共計 9439 個字符,預計需要花費 24 分鐘才能閱讀完成。
這篇文章主要講解了“PostgreSQL 中如何監控 VACUUM 的處理過程”,文中的講解內容簡單清晰,易于學習與理解,下面請大家跟著丸趣 TV 小編的思路慢慢深入,一起來研究和學習“PostgreSQL 中如何監控 VACUUM 的處理過程”吧!
概覽
PG 的 MVCC 要求“過期”的數據不能馬上被物理清除,而是標記為 dead rows,這些 dead rows 后續會通過 vacuuming 過程清理。
vacuuming 通過以下方式讓數據庫保持健康:
1. 標記 dead rows 可用于存儲新數據,這樣可以避免不必要的磁盤浪費以及可以跳過 dead rows 以提升順序掃描的性能;
2. 更新 vm(用于跟蹤過期或已廢棄的數據,反應在 pages 上)。這可以提升 index-only scans 的性能;
3. 避免出現事務 ID 回卷失敗。
PG 提供了 autovacuum 機制,通過周期性的運行 ANALYZE 來收集最近頻繁更新的數據表統計信息。
autovacuum 日志信息
這一小節介紹了配置參數中與 vacuum 相關的參數,包括 log_autovacuum_min_duration、autovacuum_naptime 等。
log_autovacuum_min_duration
如 autovacuum 的執行時間超過了該參數配置的時間(ms 為單位)則在日志中記錄。
設置為較低的閾值:10ms
[pg12@localhost pg121db]$ grep log_autovacuum postgresql.conf
log_autovacuum_min_duration = 10 # -1 disables, 0 logs all actions and
執行 SQL
[local:/data/run/pg12]:5120 pg12@testdb=#
[local:/data/run/pg12]:5120 pg12@testdb=# drop table t_autovacuum_1;
DROP TABLE
[local:/data/run/pg12]:5120 pg12@testdb=# create table t_autovacuum_1(id int,c1 varchar(20),c2 varchar(20));
CREATE TABLE
[local:/data/run/pg12]:5120 pg12@testdb=#
[local:/data/run/pg12]:5120 pg12@testdb=# insert into t_autovacuum_1 select x, c1 ||x, c2 ||x from generate_series(1,10000) as x;
INSERT 0 10000
[local:/data/run/pg12]:5120 pg12@testdb=#
[local:/data/run/pg12]:5120 pg12@testdb=#
[local:/data/run/pg12]:5120 pg12@testdb=# drop table t_big_autovacuum_1;
DROP TABLE
[local:/data/run/pg12]:5120 pg12@testdb=# create table t_big_autovacuum_1(id int,c1 varchar(20),c2 varchar(20));
CREATE TABLE
[local:/data/run/pg12]:5120 pg12@testdb=#
[local:/data/run/pg12]:5120 pg12@testdb=# insert into t_big_autovacuum_1 select x, c1 ||x, c2 ||x from generate_series(1,1000000) as x;
INSERT 0 1000000
[local:/data/run/pg12]:5120 pg12@testdb=#
[local:/data/run/pg12]:5120 pg12@testdb=# explain (analyze,verbose) select * from t_autovacuum_1 a join t_big_autovacuum_1 b on a.id = b.id;
QUERY PLAN
------------------------------------------------------------------------------------------------------------------
---------------------------------------
Merge Join (cost=63985.35..155911.33 rows=6068861 width=240) (actual time=480.678..492.453 rows=10000 loops=1)
Output: a.id, a.c1, a.c2, b.id, b.c1, b.c2
Merge Cond: (a.id = b.id)
- Sort (cost=301.37..310.03 rows=3465 width=120) (actual time=3.061..3.835 rows=10000 loops=1)
Output: a.id, a.c1, a.c2
Sort Key: a.id
Sort Method: quicksort Memory: 1213kB
- Seq Scan on public.t_autovacuum_1 a (cost=0.00..97.65 rows=3465 width=120) (actual time=0.020..1.796
rows=10000 loops=1)
Output: a.id, a.c1, a.c2
- Materialize (cost=63683.99..65435.46 rows=350295 width=120) (actual time=477.603..481.446 rows=10001 loops
Output: b.id, b.c1, b.c2
