PostgreSQL, performance for queries with grouping

1/44
PostgreSQL
Optimisation of queries with grouping
Alexey Bashtanov, Brandwatch
28 Jan 2016

2/44
What is it all about?
This talk will cover optimisation of
Grouping
Aggregation
Unfortunately it will not cover optimisation of
Getting the data
Filtering
Joins
Window functions
Other data transformations

3/44
Outline
1 What is a grouping?
2 How does it work?
Aggregation functions under the hood
Grouping algorithms
3 Optimisation
Avoiding sorts
Summation
Denormalized data aggregation
Arg-maximum
4 Still slow?

5/44
What is a grouping?
What do we call a grouping/aggregation operation?
An operation of splitting input data into several classes and
then compilation each class into one row.
3
32 21 1
3
3
3
3
1
1
2
2 15
2
2
2
2
3
3
1
1 8
3
3
2
2
3
3
1
1 9

6/44
Examples
SELECT department_id,
avg(salary)
FROM employees
GROUP BY department_id
SELECT DISTINCT department_id
FROM employees

7/44
Examples
SELECT DISTINCT ON (department_id)
department_id,
employee_id,
salary
FROM employees
ORDER BY department_id,
salary DESC

8/44
Examples
SELECT max(salary)
FROM employees
SELECT salary
FROM employees
ORDER BY salary DESC
LIMIT 1

10/44
INITCOND SFUNC
Input data
state SFUNC
Input data
state SFUNC
Input data
state
FINALFUNC
Result
An aggregate function is deﬁned by:
State, input and output types
Initial state (INITCOND)
Transition function (SFUNC)
Final function (FINALFUNC)

10/44
state = 0 state += input
2
2 state += input
3
5 state += input
7
12
=
sum=12
SELECT sum(column1),
avg(column1)
FROM (VALUES (2), (3), (7)) _

10/44
cnt = 0
sum = 0
cnt++
sum+=input
2
cnt=1
sum=2
cnt++
sum+=input
3
cnt=2
sum=5
cnt++
sum+=input
7
cnt=3
sum=12
sum / cnt
avg=4
SELECT sum(column1),
avg(column1)
FROM (VALUES (2), (3), (7)) _

11/44
SFUNC and FINALFUNC functions can be written in
C — fast (SFUNC may modify input state and return it)
SQL
PL/pgSQL — SLOW!
any other language
SFUNC and FINALFUNC functions can be declared STRICT
(i.e. not called on null input)

12/44
Grouping algorithms
PostgreSQL uses 2 algorithms to feed aggregate functions by
grouped data:
GroupAggregate: get the data sorted and apply
aggregation function to groups one by one
HashAggregate: store state for each key in a hash table

13/44
GroupAgg
1 3 1 2 2 3 1 3 2 1 state: 0

13/44
GroupAgg
1 3 1 2 2 3 1 3 2 1 state: 0
1 3 1 2 2 3 1 3 state: 3

13/44
GroupAgg
1 3 1 2 2 3 1 3 2 1 state: 0
1 3 1 2 2 3 1 3 state: 3
1 3 1 2 2 state: 4 6

13/44
GroupAgg
1 3 1 2 2 3 1 3 2 1 state: 0
1 3 1 2 2 3 1 3 state: 3
1 3 1 2 2 state: 4 6
1 3 1 state: 0 8 6

13/44
GroupAgg
1 3 1 2 2 3 1 3 2 1 state: 0
1 3 1 2 2 3 1 3 state: 3
1 3 1 2 2 state: 4 6
1 3 1 state: 0 8 6
5 8 6

14/44
HashAggregate
1 2 3 2 3 1 2 1 3 1 state: 0

14/44
HashAggregate
1 2 3 2 3 1 2 1 3 1 state: 0
1 2 3 2 3 1 2 1 3 state: 1

14/44
HashAggregate
1 2 3 2 3 1 2 1 3 1 state: 0
1 2 3 2 3 1 2 1 3 state: 1
1 2 3 2 3 1 2 1
state: 1
state: 3

