>whoami • Cedric Chin (aka Eli James) - @ejames_c • Project Manager, Floating Cube Studios (D7) • Programmer, not a database expert • But a heavy database user • Most of us are heavy database users
This Talk • You will walk away with a basic understanding of database indexes. • In general! Not specific to a particular DB • A basic understanding is usually good enough • Developers need to know how to make queries fast • The rest of the DB can be a black box.
Indexes Make Queries Fast • Indexes are conceptually similar. • But unlike a dictionary, databases are constantly updated • This means that indexes have to be constantly updated.
Indexes Make Queries Fast • In a RDBMs, indexes are like a map or guide that allows you to quickly find the data that you’re looking for. • But let’s start from the basics. Pretend we don’t care or know about how DBs work.
Creating an Index • You can create an index on any field. • Say you have a new, large table called clients: pk name age gender 1 Binh Dang 23 Male 2 Connor Tan 34 Male … … … … • All tables have an index on create: the primary key index
Two Data Structures • All database indexes consist of two data structures: • A doubly linked list • A balanced search tree • Note: this is just basic stuff! Real implementation has modifications.
Table Data • Database data is not organised sequentially on disk like books in a library. • (Or pages in a dictionary.) • Instead it is stored in blocks all across the disk.
Table Data • Table data is stored in a table block, in a heap structure • No relationships between table blocks or rows • Is not sorted. col1 col2 col3 Binh 23 A Connor 45 X Vivian 12 X Bob 98 A Visual example of table block
Ordered Table Data col1 col2 col3 Binh 67 A Connor 45 X Vivian 12 X Bob 98 A col1 col2 col3 Sean 23 A Loki 18 X Thor 33 X Trung 29 A col2 rowID 12 3A 2F 13 2F AE 18 2C 50 23 5B 78 col2 rowID 26 65 2F 29 2F 0E 33 3D A0 33 5B F9 Index Leaf Nodes
Ordered Table Data col1 col2 col3 Binh 67 A Connor 45 X Vivian 12 X Bob 98 A col1 col2 col3 Sean 23 A Loki 18 X Thor 33 X Trung 29 A col2 rowID 12 3A 2F 13 2F AE 18 2C 50 23 5B 78 col2 rowID 26 65 2F 29 2F 0E 33 3D A0 33 5B F9 Index Leaf Nodes
Ordered Table Data col1 col2 col3 Binh 67 A Connor 45 X Vivian 12 X Bob 98 A col1 col2 col3 Sean 23 A Loki 18 X Thor 33 X Trung 29 A Index Leaf Nodes col2 rowID 12 3A 2F 13 2F AE 18 2C 50 23 5B 78 col2 rowID 26 65 2F 29 2F 0E 33 3D A0 33 5B F9
Ordered Table Data • A database uses a sorted doubly linked list to keep track of order • Doubly linked list means that the DB can traverse back and forth. • O(n) transversal
Balanced Search Tree • The index leaf nodes are connected by a balanced search tree • The b-tree has equal depth at every point • Searching the b-tree is O(logn) • B-tree growth is also O(logn)
Ordered Table Data col1 col2 col3 Binh 67 A Connor 45 X Vivian 12 X Bob 98 A col1 col2 col3 Sean 23 A Loki 18 X Thor 54 X Trung 29 A col2 rowID 12 3A 2F 13 2F AE 18 2C 50 23 5B 78 col2 rowID 26 65 2F 26 2F 0E … .. .. 54 5B F9 Index Leaf Nodes
Database Query • There are 3 steps to a database query 1. Tree traversal - O(logn), fast 2. Leaf node chain traversal - O(n), slow 3. Table data retrieval - not stored physically in same location, slow
Database Query • There are 3 steps to a database query 1. Tree traversal - O(logn), fast 2. Leaf node chain traversal - O(n), slow 3. Table data retrieval - not stored physically in same location, slow
Database Queries • Assuming we have indexed `age` • Tree-traversal (fast) • Table data retrieve (1 row only, fast). SELECT * FROM clients WHERE age = 34
Database Queries • Assuming we have indexed `age` • Tree-traversal to get age = 20 (fast) • Leaf node chain traversal = ? (if many, slow) • Table data retrieve (if thousands of rows, slow). SELECT * FROM clients WHERE age >= 20 AND age <= 50
The Full Table Scan • The DB simply returns every single block in the table. • Can sometimes be more efficient, if you’re returning a large % of the data. • Why? DB executes multi-block reads, optimises for fewer read operations compared to index scan SELECT * FROM clients
EXPLAIN • How do you actually know how the DB is executing your queries? • Use the EXPLAIN statement. • Just add in front of the query; all RDBMs have some version of this. EXPLAIN SELECT * FROM clients WHERE age = 34
EXPLAIN select_type table type possible_keys key rows extra SIMPLE clients range age age 1342561 Using where MySQL example; different for others EXPLAIN SELECT * FROM clients WHERE age >= 20 AND age <= 50
EXPLAIN select_type table type possible_keys key rows extra SIMPLE clients range age age 1341 Using where MySQL example; different for others EXPLAIN SELECT * FROM clients WHERE age >= 20 AND age <= 50
MySQL types • Some of the more important types: • eq_ref - tree traversal only, unique index • ref/range - tree traversal, then leaf node traversal • index - the entire index is scanned (leaf node traversal) • full - full table scan, everything is read
EXPLAIN EXPLAIN SELECT * FROM clients WHERE age = 34 select_type table type possible_keys key rows extra SIMPLE clients ref age age 1 MySQL example; different for others
EXPLAIN EXPLAIN SELECT * FROM clients WHERE id = 2 select_type table type possible_keys key rows extra SIMPLE clients const PRIMARY PRIMARY 1 MySQL example; different for others
Database Query 1. Tree traversal - O(logn), fast 2. Full leaf node chain traversal - O(n), slow 3. Range leaf node chain traversal - O(k), ok la 4. Full table scan Takeaway: understand your RDBM’s equivalent to:
INSERT • Adding an index means INSERT operations now have more work to do. • Find a table block to store the new data • Update the index (e.g. balance the tree) for each index on the table! SQL Performance Explained, Page 160
INSERT • In practice, not that bad. • Speed is affected by size of table and number of indexes • MySQL documentation: size of table slows down index insert by log(N) • Point: don’t add redundant or unnecessary indexes.