- Sort (cost=63683.99..64559.72 rows=350295 width=120) (actual time=477.598..479.173 rows=10001 loops=
Output: b.id, b.c1, b.c2
Sort Key: b.id
Sort Method: external merge Disk: 31144kB
- Seq Scan on public.t_big_autovacuum_1 b (cost=0.00..9871.95 rows=350295 width=120) (actual tim
e=0.019..176.859 rows=1000000 loops=1)
Output: b.id, b.c1, b.c2
Planning Time: 0.315 ms
Execution Time: 498.336 ms
(19 rows)
[local:/data/run/pg12]:5120 pg12@testdb=#
日志顯示 autovacuum 的 analyze 操作
2019-12-11 12:31:48.655 CST,,,57525,,5df07134.e0b5,1,,2019-12-11 12:31:48 CST,4/3,3091,LOG,00000, automatic analyze of table testdb.public.t_autovacuum_1 system usage: CPU: user: 0.09 s, system: 0.00 s, elapsed: 0.10 s ,,,,,,,,,
2019-12-11 12:31:49.073 CST,,,57525,,5df07134.e0b5,2,,2019-12-11 12:31:48 CST,4/5,3092,LOG,00000, automatic analyze of table testdb.public.t_big_autovacuum_1 system usage: CPU: user: 0.32 s, system: 0.00 s, elapsed: 0.40 s ,,,,,,,,,
autovacuum_naptime
該參數控制 autovacuum 的調度時間,如有多個數據庫,則每個數據庫每隔一定的時間(autovacuum_naptime/ 數據庫個數)啟動 autovacuum 來進行處理。
從 PG 的源碼可見,通過函數 rebuild_database_list 來構建出現變化后的 DatabaseList,鏈表中的數據庫應出現在 pgstats 中,在 autovacuum_naptime 所設定的時間間隔范圍內均勻分布。
比如 autovacuum_naptime = 60s,有 4 個數據庫 db1- db4,那么每隔 60s/ 4 就會有啟動一個 autovacuum worker 對相應的 DB 進行處理。
可能的一個處理時間序列是:db1- XX(時):XX(分):18(秒),db4- XX:XX:33,db4- XX:XX:48,db4- XX:XX:03
后續如需要對 db1- db4 進行 vacuum,那么 db1- db4 會在下一個 18 秒、33 秒、48 秒和 03 秒觸發 autovacuum。
測試場景:創建 3 個數據庫,db1 – db3,每個數據庫每隔 10s 執行全表更新,通過日志觀察 autovacuum 的行為。
-- session 1
\c db1
drop table t_autovacuum_db1;
create table t_autovacuum_db1(id int,c1 varchar(20),c2 varchar(20));
insert into t_autovacuum_db1 select x, c1 ||x, c2 ||x from generate_series(1,500000) as x;
declare
begin
for i in 1..20 loop
update t_autovacuum_db1 set id = i;
commit;
perform pg_sleep(10);
end loop;
-- session 2
\c db2
drop table t_autovacuum_db2;
create table t_autovacuum_db2(id int,c1 varchar(20),c2 varchar(20));
insert into t_autovacuum_db2 select x, c1 ||x, c2 ||x from generate_series(1,500000) as x;
declare
begin
for i in 1..20 loop
update t_autovacuum_db2 set id = i;
commit;
perform pg_sleep(10);
end loop;
-- session 3
\c db3
drop table t_autovacuum_db3;
create table t_autovacuum_db3(id int,c1 varchar(20),c2 varchar(20));
insert into t_autovacuum_db3 select x, c1 ||x, c2 ||x from generate_series(1,500000) as x;
declare
begin
for i in 1..20 loop
update t_autovacuum_db3 set id = i;
commit;
perform pg_sleep(10);
end loop;
$$;
對應的日志輸出
2019-12-11 15:34:05.298 CST,,,1870,,5df09bec.74e,1,,2019-12-11 15:34:04 CST,6/37,3406,LOG,00000, automatic analyze of table db2.public.t_autovacuum_db2 system usage: CPU: user: 0.38 s, system: 0.01 s, elapsed: 0.52 s ,,,,,,,,,
