PostgreSQL - Decoding Partitions

© Copyright EnterpriseDB Corporation, 2019. All rights reserved.
February 14, 2019
PostgreSQL:
Decoding Partition
Beena Emerson
1

© Copyright EnterpriseDB Corporation, 2019. All rights reserved.© Copyright EnterpriseDB Corporation, 2019. All rights reserved.
INTRODUCTION
● What is Partitioning?
● Why partition?
● When to Partition?
2

What is Partitioning?
Subdividing a database table into smaller parts.
PARENT TABLE
APPLICATION
INTERNAL
PARTITIONS
3

Why Partition?
10 mn
rows
2
mn
A = 100?
Faster Data Access
Scanning smaller tables takes less time.
4

Why Partition?
TABLE SIZE
PERFORMANCEScalability
With partition, we would be dealing with multiple tables of smaller sizes.
5

Why Partition?
Easy Maintenance
5000
ms
200
ms
180
ms
Commands are faster and can be run in parallel
6

When to Partition?
Table Size
● Rule of thumb: size of the table exceeds physical
memory of the database server.
● Extensive DML operations
7

When to Partition?
Targeted Columns
Col 1 Col 2 Col 3 Col4
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
75
%
Good partition key candidate
WHERE col3 = xx
8

When to Partition?
Hot and Cold Data
FEB
2019
JAN
2019
DEC
2018
NOV
2018
FASTER DISK
9

CAUTION!
Bad partitioning is worse than no partitioning!
Choose partitioning keys, strategy and no of partitions
after thorough analysis and testing.
10

PARTITIONING IN POSTGRESQL
● Inheritance (Trigger- based) Partition
● Declarative Partitioning
11

● Since PostgreSQL 8.1
● Table inheritance simulating partitioning
A = 3
PARENT TABLE TRIGGER
CHECK
CHILD TABLES
Inheritance (Trigger- based) Partition
12

Inheritance (Trigger- based) Partition
Problems
● Partitions tracking is manual and error-prone.
● Slower Performance
● Constraints may not be mutually exclusive
13

Declarative Partitioning
● Introduced in PostgreSQL 10
● Somewhat automates creation of partitions
● Easy partition maintenance
○ No overlapping boundaries
○ Add/Delete partition easily
○ Simpler syntax
● Scope for better performance
14

Terms
● Partitioned table - parent table
● Partition key - column on which table is
partitioned
● Partitions - the smaller child tables
● Partition bounds - constraints of each partition
15

● Create Partitioned Table
CREATE TABLE parent ( <col list > )
PARTITION BY <partition strategy>
( <partition key> );
● Create Partitions
CREATE TABLE child
PARTITION OF parent <partition bounds>;
… repeat for required partitions
16

● Existing table can be added as a partition to a
partitioned table
ALTER TABLE parent ATTACH PARTITION p1 <bounds>;
All entries in the table will be checked.
● A partition can be removed and made standalone
ALTER TABLE parent DETACH PARTITION child;
17

PARTITION STRATEGY AND TYPE
● Strategy
○ List
○ Range
○ Hash
● Types
○ Multi-column partitioning
○ Multi-level partitioning
18

List Strategy
MUMBAI,
PUNE
GOA
CHENNAI,
OOTY
BANGLORE
● PostgreSQL 10
● Explicitly specify key values
○ Single or multiple value per
partition
19

List Strategy
PARTITION BY LIST ( <partition key> );
CREATE TABLE child PARTITION OF parent
FOR VALUES
IN ( <val1> [, <val2>] ) ;
20

Range Strategy
● PostgreSQL 10
● Range boundaries
○ Lower inclusive (>=)
○ Upper exclusive (<)
● Unbounded values
○ MINVALUE
○ MAXVALUE
1 - 100
100 - 200
200 - 300
300 - 400
21

Range Strategy
PARTITION BY RANGE ( <partition key> );
FOR VALUES
FROM ( <lower> ) TO ( <upper> ) ;
22

Hash Strategy
● PostgreSQL 11
● Data does not have natural boundaries
● Specify modulus and remainder
○ Modulus - non-zero positive integer
○ Remainder - non-negative integer
○ remainder < modulus
● Rows spread on hash value of the partition key
23

Hash Strategy
Div
m
ParentA = 3
Hash
fn
24

Hash Strategy
PARTITION BY HASH ( <partition key> );
FOR VALUES
WITH ( MODULUS <m>, REMAINDER <r> );
25

Multi-Column Partitioning
● Multiple columns as partition key
● Supported for range and hash
● Column limit : 32
26

Range Partition
CREATE TABLE rparent (col1 int, col2 int)
PARTITION BY RANGE (col1, col2);
CREATE TABLE rpart_1 PARTITION OF rparent
FOR VALUES FROM (0, 0) TO (100, 50);
CREATE TABLE rpart_2 PARTITION OF rparent
FOR VALUES FROM (100, 50)
TO (MAXVALUE, MAXVALUE);
27

Range Partition
● Every column following MAXVALUE / MINVALUE must
also be the same.
● The row comparison operator is used for insert
○ Elements are compared left-to-right, stopping at
first unequal pair of elements.
Consider partition (0, 0) TO (100, 50)
(0, 199), (100, 49) fits while
(100, 50) does not.
28

