index {x:1}?
Indexes cannot be used for $where type queries, but if there are non-where elements in the query then indexes can be used for the non-where elements.

Fast document range scan
db.c.find( {x:{$gt:2}} ), using index {x:1}
db.c.find( {x:{$gt:2,$lt:5}} ), using index {x:1}
db.c.find( {x:/^a/} ), using index {x:1}
QUESTION: What about db.c.find( {x:/a/} ), using index {x:1}?
The letter ‘a’ can appear anywhere in a matching string, so lexicographic ordering on strings won’t help. However, we can use the index to find the range of documents where x is string (eg not a number) or x is the regular expression /a/.

Other operations
db.c.count( {x:2} ) using index {x:1}
db.c.distinct( {x:2} ) using index {x:1}
db.c.update( {x:2}, {x:3} ) using index {x:1}
db.c.remove( {x:2} ) using index {x:1}
QUESTION: What about db.c.update( {x:2}, {$inc:{x:3}} ), using index {x:1}?
Older versions of mongoDB didn’t support modifiers on indexed fields, but we now support this.

Fast document ordering
db.c.find( {} ).sort( {x:1} ), using index {x:1}
db.c.find( {} ).sort( {x:-1} ), using index {x:1}
db.c.find( {x:{$gt:4}} ).sort( {x:-1} ), using index {x:1}
db.c.find( {} ).sort( {‘x.a’:1} ), using index {‘x.a’:1}
QUESTION: What about db.c.find( {y:1} ).sort( {x:1} ), using index {x:1}?
The index will be used to ensure ordering, provided there is no better index.

Missing fields
db.c.find( {x:null} ), using index {x:1}
Matches {_id:5}
db.c.find( {x:{$exists:false}} ), using index {x:1}
Matches {_id:5}, but not {_id:6,x:null}
QUESTION: What about db.c.find( {x:{$exists:true}} ), using index {x:1}?
The index is not currently used, though we may use the index in a future version of mongoDB.

Array matching
All the following match {_id:6,x:[2,10]} and use index {x:1}
db.c.find( {x:2} )
db.c.find( {x:10} )
db.c.find( {x:{$gt:5}} )
db.c.find( {x:[2,10]} )
db.c.find( {x:{$in:[2,5]}} )
QUESTION: What about db.c.find( {x:{$all:[2,10]}} )?
The index will be used to look up all documents matching {x:2}.

Compound Indexes
db.c.find( {x:10,y:20} ), using index {x:1,y:1}
db.c.find( {x:10,y:20} ), using index {x:1,y:-1}
db.c.find( {x:{$in:[10,20]},y:20} ), using index {x:1,y:1}
db.c.find().sort( {x:1,y:1} ), using index {x:1,y:1}
db.c.find().sort( {x:-1,y:1} ), using index {x:1,y:-1}
db.c.find( {x:10} ).sort( {y:1} ), using index {x:1,y:1}
QUESTION: What about db.c.find( {y:10} ).sort( {x:1} ), using index {x:1,y:1}?
The index will be used to ensure ordering, provided no better index is available.

When indexes are less helpful
db.c.find( {x:{$ne:1}} )
db.c.find( {x:{$mod:[10,1]}} )
Uses index {x:1} to scan numbers only
db.c.find( {x:{$not:/a/}} )
db.c.find( {x:{$gte:0,$lte:10},y:5} ) using index {x:1,y:1}
Currently must scan all elements from {x:0,y:5} to {x:10,y:5}, but some improvements may be possible
db.c.find( {$where:’this.x = 5’} )
QUESTION: What about db.c.find( {x:{$not:/^a/}} ), using index {x:1}?
The index is not used currently, but will be used in mongoDB 1.6

Geospatial indexes
db.c.find( {a:[50,50]} ) using index {a:’2d’}
db.c.find( {a:{$near:[50,50]}} ) using index {a:’2d’}
Results are sorted closest - farthest
db.c.find( {a:{$within:{$box:[[40,40],[60,60]]}}} ) using index {a:’2d’}
db.c.find( {a:{$within:{$center:[[50,50],10]}}} ) using index {a:’2d’}
db.c.find( {a:{$near:[50,50]},b:2} ) using index {a:’2d’,b:1}
QUESTION: Most queries can be performed with or without an index. Is this true of geospatial queries?
No. A geospatial query requires an index.

