+ + + + + + + +
+ +
+ +
+
+ +
+ 20 December / + + + + + + + + + / + + 筆記 + + + + +
+
+

[筆記] 測試 postgresql 的pg_prewarm 對效能的影響 / test pg_prewarm in postgresql 11

+
+
+
+
+

老闆提到想要把新系統的 postgresql 資料庫都撈到記憶體裡面

+ +

但是否決了我提出的ramdisk 作法(因為當機的話，資料就沒了)

+ +

在找資料的時候，發現了這個postgresql 的 pg_prewarm extension

+ +

好像有點意思？就來測試看看吧！

+ +

只是目前還不知道該怎麼解讀測試的數據就是了…

+ +

幹！林北真的不是 DBA 啦 =.=

+ +

安裝系統、postgresql 資料庫什麼的就不提了，那不是這次的重點

+ +

修改 postgresql.conf

+ +

編輯postgresql.conf，開啟平行處理以及設定可用記憶體容量

+ +

這台測試機的環境是一台三代i7 , 24G RAM , 240G SSD，安裝debian 10(buster)

+ +
# load libiriaes
+# 其實這次不會用到pg_stat_statements ，不過出於習慣，還是加入開機自動載入吧
+shared_preload_libraries = 'pg_stat_statements'
+
+#------------------------------------------------------------------------------
+# CUSTOMIZED OPTIONS
+#------------------------------------------------------------------------------
+
+max_connections = 20
+shared_buffers = 6GB
+effective_cache_size = 18GB
+maintenance_work_mem = 1536MB
+checkpoint_completion_target = 0.7
+wal_buffers = 16MB
+default_statistics_target = 100
+random_page_cost = 1.1
+effective_io_concurrency = 200
+work_mem = 78643kB
+min_wal_size = 1GB
+max_wal_size = 2GB
+max_worker_processes = 8
+max_parallel_workers_per_gather = 4
+max_parallel_workers = 8
+
+ +

重新啟動postgresql ，準備開始測試囉！

+ +

轉換成 postgres 身份後，進入 psql

+ +

建立測試資料庫

+ +
postgres=# create database test;
+CREATE DATABASE
+postgres=#
+
+ +

連接測試資料庫、建立pg_prewarm extension

+ +
postgres=# \c test ;
+You are now connected to database "test" as user "postgres".
+test=# CREATE EXTENSION pg_prewarm;
+CREATE EXTENSION
+test=# 
+
+ +

建立測試資料表，塞入500萬筆資料

+ +
test=# \timing
+Timing is on.
+test=# CREATE TABLE test_tbl AS
+SELECT floor(random() * (9923123) + 1)::int FROM generate_series(1, 5000000) AS id;
+SELECT 5000000
+Time: 2940.602 ms (00:02.941)
+test=#
+
+ +

檢查看看剛剛建立的table 用了多少空間

+ +

哎呀，看起來用得不多啊

+ +
test=# SELECT pg_size_pretty(pg_relation_size('test_tbl'));
+ 173 MB
+
+ +

玩大一點，塞個一億筆資料好了

+ +
test=# drop table test_tbl;
+Time: 0.361 ms
+test=# CREATE TABLE test_tbl AS
+SELECT floor(random() * (99343) + 1)::int FROM generate_series(1, 100000000) AS id;
+SELECT 100000000
+Time: 6321.415 ms (00:06.321)
+
+test=# SELECT pg_size_pretty(pg_relation_size('test_tbl'));
+pg_size_pretty | 3457 MB
+
+Time: 0.589 ms
+test=#
+
+
+ +

好，現在資料庫長到3457MB了

+ +

先來執行一些初步的取得基本數據

+ +
test=# explain (analyze,buffers) select count(*) from test_tbl;
+                                   QUERY PLAN
+------------------------------------------------------------------------------------------------------------------------
+ Finalize Aggregate  (cost=755978.52..755978.53 rows=1 width=8) (actual time=3331.917..3331.918 rows=1 loops=1)
+   Buffers: shared hit=160 read=442318
+   ->  Gather  (cost=755978.10..755978.51 rows=4 width=8) (actual time=3331.876..3333.674 rows=5 loops=1)
+         Workers Planned: 4
+         Workers Launched: 4
+         Buffers: shared hit=160 read=442318
+         ->  Partial Aggregate  (cost=754978.10..754978.11 rows=1 width=8) (actual time=3329.279..3329.280 rows=1 loops=5)
+               Buffers: shared hit=160 read=442318
+               ->  Parallel Seq Scan on test_tbl  (cost=0.00..692478.08 rows=25000008 width=0) (actual time=0.029..1924.601 rows=20000000 loops=5)
+                     Buffers: shared hit=160 read=442318
+ Planning Time: 0.040 ms
+ Execution Time: 3333.729 ms
+(12 rows)
+
+(END)
+
+
+ +