DELETE • DELETE benefits from the WHERE clause. • Is like a SELECT, but with the extra step of deleting row and rebalancing the index, for each index on the table
Concatenated Indexes • A concatenated index is an index over multiple columns • Consider the following table: • We query first_name and last_name a lot pk first_name last_name age 1 Binh Nguyen 23 2 Connor Tan 34 … … …
Concatenated Indexes • We want to index the first_name and last_name • Are the following two indexes the same? ALTER TABLE `clients` ADD INDEX (`first_name`, `last_name`); ALTER TABLE `clients` ADD INDEX (`last_name`, `first_name`);
Concatenated Indexes • This query will benefit from the index: • SELECT * FROM clients WHERE first_name = `Binh` AND last_name = `Nguyen` • This query will not benefit from the index: • SELECT * FROM clients WHERE last_name = `Nguyen` ALTER TABLE `clients` ADD INDEX (`first_name`, `last_name`);
Concatenated Indexes first_name last_name Binh Dang Binh Nguyen Binh Pham Connor Chan Connor Tan last_name first_name Nguyen Huy Nguyen Trung Nguyen Tuan Tan Jonathan Tan Bob (`first_name`, `last_name`) (`last_name`, `first_name`)
Concatenated Indexes first_name last_name Binh Nguyen Binh Dang Binh Pham Connor Chan Connor Tan last_name first_name Nguyen Huy Nguyen Trung Nguyen Tuan Tan Jonathan Tan Bob (`first_name`, `last_name`) (` SELECT * FROM clients WHERE first_name = `Binh` AND last_name = `Nguyen`
Concatenated Indexes first_name last_name Binh Dang Binh Nguyen Binh Pham Connor Chan Connor Tan last_name first_name Nguyen Huy Nguyen Trung Nguyen Tuan Tan Jonathan Tan Bob (`first_name`, `last_name`) (` SELECT * FROM clients WHERE last_name = `Nguyen` ??
Concatenated Indexes first_name last_name Binh Dang Binh Nguyen Binh Pham Connor Chan Connor Tan last_name first_name Nguyen Huy Nguyen Trung Nguyen Tuan Tan Jonathan Tan Bob (` (`last_name`, `first_name`) SELECT * FROM clients WHERE last_name = `Nguyen`
How is this useful? • Useful when you have associations • e.g. client has many groups • Principle: index in a way such that the left-most index is always used. pk client_id group_id 1 1 3 2 1 4 (client_id, group_id)
Functions & Indexes • Some databases have functions. • e.g.: UPPER, LOWER • If you have an index of `name` and do this: • SELECT * FROM client WHERE UPPER(name) = `HUY NGUYEN` • will it use the index? • Answer: NO!
Functions & Indexes • The index is unable to look at the result of functions • But in some DBs you can create a function-based index: • CREATE INDEX up_name ON clients (UPPER(name)) • MySQL < 5.6 does not have function-based indexing. LOL
What Did You Learn? • What is an index? • How do indexes work? • What does an indexed query consist of? • The EXPLAIN statement • Drawbacks of indexes • Concatenated indexes • Indexing database functions
More Stuff • http://use-the-index-luke.com/ • There is quite a bit more to indexes than this talk • But the basics are now covered • Enjoy speedier queries =)
sbtechnight
ryomaru0825
kthrtty
takakuni
xibuka
nomu
makamaka
superbrothers
ytsuzaki
arthur1
nozomiito
zakiwarfel
searls
kneath
mongodb
maltzj
rocio
eileencodes
smashingmag
sugarenia
jrom
samanthasiow
scottboms