Creating indexes
{_id:1} index created automatically
For non-capped collections
db.c.ensureIndex( {x:1} )
Can create an index at any time, even when you already have plenty of data in your collection
Creating an index will block mongoDB unless you specify background index creation
db.c.ensureIndex( {x:1}, {background:true} )
Background index creation is a still impacts performance – run at non peak times if you’re concerned
QUESTION: Can an index be removed during background creation?
Not at this time.

Unique key constraints
db.c.ensureIndex( {x:1}, {unique:true} )
Don’t allow {_id:10,x:2} and {_id:11,x:2}
Don’t allow {_id:12} and {_id:13} (both match {x:null}
What if duplicates exist before index is created?
Normally index creation fails and the index is removed
db.ensureIndex( {x:1}, {unique:true,dropDups:true} )
QUESTION: In dropDups mode, which duplicates will be removed?
The first document according to the collection’s “natural order” will be preserved.

Cleaning up indexes
db.system.indexes.find( {ns:’db.c’} )
db.c.dropIndex( {x:1} )
db.c.dropIndexes()
db.c.reIndex()
Rebuilds all indexes, removing index cruft that has built up over large numbers of updates and deletes. Index cruft will not exist in mongoDB 1.6, so this command will be deprecated.
QUESTION: Why would you want to drop an index?
See next slide…

Limits and Tradeoffs
Max 40 indexes per collection
Logically equivalent indexes are not prevented (eg {x:1} and {x:-1})
Indexes can improve speed of queries, but make inserts slower
More specific indexes {a:1,b:1,c:1} can be more helpful than less specific indexes {a:1}, but sorting compound keys may not be as fast as sorting simple keys
QUESTION: Do indexes make updates slower? How about deletes?
It depends – finding your document might be faster, but if any indexed fields are changed the indexes must be updated.

Query Optimizer
In charge of picking which index to use for a query/count/update/delete/etc
Implementation is part of the magic of mongo (you can read about it online – not covering today)
Usually it does a good job, but if you know what you’re doing you can override it
db.c.find( {x:2,y:3} ).hint( {y:1} )
Use index {y:1} and avoid trying out {x:1}
As your data changes, different indexes may be chosen. Ordering requirements should be made explicit using sort().
QUESTION: How can you force a full collection scan instead of using indexes?
db.c.find( {x:2,y:3} ).hint( {$natural:1} )

Mongod log output
query test.c ntoreturn:1 reslen:69 nscanned:100000 { i: 99999.0 } nreturned:1 157ms
query test.$cmd ntoreturn:1 command: { count: "c", query: { type: 0.0, i: { $gt: 99000.0 } }, fields: {} } reslen:64 256ms
query:{ query: {}, orderby: { i: 1.0 } } ... query test.c ntoreturn:0 exception 1378ms ... User Exception 10128:too much key data for sort() with no index. add an index or specify a smaller limit
query test.c ntoreturn:0 reslen:4783 nscanned:100501 { query: { type: 500.0 }, orderby: { i: 1.0 } } nreturned:101 390ms
Occasionally may see a slow operation as a result of disk activity or mongo cleaning things up – some messages about slow ops are spurious
Keep this in mind when running the same op a massive number of times, and it appears slow very rarely

Profiling
Record same info as with log messages, but in a database collection
> db.system.profile.find()
{"ts" : "Thu Jan 29 2009 15:19:32 GMT-0500 (EST)" , "info" : "query test.$cmd ntoreturn:1 reslen:66 nscanned:0 <br>query: { profile: 2 } nreturned:1 bytes:50" , "millis" : 0}...
> db.system.profile.find( { info: /test.foo/ } )
> db.system.profile.find( { millis : { $gt : 5 } } )
> db.system.profile.find().sort({$natural:-1})
Enable explicitly using levels (0:off, 1:slow ops (>100ms), 2:all ops)
> db.setProfilingLevel(2);
{"was" : 0 , "ok" : 1}
> db.getProfilingLevel()
2
> db.setProfilingLevel( 1 , 10 ); // slow means > 10ms
Profiling impacts performance, but not severely