14/44
HashAggregate
1 2 3 2 3 1 2 1 3 1 state: 0
1 2 3 2 3 1 2 1 3 state: 1
1 2 3 2 3 1 2 1
state: 1
state: 3
1 2 3
state: 6
state: 6
state: 1

14/44
HashAggregate
1 2 3 2 3 1 2 1 3 1 state: 0
1 2 3 2 3 1 2 1 3 state: 1
1 2 3 2 3 1 2 1
state: 1
state: 3
1 2 3
state: 6
state: 6
state: 1
state: 6
state: 8
state: 5

14/44
HashAggregate
1 2 3 2 3 1 2 1 3 1 state: 0
1 2 3 2 3 1 2 1 3 state: 1
1 2 3 2 3 1 2 1
state: 1
state: 3
1 2 3
state: 6
state: 6
state: 1
state: 6
state: 8
state: 5
68 5

15/44
GroupAggregate vs. HashAggregate
GroupAggregate
− Requires sorted data
+ Needs less memory
+ Returns sorted data
+ Returns data on the ﬂy
+ Can perform
count(distinct ...),
array_agg(... order by ...)
etc.
HashAggregate
+ Accepts unsorted data
− Needs more memory
− Returns unsorted data
− Returns data at the end
− Can perform only basic
aggregation

17/44
Avoiding sorts
Sorts are really slow. Prefer HashAggregation if possible.

17/44
Avoiding sorts
What to do if you get something like this?
EXPLAIN
SELECT region_id,
avg(age)
FROM people
GROUP BY region_id
GroupAggregate (cost=149244.84..156869.46 rows=9969 width=10)
-> Sort (cost=149244.84..151744.84 rows=1000000 width=10)
Sort Key: region_id
-> Seq Scan on people (cost=0.00..15406.00 rows=1000000 width=10)
1504.474 ms

17/44
Avoiding sorts
EXPLAIN
SELECT region_id,
avg(age)
FROM people
GROUP BY region_id
set enable_sort to off?

17/44
Avoiding sorts
EXPLAIN
SELECT region_id,
avg(age)
FROM people
GROUP BY region_id
set enable_sort to off? No!
GroupAggregate (cost=10000149244.84..10000156869.46 rows=9969 width=10)
-> Sort (cost=10000149244.84..10000151744.84 rows=1000000 width=10)
Sort Key: region_id
1497.167 ms

17/44
Avoiding sorts
EXPLAIN
SELECT region_id,
avg(age)
FROM people
GROUP BY region_id
Increase work_mem: set work_mem to ’100MB’
HashAggregate (cost=20406.00..20530.61 rows=9969 width=10)
685.689 ms

17/44
Avoiding sorts
EXPLAIN
SELECT region_id,
avg(age)
FROM people
GROUP BY region_id
Increase work_mem: set work_mem to ’100MB’
HashAggregate (cost=20406.00..20530.61 rows=9969 width=10)
685.689 ms
Increase sanely to avoid OOM

18/44
Avoiding sorts
How to spend less memory to allow HashAggregation?
Don’t aggregate joined
SELECT p.region_id,
d.region_description,
avg(age)
FROM people p
JOIN regions r using (region_id)
GROUP BY region_id,
region_description
Join aggregated instead
SELECT a.region_id,
r.region_description,
a.avg_age
FROM (
SELECT region_id,
avg(age) avg_age
FROM people p
GROUP BY region_id
) a
JOIN regions r using (region_id)

19/44
Avoiding sorts
How to avoid sorts for count(DISTINCT ...)?
SELECT date_trunc(’month’, visit_date),
count(DISTINCT visitor_id)
FROM visits
GROUP BY date_trunc(’month’, visit_date)
GroupAggregate (actual time=7685.972..10564.358 rows=329 loops=1)
-> Sort (actual time=7680.426..9423.331 rows=4999067 loops=1)
Sort Key: (date_trunc(’month’::text, visit_date))
Sort Method: external merge Disk: 107496kB
-> Seq Scan on visits (actual time=10.941..2966.460 rows=4999067 loops=1)

