Webinar: Index Tuning and Evaluation

2
Problem statement:
I inherited a MongoDB database server with
60 collections and 100 or so indexes.
The business users are complaining about
slow report completion times.
What can I do to improve performance ?

3
Scope:
System tuning-
Memory
Process
Disk
Network
Application tuning-
Application architecture
Statement design
Data model design
Indexing, Query optimization
(Relational, so that we may compare/contrast)

4
What you will
leave with in 60
minutes:
• Detail command processing stages
• Can apply the 5 rules to a Rule Based query optimizer
• Apply 3 Index Negation guidelines
• Repair common query design problems-
• Psuedo order by clause
• OR topped queries
• (And topped queries, non-compound)
• Drop and analyze query plans
• Articulate/control FJS, & ESR query processing
patterns

5
Queries 1 & 2:
SELECT * FROM phonebook
WHERE lastName LIKE “?son”;
SELECT * FROM phonebook
WHERE firstName = “Jennifer”;

6
Query 3:
CREATE TABLE t1 (col1, col_2, .. 80 more columns);
CREATE INDEX i1 ON t1 (col_gender);
SELECT * FROM t1
WHERE col_gender = “F” AND col_age > 50;

7
Negation of an index
1.Non-initial substring
2.Non-anchored compound/composite key
3.(Poor) selectivity of a filter
Partial negation of an index-
CREATE INDEX i1 ON t2 (col1, col2);
//
SELECT * FROM t1
WHERE col1 > 100 AND col2 = “x”;
Exception to all above-
Covered query/key-only

8
(n) Stage database server back end

9
Query Optimizers
Rule Cost ?

10
5 Rules to a rule based optimizer
1.Outer table joins
2.(Non-outer) table joins
3.Filter criteria (predicates)
4.Table size
5.Table cardinality
SELECT *
FROM orderHeader, orderLineItems
WHERE
oh.orderNumber =
oi.orderNumber;
SELECT *
FROM persons, OUTER automobiles
WHERE
p.personId = a.personId;

11
Query 4: final/larger example using SQL

14
Collection access method: collection scan versus index scan

15
Query 4: Join and predicate

17
FJS versus ESR: MongoDB
SELECT *
FROM collection
WHERE
col1 = ‘x’ and col2 > ‘y’
ORDER BY col3;
Filter -> Join -> Sort (FJS)
Equality -> Sort -> Range (ESR)

18
Problem statement:
I inherited a MongoDB database server with
60 collections and 100 or so indexes.
The business users are complaining about
slow report completion times.
What can I do to improve performance ?

19
Skills/tools we need:
• Which server
• Which logfile
• Server profiling Level
• Which queries
• Cloud/Ops Manager !!
• mtools
• Text processing
• Drop the query plan
• Analyze the query plan
• Which indexes get used
• Other

20
Sample data set: zips.json
> db.zips.findOne()
{
"_id" : ObjectId("570 .. 1c1f2"),
"city" : "ACMAR",
"zip" : "35004",
"loc" : {
"y" : 33.584132,
"x" : 86.51557
},
"pop" : 6055,
"state" : "AL"
}
> db.zips.count()
29470
>db.zips.find( { "state" : "WI", "pop" : { "$lt"
: 50 } } ).sort( { "city" : 1 } )

21
Query 5: Dumping the query plan
db.zips.find( { "state" : "WI",
"pop" : { "$lt" : 50 } } ).sort(
{ "city" : 1 }
).explain("executionStats")
"winningPlan" : {
"stage" : "SORT", "sortPattern" : { "city" : 1 },
"inputStage" : {
"stage" : "COLLSCAN",
"filter" : {
"$and" : [ "state" : { "$eq" : "WI" "pop" : { "$lt" : 50
"rejectedPlans" : [ ]
"executionStats" : {
"nReturned" : 4,
"executionTimeMillis" : 16,
"totalKeysExamined" : 0,
"totalDocsExamined" : 29470,

22
Query 5: get indexes
> db.zips.getIndexes()
[ {
"v" : 1, "key" : { "_id" : 1 },
"name" : "_id_",
"ns" : "test_db.zips"
} ]
db.zips.createIndex( { "state" : 1 , "pop" : 1 } )

23
Query 5: attempt 2
winningPlan" : {
"stage" : "SORT", "sortPattern" : { "city" : 1 },
"inputStage" : {
"stage" : "FETCH",
"inputStage" : {
"stage" : "IXSCAN",
"keyPattern" : { "state" : 1, "pop" : 1 },
"indexBounds" : {
"state" : [ "["WI", "WI"]" ],
"pop" : [ "[-inf.0, 50.0)" ]
"nReturned" : 4,