可以看到打中buffer 的部份其實很少，只有 160 ，大部分都是讀進去buffer (442318)

+ +

來看看 buffer 的使用狀況

+ +
test=# CREATE EXTENSION pg_buffercache;
+CREATE EXTENSION
+test=# select c.relname,pg_size_pretty(count(*) * 8192) as buffered, 
+test-#         round(100.0 * count(*) / ( 
+test(#            select setting from pg_settings 
+test(#            where name='shared_buffers')::integer,1)
+test-#         as buffer_percent, 
+test-#         round(100.0*count(*)*8192 / pg_table_size(c.oid),1) as percent_of_relation
+test-# from pg_class c inner join pg_buffercache b on b.relfilenode = c.relfilenode inner 
+test-# join pg_database d on ( b.reldatabase =d.oid and d.datname =current_database()) 
+test-# group by c.oid,c.relname order by 3 desc limit 10;
+   relname    |  buffered  | buffer_percent | percent_of_relation 
+--------------+------------+----------------+---------------------
+ test_tbl     | 18 MB      |            0.3 |                 0.5
+ pg_am        | 8192 bytes |            0.0 |                20.0
+ pg_index     | 24 kB      |            0.0 |                37.5
+ pg_amproc    | 32 kB      |            0.0 |                50.0
+ pg_cast      | 16 kB      |            0.0 |                33.3
+ pg_depend    | 64 kB      |            0.0 |                13.3
+ pg_amop      | 48 kB      |            0.0 |                54.5
+ pg_namespace | 8192 bytes |            0.0 |                20.0
+ pg_opclass   | 16 kB      |            0.0 |                28.6
+ pg_aggregate | 8192 bytes |            0.0 |                16.7
+(10 rows)
+
+Time: 148.719 ms
+test=# 
+
+ +

可以看到這個 test_tbl 只有0.5% 被撈到shared_buffers 裡面

+ +

接下來就把這個table全部推到shared_buffers 裡面去

+ +
test=# select pg_prewarm('test_tbl','buffer');
+ pg_prewarm 
+------------
+     442478
+(1 row)
+
+Time: 1938.043 ms (00:01.938)
+test=#
+
+ +

然後再來看一次shared_buffers的使用狀況

+ +
test=# select c.relname,pg_size_pretty(count(*) * 8192) as buffered, 
+        round(100.0 * count(*) / ( 
+           select setting from pg_settings 
+           where name='shared_buffers')::integer,1)
+        as buffer_percent, 
+        round(100.0*count(*)*8192 / pg_table_size(c.oid),1) as percent_of_relation
+from pg_class c inner join pg_buffercache b on b.relfilenode = c.relfilenode inner 
+join pg_database d on ( b.reldatabase =d.oid and d.datname =current_database()) 
+group by c.oid,c.relname order by 3 desc limit 10;
+   relname    |  buffered  | buffer_percent | percent_of_relation 
+--------------+------------+----------------+---------------------
+ test_tbl     | 3457 MB    |           56.3 |               100.0
+ pg_am        | 8192 bytes |            0.0 |                20.0
+ pg_index     | 24 kB      |            0.0 |                37.5
+ pg_amproc    | 32 kB      |            0.0 |                50.0
+ pg_cast      | 16 kB      |            0.0 |                33.3
+ pg_depend    | 64 kB      |            0.0 |                13.3
+ pg_amop      | 48 kB      |            0.0 |                54.5
+ pg_namespace | 8192 bytes |            0.0 |                20.0
+ pg_opclass   | 16 kB      |            0.0 |                28.6
+ pg_aggregate | 8192 bytes |            0.0 |                16.7
+(10 rows)
+
+Time: 2778.354 ms (00:02.778)
+test=# 
+
+ +