20/44
Avoiding sorts
Two levels of HashAggregate could be faster!
SELECT visit_month,
count(*)
FROM (
SELECT DISTINCT
date_trunc(’month’, visit_date)
as visit_month,
visitor_id
FROM visits
) _
GROUP BY visit_month
HashAggregate (actual time=2632.322..2632.354 rows=329 loops=1)
-> HashAggregate (actual time=2496.010..2578.779 rows=329000 loops=1)

21/44
Avoiding sorts
How to avoid sorts for array_agg(...ORDER BY ...)?
SELECT
visit_date,
array_agg(visitor_id ORDER BY visitor_id)
FROM visits
GROUP BY visit_date
GroupAggregate (actual time=5433.658..8010.309 rows=10000 loops=1)
-> Sort (actual time=5433.416..6769.872 rows=4999067 loops=1)
Sort Key: visit_date
Sort Method: external merge Disk: 107504kB

22/44
Avoiding sorts
Might be better to sort each line separately
SELECT
visit_date,
(
select array_agg(i ORDER BY i)
from unnest(visitors_u) i
)
FROM (
SELECT visit_date,
array_agg(visitor_id) visitors_u
FROM visits
GROUP BY visit_date
) _
Subquery Scan on _ (actual time=2504.915..3767.300 rows=10000 loops=1)
-> HashAggregate (actual time=2504.757..2555.038 rows=10000 loops=1)
SubPlan 1
-> Aggregate (actual time=0.120..0.121 rows=1 loops=10000)
-> Function Scan on unnest i (actual time=0.033..0.055 rows=500 loops=10000)

23/44
Summation
There are three sum functions in PostgreSQL:
sum(int) returns bigint
sum(bigint) returns numeric — SLOW
(needs to convert every input value)
sum(numeric) returns numeric
Do not use bigint as a datatype for a value to be summed,
prefer numeric. BTW small numeric numbers spend less
space bytes on disk than bigint.
It might be worth writing a custom aggregate function
sum(bigint) returns bigint . . .

24/44
Summation
Straightforward solution, to be used if there are few zero values:
SELECT sum(cat_cnt)
FROM cities
Can speed up up to 7 times. Worth considering if >50% zeroes:
SELECT coalesce(sum(tiger_cnt), 0)
FROM cities
WHERE tiger_cnt <> 0
Can help only if the type is numeric and we cannot ﬁlter out:
SELECT coalesce(sum(nullif(tiger_cnt, 0)), 0),
sum(cat_cnt)
FROM cities

25/44
Summation
Better in any case to replace all zeroes by nulls:
UPDATE cities
SET cat_cnt = nullif(cat_cnt, 0),
tiger_cnt = nullif(tiger_cnt, 0);
VACUUM FULL cities;
Additionally this will dramatically reduce space occupied.

26/44
Sometimes we need to aggregate denormalized data
Most common solution is
SELECT account_id,
account_name,
sum(payment_amount)
FROM payments
GROUP BY account_id,
account_name
Planner does not know that account_id and account_name
correlate. It can lead to wrong estimates and suboptimal plan.

27/44
A bit less-known approach is
SELECT account_id,
min(account_name),
sum(payment_amount)
FROM payments
GROUP BY account_id
Works only if the type of "denormalized payload" supports
comparison operator.