Hash Partition
CREATE TABLE hparent (col1 int, col2 int)
PARTITION BY HASH (col1, col2);
CREATE TABLE hpart_2 PARTITION OF hparent FOR
VALUES WITH (MODULUS 3, REMAINDER 2);
29

Hash Partition
● Only a single modulus, remainder pair is used.
● The hash of each of partition key is calculated and
then combined to get a single hash value which is
divided by the modulus specified.
30

Multi-level Partitioning
31

Multi-level Partitioning
● Create a partitioned partition
CREATE TABLE child1 PARTITION OF parent
FOR VALUES FROM (0) TO (100)
PARTITION BY LIST (col3);
● Attach a partitioned table as a partition.
● There is no check to ensure that the sub partition
bounds are a subset of the partition’s own bound.
CREATE TABLE child1_1 PARTITION OF child1
FOR VALUES IN (500, 2000);
32

OTHER FEATURES
● Default Partition
● Handling NULL values
● Inherit Constraints
● Update tuple routing
● Foreign Table Partitions
● Partition Pruning
33

● PostgreSQL 11
● Catch tuples that do not match
partition bounds of others.
● Support for: list, range
● Syntax:
CREATE TABLE child
PARTITION OF parent
DEFAULT
A
B
C
DEFAULT
Default Partition
Z
(NOT (col1 IS NOT NULL) AND
(col1 = ANY (ARRAY[ 'A', 'B', 'C' ])))
34

Default Partition
Add new partition
Default partition should not
have rows that satisfy the
new partition.
ERROR: updated partition constraint for default
partition "part_def" would be violated by some row
1 - 100
100 - 200
DEF: 205, 410
200 - 300
35

Handling NULL values
● Hash routes based on the hash.
● Range: can be inserted in default partition.
● List: A partition can be made to accept NULL else
it is routed to default partition.
CREATE TABLE child PARTITION OF lparent
FOR VALUES
IN ( NULL ) ;
36

Inherit Constraints
Constraint Types
● Check
● Not-Null
● Unique
● Primary Keys
● Foreign Keys
● Exclude
37

Inherit Constraints
PostgreSQL 10
● Supported: Check, Not Null
● Other constraints can be explicitly added to
individual partition.
● No cascading inheritance
38

Inherit Constraints
PostgreSQL 11
● Added Support: unique, primary and foreign keys
● Not supported: exclude constraints
● Cascading: New constraints added to parent
propagates to partitions.
● Other constraints can be explicitly added to
individual partition.
39

Update Tuple Routing
PostgreSQL 11
● When an updated tuple no longer fits the partition,
route it to the correct one.
1 - 100
100 - 200
200 - 300
300 - 400
150
380
40

Foreign Table Partitions
● Can add foreign table as a partition.
CREATE FOREIGN TABLE lpart_f1 PARTITION OF
lparent FOR VALUES IN ('C' , 'E') SERVER s0
● Tuples can be inserted from v11.
● Cannot be partitioned.
● Updated tuples can me moved to foreign
partitions but not from them.
41

Partition Pruning
0 - 100
100 - 200
200 - 300
300 - 400
180 - 220
Planner Level Pruning
Depending on the query conditions
on partition key, scan only the
relevant partitions.
col1
42

Partition Pruning
Planner Level Pruning
=# EXPLAIN (COSTS OFF, TIMING OFF, ANALYZE) SELECT *
FROM rparent WHERE col1 BETWEEN 180 AND 220;
Append (actual rows=41 loops=1)
-> Seq Scan on rpart_2 (actual rows=20 loops=1)
Filter: ((col1 >= 180) AND (col1 <= 220))
Rows Removed by Filter: 80
-> Seq Scan on rpart_3 (actual rows=21 loops=1)
Filter: ((col1 >= 180) AND (col1 <= 220))
Rows Removed by Filter: 79
43

Partition Pruning
Runtime Partition Pruning
● PostgreSQL 11
● Performed at two levels
○ Executor Initialization - prepared query
○ Actual Execution - nested loop join
44

Partition Pruning
Runtime Pruning - Executor Initialization
=# PREPARE r_q1 (int) as SELECT * from rparent where
col1 < $1;
=# EXPLAIN execute r_q1(150);
Append (cost=0.00..168.06 rows=3012 width=8)
Subplans Removed: 2
-> Seq Scan on rpart_1
Filter: (col1 < $1)
-> Seq Scan on rpart_2
Filter: (col1 < $1)
(6 rows)
45

Partition Pruning
Runtime Pruning - Actual execution
Nested Loop Join
100 - 200
0 - 100
200 - 300
300 - 400
10000 rows
1-50
Rest > 500
(50 times)
46

Partition Pruning
Runtime Pruning - Actual execution
=#EXPLAIN SELECT * FROM tbl JOIN rparent ON id= col1;
Nested Loop (actual rows=50 loops=1)
-> Seq Scan on tbl (actual rows=10000 loops=1)
-> Append (actual rows=0 loops=10000)
-> Index Scan using rpart_1_pkey on rpart_1 ( actual rows=1
loops=50)
Index Cond: (col1 = tbl.id)
-> Index Scan using rpart_2_pkey on rpart_2 (never executed)
47

THANK YOU!
48

PostgreSQL - Decoding Partitions

Recommended

More Related Content

Similar to PostgreSQL - Decoding Partitions (20)

Recently uploaded (20)

PostgreSQL - Decoding Partitions