24
Query 5:
attempt 3
db.zips.createIndex( { "state" : 1 , "city" : 1 , "pop" : 1 } )
"winningPlan" : {
"stage" : "SORT", "sortPattern" : { "city" : 1
"inputStage" : {
"stage" : "FETCH",
"inputStage" : {
"stage" : "IXSCAN",
"keyPattern" : { "state" : 1, "pop" : 1
"rejectedPlans" : [
"stage" : "FETCH",
"inputStage" : {
"stage" : "IXSCAN",
"keyPattern" : { "state" : 1, "city" : 1, "pop" : 1 },
"indexBounds" : {
"state" : [ "["WI", "WI"]" ],
"city" : [ "[MinKey, MaxKey]" ],
"pop" : [ "[-inf.0, 50.0)" ]

25
Query 6: (pseudo order by clause)
db.zips.find( { "state" : "CO" }
).sort( { "pop" : 1 } )
SELECT * FROM t1
WHERE col1 = ‘x’
ORDER BY col2;
SELECT * FROM t1
ORDER BY col1, col2;
SELECT * FROM t1
ORDER BY ‘x’, col2;

26
Query 6: query plan
"stage" : "FETCH",
"inputStage" : {
"stage" : "IXSCAN",
"keyPattern" : { "state" : 1, "city" : 1, "pop" : 1 },
"indexBounds" : {
"state" : [ "["CO", "CO"]" ],
"city" : [ "[MinKey, MaxKey]" ],
"pop" : [ "[MinKey, MaxKey]" ]
"nReturned" : 416,
"winningPlan" : {
"stage" : "FETCH",
"inputStage" : {
"stage" : "IXSCAN",
"indexBounds" : {
"state" : [ "["CO", "CO"]" ],
"rejectedPlans" : [
"stage" : "SORT",
"sortPattern" : { "pop" : 1 },
"inputStage" : {

27
Review the indexes we have so far
> db.zips.getIndexes()
_id
db.zips.createIndex( { "state" : 1 , "pop" : 1 } )
db.zips.createIndex( { "state" : 1 , "city" : 1 , "pop" : 1 } )

28
Query 7: OR topped query
db.zips.find( { "$or" : [ { "state" : "UT" }, { "pop" : 2 } ] } )
"winningPlan" : {
"inputStage" : {
"stage" : "COLLSCAN",
"filter" : {
"$or" : [
"pop" : { "$eq" : 2 }
"state" : { "$eq" : "UT" }
"nReturned" : 215,
SELECT * FROM t1
WHERE
order_date = TODAY
OR
ship_weight < 10;

29
Query 7: solution
"stage" : "IXSCAN",
"keyPattern" : { "pop" : 1 },
"indexBounds" : {
"pop" : [ "[2.0, 2.0]" ]
"nReturned" : 215,
db.zips.createIndex( { "pop" : 1 } )
"winningPlan" : {
"inputStage" : {
"stage" : "FETCH",
"inputStage" : {
"stage" : "OR",
"inputStages" : [
"stage" : "IXSCAN",
"indexBounds" : {
"state" : [ "["UT", "UT"]" ],

30
Topics not previously
covered
• How to tell which indexes are being used
• How to tell if an index is unique
• Smoke tests
• Covered queries
• MongoDB index types
• When do winning query plans get evacuated
• Index intersection
• Building indexes (online/offline)
• Sharding and queries, query plans
• Capped collections, tailable cursors
• Optimizer hints
• Memory limits
• Query rewrite (aggregation pipeline optimization)
• Which server
• Which logfile
• (Server profiling Level)
• Which queries
• mtools
• Text processing

31
Resources:
• The parent to this preso,
https://github.com/farrell0/MongoDB-Developers-Notebook
• An excellent query primer (110 pages)
http://www.redbooks.ibm.com/abstracts/sg247138.html?Open
(Chapters 10 and 11.)
• University.MongoDB.com
• https://www.mongodb.com/consulting-test#performance_evaluation
• zips.json
http://media.mongodb.org/zips.json
• Call Dave Lutz, at home, .. .. On Sunday (early)
(512)555/1212

How to tell which indexes are being used
db.zips.aggregate( [ { "$indexStats" : {} } ] ).pretty()
{ "name" : "pop_1",
"key" : { "pop" : 1 },
"host" : "rhhost00.grid:27017",
"accesses" : {
"ops" : NumberLong(15),
"since" : ISODate("2016-04-19T07:13:44.546Z") } }
{ "name" : "state_1_city_1_pop_1",
"key" : { "state" : 1, "city" : 1, "pop" : 1 },
"host" : "rhhost00.grid:27017",
"accesses" : {
"ops" : NumberLong(0),
"since" : ISODate("2016-04-19T06:49:11.765Z") } }