28/44
Also we can write a custom aggregate function
CREATE FUNCTION frst (text, text)
RETURNS text IMMUTABLE LANGUAGE sql AS
$$ select $1; $$;
CREATE AGGREGATE a (text) (
SFUNC=frst,
STYPE=text
);
SELECT account_id,
a(account_name),
sum(payment_amount)
FROM payments
GROUP BY account_id

29/44
Or even write it in C
SELECT account_id,
anyold(account_name),
sum(payment_amount)
FROM payments
GROUP BY account_id
Sorry, no source code for anyold

30/44
And what is the fastest?
It depends on the width of "denormalized payload":
1 10 100 1000 10000
dumb 366ms 374ms 459ms 1238ms 53236ms
min 375ms 377ms 409ms 716ms 16747ms
SQL 1970ms 1975ms 2031ms 2446ms 2036ms
C 385ms 385ms 408ms 659ms 436ms

30/44
And what is the fastest?
It depends on the width of "denormalized payload":
1 10 100 1000 10000
dumb 366ms 374ms 459ms 1238ms 53236ms
min 375ms 377ms 409ms 716ms 16747ms
SQL 1970ms 1975ms 2031ms 2446ms 2036ms*
C 385ms 385ms 408ms 659ms 436ms*
* — The more data the faster we proceed?
It is because we do not need to extract TOASTed values.

31/44
Arg-maximum
Max
Population of the largest
city in each country
Date of last tweet by each
author
The highest salary in each
department

31/44
Arg-maximum
Max
Population of the largest
city in each country
Date of last tweet by each
author
The highest salary in each
department
Arg-max
What is the largest city in
each country
What is the last tweet by
each author
Who gets the highest
salary in each department

32/44
Arg-maximum
Max is built-in. How to perform Arg-max?
Self-joins?
Window-functions?

32/44
Arg-maximum
Self-joins?
Window-functions?
Use DISTINCT ON() (PG-speciﬁc, not in SQL standard)
SELECT DISTINCT ON (author_id)
author_id,
twit_id
FROM twits
ORDER BY author_id,
twit_date DESC

32/44
Arg-maximum
Self-joins?
Window-functions?
Use DISTINCT ON() (PG-speciﬁc, not in SQL standard)
SELECT DISTINCT ON (author_id)
author_id,
twit_id
FROM twits
ORDER BY author_id,
twit_date DESC
But it still can be performed only by sorting, not by hashing :(

33/44
Arg-maximum
We can emulate Arg-max by ordinary max and dirty hacks
SELECT author_id,
(max(array[
twit_date,
date’epoch’ + twit_id
]))[2] - date’epoch’
FROM twits
GROUP BY author_id;
But such types tweaking is not always possible.

34/44
Arg-maximum
It’s time to write more custom aggregate functions
CREATE TYPE amax_ty AS (key_date date, payload int);
CREATE FUNCTION amax_t (p_state amax_ty, p_key_date date, p_payload int)
RETURNS amax_ty IMMUTABLE LANGUAGE sql AS
$$
SELECT CASE WHEN p_state.key_date < p_key_date
OR (p_key_date IS NOT NULL AND p_state.key_date IS NULL)
THEN (p_key_date, p_payload)::amax_ty
ELSE p_state END
$$;
CREATE FUNCTION amax_f (p_state amax_ty) RETURNS int IMMUTABLE LANGUAGE sql AS
$$ SELECT p_state.payload $$;
CREATE AGGREGATE amax (date, int) (
SFUNC = amax_t,
STYPE = amax_ty,
FINALFUNC = amax_f,
INITCOND = ’(,)’
);
SELECT author_id,
amax(twit_date, twit_id)
FROM twits
GROUP BY author_id;

35/44
Arg-maximum
Argmax is similar to amax, but written in C
SELECT author_id,
argmax(twit_date, twit_id)
FROM twits
GROUP BY author_id;

36/44
Arg-maximum
Who wins now?
1002 3332 10002 33332 50002
DISTINCT ON 6ms 42ms 342ms 10555ms 30421ms
Max(array) 5ms 47ms 399ms 4464ms 10025ms
SQL amax 38ms 393ms 3541ms 39539ms 90164ms
C argmax 5ms 37ms 288ms 3183ms 7176ms

36/44
Arg-maximum
Who wins now?
1002 3332 10002 33332 50002
SQL amax ﬁnally outperforms DISTINCT ON on 109-ish rows

38/44
Still slow?
Slow max, arg-max or distinct query?
Sometimes we can fetch the rows one-by-one using index:
3 2 1 4 2 2 1 3 31 0
CREATE INDEX ON twits(author_id, twit_date DESC);
-- for the very first author_id fetch the row with latest date
SELECT twit_id,
twit_date,
author_id
FROM twits
ORDER BY author_id,
twit_date DESC
LIMIT 1;
-- find the next author_id and fetch the row with latest date
SELECT twit_id,
twit_date,
author_id
FROM twits
WHERE author_id > ?
ORDER BY author_id,
twit_date DESC
LIMIT 1;
...