Query explain
> db.c.find( {x:1000,y:0} ).explain()
{
"cursor" : "BtreeCursor x_1",
"indexBounds" : [
[
{
"x" : 1000
},
{
"x" : 1000
}
]
],
"nscanned" : 10,
"nscannedObjects" : 10,
"n" : 10,
"millis" : 0,
"oldPlan" : {
"indexBounds" : [
[
{
"x" : 1000
},
{
"x" : 1000
}
]
]
},
"allPlans" : [
{
"indexBounds" : [
[
{
"x" : 1000
},
{
"x" : 1000
}
]
]
},
{
"cursor" : "BtreeCursor y_1",
"indexBounds" : [
[
{
"y" : 0
},
{
"y" : 0
}
]
]
},
{
"cursor" : "BasicCursor",
"indexBounds" : [ ]
}
]
}

Example 1
> db.c.findOne( {i:99999} )
{ "_id" : ObjectId("4bb962dddfdcf5761c1ec6a3"), "i" : 99999 }
query test.c ntoreturn:1 reslen:69 nscanned:100000 { i: 99999.0 } nreturned:1 157ms
> db.c.find( {i:99999} ).limit(1).explain()
{
"indexBounds" : [ ],
"nscanned" : 100000,
"n" : 1,
"millis" : 161,
"allPlans" : [
{
"indexBounds" : [ ]
}
]
}
> db.c.ensureIndex( {i:1} );
> for( i = 0; i < 100000; ++i ) { db.c.save( {i:i} ); }

Example 2
> db.c.count( {type:0,i:{$gt:99000}} )
499
query test.$cmd ntoreturn:1 command: { count: "c", query: { type: 0.0, i: { $gt: 99000.0 } }, fields: {} } reslen:64 256ms
> db.c.find( {type:0,i:{$gt:99000}} ).limit(1).explain()
{
"cursor" : "BtreeCursor type_1",
"indexBounds" : [
[
{
"type" : 0
},
{
"type" : 0
}
]
],
"nscanned" : 49502,
"n" : 1,
"millis" : 349,
...
> db.c.ensureIndex( {type:1,i:1} );
> for( i = 0; i < 100000; ++i ) { db.c.save( {type:i%2,i:i} ); }

Example 3
> db.c.find().sort( {i:1} )
error: {
"$err" : "too much key data for sort() with no index. add an index or specify a smaller limit"
}
> db.c.find().sort( {i:1} ).explain()
JS Error: uncaught exception: error: {
"$err" : "too much key data for sort() with no index. add an index or specify a smaller limit"
}
> db.c.ensureIndex( {i:1} );
> for( i = 0; i < 1000000; ++i ) { db.c.save( {i:i} ); }

Example 4
> db.c.find( {type:500} ).sort( {i:1} )
{ "_id" : ObjectId("4bba4904dfdcf5761c2f917e"), "i" : 500, "type" : 500 }
{ "_id" : ObjectId("4bba4904dfdcf5761c2f9566"), "i" : 1500, "type" : 500 }
...
query test.c ntoreturn:0 reslen:4783 nscanned:100501 { query: { type: 500.0 }, orderby: { i: 1.0 } } nreturned:101 390ms
> db.c.find( {type:500} ).sort( {i:1} ).explain()
{
"cursor" : "BtreeCursor i_1",
"indexBounds" : [
[
{
"i" : {
"$minElement" : 1
}
},
{
"i" : {
"$maxElement" : 1
}
}
]
],
"nscanned" : 1000000,
"n" : 1000,
"millis" : 5388,
...
> db.c.ensureIndex( {type:1,i:1} );
> for( i = 0; i < 1000000; ++i ) { db.c.save( {i:i,type:i%1000} ); }

Questions?
Follow @mongodb
Get involved www.mongodb.org
Upcoming events www.mongodb.org/display/DOCS/Events
MongoSF April 30
SF office hours every Mon 4-6pm Epicenter Cafe
Commercial support www.10gen.com
jobs@10gen.com

http://nosql.pl	977
http://www.slideshare.net	179
http://www.10gen.com	171
http://www.mongodb.org	3
http://nosqlpl.tumblr.com	2
http://www.tumblr.com	1
http://webcache.googleusercontent.com	1
http://www.babbleapp.org	1
http://dl.mongodb.org	1
https://www.10gen.com	1
http://feeds.feedburner.com	1

Indexing with MongoDB

by Dwight Merriman on Apr 06, 2010

Accessibility

Tags

Upload Details

Usage Rights

Embed Views

Statistics

Indexing with MongoDB — Presentation Transcript

Get analytics right now with SlideShare PRO

An effective way to find out who is viewing your content

Remove ads from all your pages with SlideShare PRO

Your viewers will engage with your content, not ads