OK ，可以看到 test_tbl 已經通通被載入 shared_buffers 中

+ +

buffered 表示表格被載入shared_buffers的大小

+ +

buffer_percent 表示這個表格佔用多少shared_buffers 的比例

+ +

percent_of_relation 表示這個表格有多少比例被載入 shared_buffers

+ +

再來跑一次explain看看狀況

+ +
test=# explain (analyze,buffers) select count(*) from test_tbl;
+Time: 3551.785 ms (00:03.552)
+ Finalize Aggregate  (cost=755978.52..755978.53 rows=1 width=8) (actual time=3427.286..3427.287 rows=1 loops=1)
+   Buffers: shared hit=442478
+   ->  Gather  (cost=755978.10..755978.51 rows=4 width=8) (actual time=3427.215..3551.326 rows=5 loops=1)
+         Workers Planned: 4
+         Workers Launched: 4
+         Buffers: shared hit=442478
+         ->  Partial Aggregate  (cost=754978.10..754978.11 rows=1 width=8) (actual time=3423.659..3423.659 rows=1 loops=5)
+               Buffers: shared hit=442478
+               ->  Parallel Seq Scan on test_tbl  (cost=0.00..692478.08 rows=25000008 width=0) (actual time=0.017..1976.744 rows=20000000 loops=5)
+                     Buffers: shared hit=442478
+ Planning Time: 0.039 ms
+ Execution Time: 3551.365 ms
+(12 rows)
+
+
+ +

這邊就可以看到都是從buffer 讀出來所以 hit=442478

+ +

看樣子表格還是太小，所以沒有完全發揮？那再來把表格加大！

+ +

先重開一次 postgresql 清除buffer

+ +

然後重新建立表格

+ +
test=# drop table test_tbl;
+DROP TABLE
+Time: 297.493 ms
+test=# CREATE TABLE test_tbl AS
+test-# SELECT floor(random() * (993343) + 1)::int FROM generate_series(1, 300000000) AS id;
+SELECT 300000000
+Time: 290660.607 ms (04:50.661)
+test=# 
+
+ +

一樣，看看用了多少容量

+ +
test=# SELECT pg_size_pretty(pg_relation_size('test_tbl'));
+ pg_size_pretty 
+----------------
+ 10 GB
+(1 row)
+
+Time: 0.474 ms
+test=# 
+
+ +

哇哈哈，用了10G ，這次還不撐爆你！

+ +

跑explain 看看狀況

+ +
test=# explain (analyze,buffers) select count(*) from test_tbl;
+Time: 22909.065 ms (00:22.909)
+
+                                              QUERY PLAN
+---------------------------------------------------------------------------------------------------------------------------
+ Finalize Aggregate  (cost=2265934.72..2265934.73 rows=1 width=8) (actual time=22906.045..22906.045 rows=1 loops=1)
+   Buffers: shared hit=2080 read=1325354 dirtied=1295425 written=1295265
+   ->  Gather  (cost=2265934.30..2265934.71 rows=4 width=8) (actual time=22905.997..22908.522 rows=5 loops=1)
+         Workers Planned: 4
+         Workers Launched: 4
+         Buffers: shared hit=2080 read=1325354 dirtied=1295425 written=1295265
+         ->  Partial Aggregate  (cost=2264934.30..2264934.31 rows=1 width=8) (actual time=22903.473..22903.474 rows=1 loops=5)
+               Buffers: shared hit=2080 read=1325354 dirtied=1295425 written=1295265
+               ->  Parallel Seq Scan on test_tbl  (cost=0.00..2077434.24 rows=75000024 width=0) (actual time=0.040..18374.277 rows=60000000 loops=5)
+                     Buffers: shared hit=2080 read=1325354 dirtied=1295425 written=1295265
+ Planning Time: 0.094 ms
+ Execution Time: 22908.571 ms
+(12 rows)
+
+
+ +

看一下現在 shared_buffers 使用狀況

+ +

可以看到這個 test_tbl 幾乎沒被放入 shared_buffers 中

+ +
test=# select c.relname,pg_size_pretty(count(*) * 8192) as buffered, 
+        round(100.0 * count(*) / ( 
+           select setting from pg_settings 
+           where name='shared_buffers')::integer,1)
+        as buffer_percent, 
+        round(100.0*count(*)*8192 / pg_table_size(c.oid),1) as percent_of_relation
+from pg_class c inner join pg_buffercache b on b.relfilenode = c.relfilenode inner 
+join pg_database d on ( b.reldatabase =d.oid and d.datname =current_database()) 
+group by c.oid,c.relname order by 3 desc limit 10;
+   relname    |  buffered  | buffer_percent | percent_of_relation 
+--------------+------------+----------------+---------------------
+ test_tbl     | 18 MB      |            0.3 |                 0.2
+ pg_am        | 8192 bytes |            0.0 |                20.0
+ pg_index     | 24 kB      |            0.0 |                37.5
+ pg_amproc    | 32 kB      |            0.0 |                50.0
+ pg_cast      | 16 kB      |            0.0 |                33.3
+ pg_depend    | 64 kB      |            0.0 |                13.3
+ pg_amop      | 48 kB      |            0.0 |                54.5
+ pg_namespace | 8192 bytes |            0.0 |                20.0
+ pg_opclass   | 16 kB      |            0.0 |                28.6
+ pg_aggregate | 8192 bytes |            0.0 |                16.7
+(10 rows)
+
+Time: 163.936 ms
+test=# 
+
+ +