38/44
Still slow?
Slow max, arg-max or distinct query?
Sometimes we can fetch the rows one-by-one using index:
3 2 1 4 2 2 1 3 31 0
CREATE INDEX ON twits(author_id, twit_date DESC);
CREATE FUNCTION f1by1() RETURNS TABLE (o_twit_id int, o_twit_date date) AS $$
DECLARE l_author_id int := -1; -- to make the code a bit more simple
BEGIN
LOOP
SELECT twit_id,
twit_date,
author_id
INTO o_twit_id,
o_twit_date,
l_author_id
FROM twits
WHERE author_id > l_author_id
ORDER BY author_id,
twit_date DESC
LIMIT 1;
EXIT WHEN NOT FOUND;
RETURN NEXT;
END LOOP;
END;
$$ LANGUAGE plpgsql;
SELECT * FROM f1by1();

39/44
Still slow?
Let us use pure SQL instead, it is a bit faster as usual
WITH RECURSIVE d AS (
(
SELECT array[author_id, twit_id] ids
FROM twits
ORDER BY author_id,
twit_date DESC
LIMIT 1
)
UNION
SELECT (
SELECT array[t.author_id, t.twit_id]
FROM twits t
WHERE t.author_id > d.ids[1]
ORDER BY t.author_id,
t.twit_date DESC
LIMIT 1
) q
FROM d
)
SELECT d.ids[1] author_id,
d.ids[2] twit_id
FROM d;

40/44
Still slow?
One-by-one retrieval by index
+ Incredibly fast unless returns too many rows
− Needs an index
− SQL version needs tricks if the data types differ
Authors × Twits-per-author:
106 × 101 105 × 102 104 × 103 102 × 105
C argmax 3679ms 3081ms 2881ms 2859ms
1-by-1 proc 12750ms 1445ms 152ms 2ms
1-by-1 SQL 6250ms 906ms 137ms 2ms

40/44
Still slow?
One-by-one retrieval by index
+ Incredibly fast unless returns too many rows
− Needs an index
− SQL version needs tricks if the data types differ
1002 3332 10002 33332 50002
1-by-1 proc 2ms 6ms 12ms 42ms 63ms
1-by-1 SQL 1ms 4ms 11ms 29ms 37ms

41/44
Still slow?
Slow HashAggregate?
Use parallel aggregation extension:
http://www.cybertec.at/en/products/
agg-parallel-aggregations-postgresql/
+ Up to 30 times faster
+ Speeds up SeqScan as well
− Mostly useful for complex row operations
− Requires PG 9.5+
− No magic: it loads up several of your cores

42/44
Still slow?
Slow count(DISTINCT ...)?
Use HyperLogLog: reliable and efﬁcient approximate algorithm
https://en.wikipedia.org/wiki/HyperLogLog
https://github.com/aggregateknowledge/postgresql-hll
Or fetch approximate values from pg_stats

43/44
Still slow?
Slow in typing? ;)
SELECT department_id,
avg(salary)
FROM employees
GROUP BY 1 -- same as GROUP BY department_id
SELECT count(*)
FROM employees
GROUP BY true -- same as HAVING count(*) > 0
-- or use MySQL
SELECT account_id,
account_name,
sum(payment_amount)
FROM payments
GROUP BY 1

PostgreSQL, performance for queries with grouping

Recommended

More Related Content

What's hot (20)

Viewers also liked (11)

Similar to PostgreSQL, performance for queries with grouping (20)

Recently uploaded (20)

PostgreSQL, performance for queries with grouping