How to tell if an index is unique
db.t1.createIndex( { "k1" : 1 },
{ "unique" : true })
{
"createdCollectionAutomatically" : true,
"numIndexesBefore" : 1,
"numIndexesAfter" : 2,
"ok" : 1
}
> db.t1.getIndexes()
[ { "v" : 1,
"key" : { "_id" : 1 },
"name" : "_id_",
"ns" : "test_db.t1“ },
{ "v" : 1,
"unique" : true,
"key" : { "k1" : 1 },
"name" : "k1_1",
"ns" : "test_db.t1“ }
]

Smoke tests
Every night, gather a set of statistics about your hard disk fullness,
and about the performance of a set of queries that are strategic to
the application.
For queries we wish to record-
• The number of documents returned
• The winning query plan
• Elapsed time, disk and memory consumed
• Other

Covered queries
> db.zips.find( { "pop" : { "$lt" : 200 } },
{ "_id" : 0, "pop" : 1 } ).sort(
{ "pop" : -1 } ).explain()
"winningPlan" : {
"stage" : "PROJECTION",
"transformBy" : {
"_id" : 0,
"pop" : 1 },
"inputStage" : {
"stage" : "IXSCAN",
"keyPattern" : { "pop" : 1 },
"indexBounds" : {
"pop" : [ "(200.0, -inf.0]" ]
}

When does the winning plan get evacuated
In short, the cached query plan is re-evaluated if:
• The collection receives 1000 or more writes
• An index is added or dropped
• A reindex operation is performed
• mongod is restarted
• You run a query with explain

Index intersection
db.zips.find( { "$or" : [ { "state" : "UT" }, { "pop" : 2 } ] } )
db.zips.find( { "city" : "EAST TROY", "zip" : 53120 } )

Building indexes
db.zips.createIndex( { “zip” : 1 }, { “background” : true } )

Capped collections
db.createCollection("my_collection",
{ capped : true, size : 5242880,
max : 5000 } )
from pymongo import Connection
import time
db = Connection().my_db
coll = db.my_collection
cursor = coll.find(tailable=True)
while cursor.alive:
try:
doc = cursor.next()
print doc
except StopIteration:
time.sleep(1)

Memory limits: 32 MB, 100 MB
"executionStages" : {
"stage" : "SORT",
"nReturned" : 1,
"executionTimeMillisEstimate" : 60,
…
"sortPattern" : { "City" : 1 },
"memUsage" : 120,
"memLimit" : 33554432,
db.zips.aggregate([
{ "$group" :
{ "_id" : "$state",
//
"totalPop" : { "$sum" : "$pop" },
"cityCount" : { "$sum" : 1 }
}
} ,
{ "$sort" : { "_id" : 1 } }
],
{ "allowDiskUse" : true }
)

43
But first:
Y/N I have more than 24 months
experience with SQL
Y/N I have more than 6 months
experience with MongoDB
Y/N I have dropped a MongoDB
explain plan, understood it,
made changes, and was happy
Y/N Puppies scare me

44
Two more examples: Queries 8 and 9
find() aggregate()
• optimizer hints
• $lookup()

45
Query 8: automatic query rewrite
> db.zips.aggregate(
... [
... { "$sort" :
... { "state" : 1 }
... },
... {
... "$match" :
... { "state" : { "$gt" : "M" } }
... }
... ],
... { "explain" : true } )

46
Query 8: Explain plan
"stages" : [
"$cursor" : {
"sort" : { "state" : 1 },
"winningPlan" : {
"stage" : "FETCH",
"inputStage" : {
"stage" : "IXSCAN",
"indexName" : "state_1_city_1",
"indexBounds" : {
"state" : [ "("M", {})" ],
"city" : [ "[MinKey, MaxKey]" ]

47
Query 9: Optimizer hints
> db.zips.find( { "city" : "EAST TROY" }).hint(
{ "zip" : 1} ).explain("executionStats")
"winningPlan" : {
"stage" : "FETCH",
"filter" : { "city" : { "$eq" : "EAST TROY" } },
"inputStage" : {
"stage" : "IXSCAN", "keyPattern" : { "zip" : 1 },
"indexBounds" : { "zip" : [ "[MinKey, MaxKey]" ] }
"nReturned" : 1,

Webinar: Index Tuning and Evaluation

More Related Content

What's hot (20)

Similar to Webinar: Index Tuning and Evaluation (20)

More from MongoDB (20)

Recently uploaded (20)

Webinar: Index Tuning and Evaluation