強制把test_tbl 全部塞進 shared_buffers

+ +
test=# select pg_prewarm('test_tbl','buffer');
+ pg_prewarm 
+------------
+    1327434
+(1 row)
+
+Time: 7472.805 ms (00:07.473)
+test=# 
+
+ +

確認一下test_tbl 有沒有被整個塞進去

+ +
test=# select c.relname,pg_size_pretty(count(*) * 8192) as buffered, 
+        round(100.0 * count(*) / ( 
+           select setting from pg_settings 
+           where name='shared_buffers')::integer,1)
+        as buffer_percent, 
+        round(100.0*count(*)*8192 / pg_table_size(c.oid),1) as percent_of_relation
+from pg_class c inner join pg_buffercache b on b.relfilenode = c.relfilenode inner 
+join pg_database d on ( b.reldatabase =d.oid and d.datname =current_database()) 
+group by c.oid,c.relname order by 3 desc limit 10;
+   relname    |  buffered  | buffer_percent | percent_of_relation 
+--------------+------------+----------------+---------------------
+ test_tbl     | 6142 MB    |          100.0 |                59.2
+ pg_am        | 8192 bytes |            0.0 |                20.0
+ pg_index     | 24 kB      |            0.0 |                37.5
+ pg_amproc    | 32 kB      |            0.0 |                50.0
+ pg_cast      | 16 kB      |            0.0 |                33.3
+ pg_depend    | 24 kB      |            0.0 |                 5.0
+ pg_amop      | 40 kB      |            0.0 |                45.5
+ pg_namespace | 8192 bytes |            0.0 |                20.0
+ pg_opclass   | 16 kB      |            0.0 |                28.6
+ pg_aggregate | 8192 bytes |            0.0 |                16.7
+(10 rows)
+
+Time: 4985.366 ms (00:04.985)
+test=#
+
+ +

GOOD ！ let’s do explain again !

+ +
test=# explain (analyze,buffers) select count(*) from test_tbl;
+Time: 11451.188 ms (00:11.451)
+                             QUERY PLAN
+--------------------------------------------------------------------------------------------------------------------
+ Finalize Aggregate  (cost=2265934.72..2265934.73 rows=1 width=8) (actual time=11231.664..11231.664 rows=1 loops=1)
+   Buffers: shared hit=785963 read=541471
+   ->  Gather  (cost=2265934.30..2265934.71 rows=4 width=8) (actual time=11231.606..11450.719 rows=5 loops=1)
+         Workers Planned: 4
+         Workers Launched: 4
+         Buffers: shared hit=785963 read=541471
+         ->  Partial Aggregate  (cost=2264934.30..2264934.31 rows=1 width=8) (actual time=11228.829..11228.830 rows=1 loops=5)
+               Buffers: shared hit=785963 read=541471
+               ->  Parallel Seq Scan on test_tbl  (cost=0.00..2077434.24 rows=75000024 width=0) (actual time=0.037..6414.711 rows=60000000 loops=5)
+                     Buffers: shared hit=785963 read=541471
+ Planning Time: 0.039 ms
+ Execution Time: 11450.781 ms
+(12 rows)
+
+ +

確認一下，果然大部分都打到cache 了，但是因為shared_buffers 不夠大，所以還會從磁碟讀取一部分

+ +

而且時間也比之前還沒塞進shared_buffers 的時候要快了不少

+ +

22908.571 –> 11450.781 ms

+ +
+ +

從這次的測試看來，我想如果有足夠大的記憶體，能夠把資料表都塞入shared_buffers 中

+ +

應該可以帶來不錯的效能增幅！

+
+
+ +
+ + +

Categories + + 筆記 +

+
+ + + +

Tags + + postgresql + + +

+ + +
+ +
+ + Eric Chang + + + Written by: +
+
+ + +

塵世裡一個迷途小書僮

+ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
+
+ +
+
+ +
+
+ + +comments powered by Disqus + + + + + + +
+
+ + + + + +