This talk will first introduce the different ways PostgreSQL can use memory, from the operating system, to cluster wide and then into per session and per operation. From there we will dive into specifics around different PostgreSQL parameters like shared_buffers, work_mem, maintenance_work_mem and how to set them depending on your workload. The presentation will also cover some of the lesser known ways that PostgreSQL will consume memory, how you can diagnose what is using the memory in your PostgreSQL cluster and possible ways to avoid running out of memory. Additionally the talk we will cover the importance of hugepages for not only performance but memory usage on large memory systems.

1. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
© 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
PRACTICAL MEMORY
TUNING FOR
POSTGRESQL
Grant McAlister
Senior Principal Engineer

2. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Why you should care
0
10,000
20,000
30,000
40,000
50,000
60,000
70,000
5 105 205 305 405 505 605 705 805 905 1005 1105 1205 1305 1405 1505 1605 1705
Transactions
Per
Second
(TPS)
Seconds
sysbench read only point selects
Normal Tuning Less than Optimal Tuning
11X
buffers not in memory swap thrashing

3. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
© 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
OOM & Swap
3

4. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
What is the Out of Memory (OOM) Killer
postgres 1
postgres 2
postgres 3
postgres 4
badness_for_task = total_vm_for_task / (sqrt(cpu_time_in_seconds) * sqrt(sqrt(cpu_time_in_minutes)))

5. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
What is the Out of Memory (OOM) Killer
postgres 1
postgres 2
postgres 3
postgres 4
badness_for_task = total_vm_for_task / (sqrt(cpu_time_in_seconds) * sqrt(sqrt(cpu_time_in_minutes)))

6. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Why swap?
RAM
request 8k

7. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Why swap?
RAM
request 16k
FULL

8. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Why swap?
RAM SWAP
request 16k

9. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Why swap?
RAM SWAP
request 16k
swap out
swap in
THRASHING

10. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
© 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Overview
6

11. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Memory Overview
Operating System
shared_buffers
postgres
processes
pagecache
25%
Operating System
shared_buffers
postgres
processes
pagecache
75%

12. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Pagecache
File System Cache
Storage
shared_buffers wal archiving misc temp_buffers

13. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Pagecache
File System Cache
Storage
shared_buffers wal archiving misc temp_buffers

14. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Pagecache
File System Cache
Storage
shared_buffers wal archiving misc
shared_buffers
wal
archiving
misc
temp_buffers
temp_buffers

15. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Pagecache
File System Cache
Storage
shared_buffers wal archiving misc
shared_buffers
wal
archiving
misc
temp_buffers
temp_buffers
shared_buffers
misc temp_buffers

16. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Working Set Size - Heat
Orders_2023
Orders_2022
Orders_2021
Orders_2020
Orders_2019
Orders_2018
Orders_2017
Orders_2016
800GB
80-100GB hot working set

17. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Inventory_SEA
Inventory_LAX
Inventory_SFO
Inventory_EWR
Inventory_SYD
Inventory_YYC
Inventory_BIO
Inventory_SBP
Working Set Size - Heat
Orders_2023
Orders_2022
Orders_2021
Orders_2020
Orders_2019
Orders_2018
Orders_2017
Orders_2016
800GB
80-100GB hot working set
Inventory_SEA
Inventory_LAX
Inventory_SFO
Inventory_EWR
Inventory_SYD
Inventory_YYC
Inventory_BIO
Inventory_SBP
800GB
200-400GB hot working set

18. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Working Set Size – Indexes and Data
Random
heap
index

19. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Working Set Size – Indexes and Data
Random
heap
index

20. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Working Set Size – Indexes and Data
Random
heap
index

21. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Working Set Size – Indexes and Data
Random
heap
index

22. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Working Set Size – Indexes and Data
right leaning
heap
index

23. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Working Set Size – Indexes and Data
right leaning
heap
index

24. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Working Set Size – Indexes and Data
right leaning
heap
index

25. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Working Set Size – Indexes and Data
right leaning
heap
index

26. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
© 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Shared Buffers
12

27. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Small Cache
File System Cache
Storage
shared_buffers

28. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
shared_buffers comparison
14
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
0
100,000
200,000
300,000
400,000
500,000
600,000
700,000
800,000
900,000
25% of host memory - 80
tables
25% of host memory - 125
tables
25% of host memory - 250
tables
12.5% of host memory -
250 tables
5% of host memory - 250
tables
1% of host memory - 250
tables
Percent
of
object
type
in
memory
Transactions
Per
Second sysbench - read only point selects
TPS Index Tables
60%

29. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
shared_buffers comparison – pg_buffercache
Type
Usage
count
25% of host
memory
1% of host
memory
Index 5 855806 3630
Index 4 707472 2624
Index 3 124292 23
Index 2 22019 1754
Index 1 3799 73303
Index 0 558 71569
Table 5 316 1
Table 4 2764 3
Table 3 33572 105
Table 2 359638 180
Table 1 2933760 76505
Table 0 2606002 76303
15

30. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
shared_buffers comparison – pg_buffercache
Type
Usage
count
25% of host
memory
1% of host
memory
Index 5 855806 3630
Index 4 707472 2624
Index 3 124292 23
Index 2 22019 1754
Index 1 3799 73303
Index 0 558 71569
Table 5 316 1
Table 4 2764 3
Table 3 33572 105
Table 2 359638 180
Table 1 2933760 76505
Table 0 2606002 76303
15

31. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Big Cache
File System Cache
Storage
shared_buffers

32. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Big Cache
File System Cache
Storage
shared_buffers

33. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Big Cache
File System Cache
Storage

34. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Big Cache
File System Cache
Storage
shared_buffers

35. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Big Cache
File System Cache
Storage
shared_buffers

36. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
© 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
HugePages
17

37. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Why HugePages – page mapping
18
PageTables shared_buffers
pg 1
pg 2
pg 3
4k
4k
size of page table = # of PostgreSQL process X amount of shared buffers accessed
2M

38. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved. 19
0
10,000
20,000
30,000
40,000
50,000
60,000
70,000
5
30
55
80
105
130
155
180
205
230
255
280
305
330
355
380
405
430
455
480
505
530
555
580
605
630
655
680
705
730
755
780
805
830
855
880
905
930
955
980
1005
1030
1055
1080
1105
1130
1155
1180
1205
1230
1255
1280
1305
1330
1355
1380
1405
1430
1455
1480
1505
1530
1555
1580
1605
1630
1655
1680
1705
1730
1755
1780
Transactions
Per
Second
(TPS)
Seconds
Sysbench Read Only Point Selects – r6i.8xlarge – 250 tables x 2.5M rows – 160GB
250 Client Hupepage=on 250 Client Hupepage=off

39. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved. 20
0
10,000
20,000
30,000
40,000
50,000
60,000
70,000
5
30
55
80
105
130
155
180
205
230
255
280
305
330
355
380
405
430
455
480
505
530
555
580
605
630
655
680
705
730
755
780
805
830
855
880
905
930
955
980
1005
1030
1055
1080
1105
1130
1155
1180
1205
1230
1255
1280
1305
1330
1355
1380
1405
1430
1455
1480
1505
1530
1555
1580
1605
1630
1655
1680
1705
1730
1755
1780
Transactions
Per
Second
(TPS)
Seconds
Sysbench Read Only Point Selects – r6i.8xlarge – 250 tables x 2.5M rows – 160GB
500 Client Hupepage=on 500 Client Hupepage=off

40. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved. 21
0
10,000
20,000
30,000
40,000
50,000
60,000
70,000
5
30
55
80
105
130
155
180
205
230
255
280
305
330
355
380
405
430
455
480
505
530
555
580
605
630
655
680
705
730
755
780
805
830
855
880
905
930
955
980
1005
1030
1055
1080
1105
1130
1155
1180
1205
1230
1255
1280
1305
1330
1355
1380
1405
1430
1455
1480
1505
1530
1555
1580
1605
1630
1655
1680
1705
1730
1755
1780
Transactions
Per
Second
(TPS)
Seconds
Sysbench Read Only Point Selects – r6i.8xlarge – 250 tables x 2.5M rows – 160GB
1000 Client Hupepage=on 1000 Client Hupepage=off

41. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Cost of not setting HugePages
22
0%
5%
10%
15%
20%
25%
30%
35%
40%
45%
50%
0
20
40
60
80
100
120
140
250 500 1000
Percent
of
host
memory
GB
of
PageTable
number of postgreSQL process
sysbench read only - shard_buffer=25% of ram
PageTables size (gb) % of host memory

42. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
© 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Per Session
23

43. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
temp_buffers
shared_buffers temp_buffers
File System Cache
regular tables temp tables
global session
local

44. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
SQL Stream
logical_decoding_work_mem
OLTP
DW
SQL Stream
logical
decoding
slot
wal
log
DW2
logical
decoding
slot
SQL Stream

45. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
© 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Per Session Operation
26

46. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
work_mem
postgres=# set work_mem TO '1 GB’;
postgres=# explain analyze
select mykey::bigint, (random()*10000000000)::bigint as scratch ,
repeat('X', 1024)::char(1024) filler from generate_series(1,3800000)
as mykey order by scratch;
QUERY PLAN
------------------------------------------------------------------------------------------------------------------------------------------------
-
Sort (cost=7219224.79..7228724.79 rows=3800000 width=4116) (actual time=7083.506..8358.093 rows=3800000 loops=1)
Sort Key: (((random() * '10000000000'::double precision))::bigint)
Sort Method: external merge Disk: 3919000kB
-> Function Scan on generate_series mykey (cost=0.00..76000.00 rows=3800000 width=4116) (actual time=194.053..493.117 rows=3800000 loops=1)
Planning Time: 0.049 ms
Execution Time: 9117.344 ms
(6 rows)
~3.8GB

47. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved. 28
0
1
2
3
4
5
6
7
8
9
10
1 6 11 16 21 26 31 36 41 46 51 56
GB
RAM
Seconds
work_mem - sorting
1GB work_mem 3.8GB query

48. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
work_mem
postgres=# set work_mem TO '4 GB’;
postgres=# explain analyze
select mykey::bigint, (random()*10000000000)::bigint as scratch ,
repeat('X', 1024)::char(1024) filler from generate_series(1,3800000)
as mykey order by scratch;
QUERY PLAN
------------------------------------------------------------------------------------------------------------------------------------------------
-
Sort (cost=7219224.79..7228724.79 rows=3800000 width=4116) (actual time=2766.682..3408.706 rows=3800000 loops=1)
Sort Key: (((random() * '10000000000'::double precision))::bigint)
Sort Method: quicksort Memory: 4128025kB
-> Function Scan on generate_series mykey (cost=0.00..76000.00 rows=3800000 width=4116) (actual time=194.365..506.965 rows=3800000 loops=1)
Planning Time: 0.047 ms
Execution Time: 3825.965 ms
(6 rows)
~3.8GB

49. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved. 30
0
1
2
3
4
5
6
7
8
9
10
1 6 11 16 21 26 31 36 41 46 51 56
GB
RAM
Seconds
work_mem - sorting
4GB work_mem - 3.8GB query 1GB work_mem 3.8GB query

50. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
work_mem
postgres=# set work_mem TO '4 GB’;
explain analyze select s1.filler, s2.filler,s1.mykey, s2.mykey, s1.scratch, s2.scratch from
(select mykey::bigint, (random()*1000000000)::bigint as scratch ,
repeat('X', 1024)::char(1024) filler from generate_series(1,3800000)
as mykey order by scratch) s1,
(select mykey::bigint, (random()*10000000000)::bigint as scratch ,
repeat('X', 1024)::char(1024) filler from generate_series(1,3800000)
as mykey order by scratch) s2
where s1.scratch=s2.scratch order by s1.mykey;
QUERY PLAN
------------------------------------------------------------------------------------------------------------------------------------------------
---------------------
Sort (cost=523467712452.33..523648212452.33 rows=72200000000 width=8232) (actual time=6851.977..6852.172 rows=1494 loops=1)
Sort Key: ((mykey.mykey)::bigint)
Sort Method: quicksort Memory: 3153kB
-> Merge Join (cost=14438449.59..1097542949.59 rows=72200000000 width=8232) (actual time=5558.904..6850.847 rows=1494 loops=1)
Merge Cond: ((((random() * '1000000000'::double precision))::bigint) = (((random() * '10000000000'::double precision))::bigint))
-> Sort (cost=7219224.79..7228724.79 rows=3800000 width=4116) (actual time=2785.745..3524.489 rows=3800000 loops=1)
Sort Key: (((random() * '1000000000'::double precision))::bigint)
Sort Method: quicksort Memory: 4128025kB
-> Function Scan on generate_series mykey (cost=0.00..76000.00 rows=3800000 width=4116) (actual time=194.222..515.986
rows=3800000 loops=1)
-> Materialize (cost=7219224.79..7276224.79 rows=3800000 width=4116) (actual time=2772.383..2961.007 rows=380357 loops=1)
-> Sort (cost=7219224.79..7228724.79 rows=3800000 width=4116) (actual time=2772.377..2867.394 rows=380353 loops=1)
Sort Key: (((random() * '10000000000'::double precision))::bigint)
Sort Method: quicksort Memory: 4128025kB
-> Function Scan on generate_series mykey_1 (cost=0.00..76000.00 rows=3800000 width=4116) (actual time=198.688..511.083
rows=3800000 loops=1)
Planning Time: 0.098 ms
Execution Time: 7511.377 ms
~3.8GB

51. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved. 32
0
1
2
3
4
5
6
7
8
9
10
1 6 11 16 21 26 31 36 41 46 51 56
GB
RAM
Seconds
work_mem - sorting
4GB work_mem - 7.6GB+ query 4GB work_mem - 3.8GB query
1GB work_mem 3.8GB query

52. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
work_mem
postgres=# set work_mem TO '4 GB’;
postgres=# explain analyze
select s1.filler, s2.filler,s1.mykey, s2.mykey, s1.scratch, s2.scratch from
(select mykey::bigint, (random()*1000000000)::bigint as scratch ,
repeat('X', 1024)::char(1024) filler from generate_series(1,18800000)
as mykey order by scratch) s1,
(select mykey::bigint, (random()*10000000000)::bigint as scratch ,
repeat('X', 1024)::char(1024) filler from generate_series(1,3800000)
as mykey order by scratch) s2
where s1.scratch=s2.scratch order by s1.mykey;
QUERY PLAN
------------------------------------------------------------------------------------------------------------------------------------------------
-------------------------
Sort (cost=3853846009559.71..3854739009559.71 rows=357200000000 width=8232) (actual time=52689.953..52690.952 rows=7184 loops=1)
Sort Key: ((mykey_1.mykey)::bigint)
Sort Method: quicksort Memory: 15122kB
-> Merge Join (cost=43152211.85..5401444211.85 rows=357200000000 width=8232) (actual time=42991.198..52679.295 rows=7184 loops=1)
Merge Cond: ((((random() * '10000000000'::double precision))::bigint) = (((random() * '1000000000'::double precision))::bigint))
-> Sort (cost=7219224.79..7228724.79 rows=3800000 width=4116) (actual time=2771.496..2877.777 rows=380382 loops=1)
Sort Key: (((random() * '10000000000'::double precision))::bigint)
Sort Method: quicksort Memory: 4128025kB
-> Function Scan on generate_series mykey (cost=0.00..76000.00 rows=3800000 width=4116) (actual time=194.404..516.400
rows=3800000 loops=1)
-> Materialize (cost=35932987.06..36214987.06 rows=18800000 width=4116) (actual time=40218.833..48614.865 rows=18800001 loops=1)
-> Sort (cost=35932987.06..35979987.06 rows=18800000 width=4116) (actual time=40218.824..46815.209 rows=18800000 loops=1)
Sort Key: (((random() * '1000000000'::double precision))::bigint)
Sort Method: external merge Disk: 19388680kB
-> Function Scan on generate_series mykey_1 (cost=0.00..376000.00 rows=18800000 width=4116) (actual
time=960.638..2593.178 rows=18800000 loops=1)
Planning Time: 0.102 ms
Execution Time: 55724.412 ms
~3.8GB
~18.8GB

53. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved. 34
0
1
2
3
4
5
6
7
8
9
10
1 6 11 16 21 26 31 36 41 46 51 56
GB
RAM
Seconds
work_mem - sorting
4GB work_mem - 18.8GB+ query 4GB work_mem - 7.6GB+ query
4GB work_mem - 3.8GB query 1GB work_mem 3.8GB query

54. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
hash_mem_multiplier
postgres=# show hash_mem_multiplier;
hash_mem_multiplier
---------------------
2
postgres=# set work_mem=‘4 GB';
QUERY PLAN
---------------------------------------------------------------------------------------------------------------------------------------------------------------
Hash Join (cost=2083766.00..5865195158.01 rows=167200000000 width=8232) (actual time=4360.794..16414.748 rows=3344 loops=1)
Hash Cond: (s1.scratch = s2.scratch)
-> Subquery Scan on s1 (cost=0.00..264000.00 rows=8800000 width=4116) (actual time=374.153..1692.160 rows=8800000 loops=1)
-> Function Scan on generate_series mykey (cost=0.00..176000.00 rows=8800000 width=4116) (actual time=374.152..1161.469 rows=8800000 loops=1)
-> Hash (cost=114000.00..114000.00 rows=3800000 width=4116) (actual time=3982.004..3982.006 rows=3800000 loops=1)
Buckets: 2097152 Batches: 2 Memory Usage: 2020102kB
-> Subquery Scan on s2 (cost=0.00..114000.00 rows=3800000 width=4116) (actual time=186.934..742.574 rows=3800000 loops=1)
-> Function Scan on generate_series mykey_1 (cost=0.00..76000.00 rows=3800000 width=4116) (actual time=186.932..515.290 rows=3800000 loops=1)
Planning Time: 0.090 ms
Execution Time: 16655.195 ms
(10 rows)
postgres=# set work_mem='1 GB';
SET
QUERY PLAN
---------------------------------------------------------------------------------------------------------------------------------------------------------------
Hash Join (cost=2083766.00..5865195158.01 rows=167200000000 width=8232) (actual time=5099.725..23063.116 rows=3575 loops=1)
Hash Cond: (s1.scratch = s2.scratch)
-> Subquery Scan on s1 (cost=0.00..264000.00 rows=8800000 width=4116) (actual time=376.939..1651.354 rows=8800000 loops=1)
-> Function Scan on generate_series mykey (cost=0.00..176000.00 rows=8800000 width=4116) (actual time=376.937..1119.345 rows=8800000 loops=1)
-> Hash (cost=114000.00..114000.00 rows=3800000 width=4116) (actual time=4714.890..4714.891 rows=3800000 loops=1)
Buckets: 524288 Batches: 8 Memory Usage: 505086kB
-> Subquery Scan on s2 (cost=0.00..114000.00 rows=3800000 width=4116) (actual time=181.422..740.776 rows=3800000 loops=1)
-> Function Scan on generate_series mykey_1 (cost=0.00..76000.00 rows=3800000 width=4116) (actual time=181.420..512.871 rows=3800000 loops=1)
Planning Time: 0.089 ms
Execution Time: 23186.765 ms
~2.6GB – 16.6 seconds
~1.2GB – 23.1 seconds

55. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
© 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Per Session Item
36

56. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
prepared statements
PREPARE sbtestplan (int) AS
SELECT * FROM sbtest1 WHERE id=$1 ;
EXECUTE sbtestplan(1);
sbtest=# SELECT name, ident, level, total_bytes
FROM pg_backend_memory_contexts where name ='CachedPlanSource';
name | ident | level | total_bytes
------------------+-----------------------------------------+-------+-------------
CachedPlanSource | PREPARE sbtestplan3 (int) AS +| 2 | 4096
| SELECT * FROM sbtest3 WHERE id=$1 ; | |
CachedPlanSource | PREPARE sbtestplan2 (int) AS +| 2 | 4096
| SELECT * FROM sbtest2 WHERE id=$1 ; | |
CachedPlanSource | PREPARE sbtestplan (int) AS +| 2 | 4096
| SELECT * FROM sbtest1 WHERE id=$1 ; | |

57. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
prepared statements – 10 Million
38
0
20
40
60
80
100
1
7
13
19
25
31
37
43
49
55
61
67
73
79
85
91
97
103
109
115
121
127
133
139
145
151
157
163
169
175
181
187
193
199
205
211
217
223
229
235
241
247
253
259
265
271
277
283
289
295
301
307
313
319
325
331
337
343
349
355
361
GB
Used
Seconds
10 million prepared statements - 1 client

58. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
© 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Maintenance
39

59. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
maintenance_work_mem – Index builds
sbtest=# set maintenance_work_mem = '1 GB';
^
sbtest=# create index foo on sbtest1 (id,c,pad,k desc) ;
CREATE INDEX
Time: 45803.577 ms (00:45.804)
sbtest=# set maintenance_work_mem = '2 GB';
sbtest=# create index foo on sbtest1 (id,c,pad,k desc) ;
CREATE INDEX
Time: 45904.760 ms (00:45.905)
sbtest=# set maintenance_work_mem = '4 GB';
sbtest=# create index foo on sbtest1 (id,c,pad,k desc) ;
CREATE INDEX
Time: 46081.414 ms (00:46.081)
~1GB
~2GB
~3GB

60. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
maintenance_work_mem – how vacuum works
heap
index
index
vacuum
scan heap

61. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
maintenance_work_mem – how vacuum works
heap
index
index
vacuum
maintenance_work_mem
tid to clean
scan heap

62. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
maintenance_work_mem – how vacuum works
heap
index
index
vacuum
maintenance_work_mem
tid to clean
scan heap scan indexes

63. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
maintenance_work_mem - vacuum
sbtest=# set maintenance_work_mem='100 MB';
sbtest=# vacuum (verbose) sbtest1;
INFO: vacuuming "sbtest.public.sbtest1"
INFO: launched 1 parallel vacuum worker for index vacuuming (planned: 1)
INFO: launched 1 parallel vacuum worker for index vacuuming (planned: 1)
INFO: finished vacuuming "sbtest.public.sbtest1": index scans: 2
pages: 0 removed, 13921343 remain, 13921343 scanned (100.00% of total)
tuples: 11014287 removed, 500000000 remain, 0 are dead but not yet removable
removable cutoff: 74587243, which was 0 XIDs old when operation ended
index scan needed: 11236460 pages from table (80.71% of total) had 18915133 dead item identifiers removed
index "sbtest1_pkey": pages: 1370933 in total, 0 newly deleted, 0 currently deleted, 0 reusable
index "k_1": pages: 1202589 in total, 0 newly deleted, 0 currently deleted, 0 reusable
avg read rate: 102.408 MB/s, avg write rate: 99.348 MB/s
buffer usage: 16918231 hits, 24908382 misses, 24164093 dirtied
WAL usage: 48871286 records, 26289264 full page images, 210588376068 bytes
system usage: CPU: user: 184.21 s, system: 275.86 s, elapsed: 1900.21 s
VACUUM

64. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
maintenance_work_mem - vacuum
sbtest=# set maintenance_work_mem='1 GB’;
sbtest=# vacuum (verbose) sbtest1;
INFO: vacuuming "sbtest.public.sbtest1"
INFO: launched 1 parallel vacuum worker for index vacuuming (planned: 1)
INFO: finished vacuuming "sbtest.public.sbtest1": index scans: 1
pages: 0 removed, 13921343 remain, 13921343 scanned (100.00% of total)
tuples: 10990324 removed, 500000000 remain, 0 are dead but not yet removable
removable cutoff: 103987274, which was 0 XIDs old when operation ended
new relfrozenxid: 53987274, which is 11901571 XIDs ahead of previous value
index scan needed: 10126301 pages from table (72.74% of total) had 16347414 dead item identifiers removed
index "sbtest1_pkey": pages: 1370933 in total, 0 newly deleted, 0 currently deleted, 0 reusable
index "k_1": pages: 1202589 in total, 0 newly deleted, 0 currently deleted, 0 reusable
avg read rate: 90.361 MB/s, avg write rate: 101.013 MB/s
buffer usage: 16700415 hits, 22645066 misses, 25314603 dirtied
WAL usage: 63975507 records, 26568355 full page images, 219969619018 bytes
system usage: CPU: user: 195.27 s, system: 265.41 s, elapsed: 1957.86 s
VACUUM

65. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
maintenance_work_mem - vacuum parallel
sbtest=# set maintenance_work_mem='1 GB’;
sbtest=# vacuum (verbose) sbtest1;
INFO: vacuuming "sbtest.public.sbtest1"
INFO: launched 2 parallel vacuum workers for index vacuuming (planned: 2)

66. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
autovacuum_work_mem
max mem = autovacuum_work_mem X autovacuum_max_workers

67. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Why you should care
0
10,000
20,000
30,000
40,000
50,000
60,000
70,000
5 105 205 305 405 505 605 705 805 905 1005 1105 1205 1305 1405 1505 1605 1705
Transactions
Per
Second
(TPS)
Seconds
sysbench read only point selects
Normal Tuning Less than Optimal Tuning
11X
buffers not in memory swap thrashing

68. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Thank you!
© 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Grant McAlister