2019-12-11 15:34:20.528 CST,,,1884,,5df09bfb.75c,1,,2019-12-11 15:34:19 CST,6/41,3412,LOG,00000, automatic analyze of table db1.public.t_autovacuum_db1 system usage: CPU: user: 0.51 s, system: 0.01 s, elapsed: 0.71 s ,,,,,,,,,
2019-12-11 15:34:35.482 CST,,,1897,,5df09c0a.769,1,,2019-12-11 15:34:34 CST,6/45,3416,LOG,00000, automatic analyze of table db3.public.t_autovacuum_db3 system usage: CPU: user: 0.40 s, system: 0.01 s, elapsed: 0.61 s ,,,,,,,,,
2019-12-11 15:35:05.904 CST,,,1924,,5df09c28.784,1,,2019-12-11 15:35:04 CST,6/51,0,LOG,00000, automatic vacuum of table db2.public.t_autovacuum_db2 : index scans: 0
pages: 0 removed, 25467 remain, 0 skipped due to pins, 0 skipped frozen
tuples: 500149 removed, 500000 remain, 0 are dead but not yet removable, oldest xmin: 3422
buffer usage: 50964 hits, 2 misses, 6 dirtied
avg read rate: 0.015 MB/s, avg write rate: 0.044 MB/s
system usage: CPU: user: 0.43 s, system: 0.00 s, elapsed: 1.07 s ,,,,,,,,,
2019-12-11 15:35:21.297 CST,,,1937,,5df09c37.791,1,,2019-12-11 15:35:19 CST,6/55,0,LOG,00000, automatic vacuum of table db1.public.t_autovacuum_db1 : index scans: 0
pages: 0 removed, 31832 remain, 0 skipped due to pins, 0 skipped frozen
tuples: 500270 removed, 500000 remain, 0 are dead but not yet removable, oldest xmin: 3427
buffer usage: 63695 hits, 2 misses, 6 dirtied
avg read rate: 0.010 MB/s, avg write rate: 0.031 MB/s
system usage: CPU: user: 0.51 s, system: 0.04 s, elapsed: 1.52 s ,,,,,,,,,
2019-12-11 15:35:36.250 CST,,,1950,,5df09c46.79e,1,,2019-12-11 15:35:34 CST,6/59,0,LOG,00000, automatic vacuum of table db3.public.t_autovacuum_db3 : index scans: 0
pages: 0 removed, 29875 remain, 1 skipped due to pins, 0 skipped frozen
tuples: 270 removed, 499874 remain, 0 are dead but not yet removable, oldest xmin: 3430
buffer usage: 59780 hits, 2 misses, 6 dirtied
avg read rate: 0.011 MB/s, avg write rate: 0.034 MB/s
system usage: CPU: user: 0.43 s, system: 0.03 s, elapsed: 1.38 s ,,,,,,,,,
2019-12-11 15:38:07.146 CST,,,2081,,5df09cdc.821,1,,2019-12-11 15:38:04 CST,6/81,0,LOG,00000, automatic vacuum of table db2.public.t_autovacuum_db2 : index scans: 0
pages: 0 removed, 44551 remain, 0 skipped due to pins, 0 skipped frozen
tuples: 500509 removed, 500000 remain, 0 are dead but not yet removable, oldest xmin: 3463
buffer usage: 89140 hits, 1 misses, 5 dirtied
avg read rate: 0.003 MB/s, avg write rate: 0.017 MB/s
system usage: CPU: user: 0.58 s, system: 0.11 s, elapsed: 2.28 s ,,,,,,,,,
2019-12-11 15:38:07.963 CST,,,2081,,5df09cdc.821,2,,2019-12-11 15:38:04 CST,6/82,3463,LOG,00000, automatic analyze of table db2.public.t_autovacuum_db2 system usage: CPU: user: 0.38 s, system: 0.00 s, elapsed: 0.81 s ,,,,,,,,,
2019-12-11 15:38:25.836 CST,,,2094,,5df09ceb.82e,1,,2019-12-11 15:38:19 CST,6/86,0,LOG,00000, automatic vacuum of table db1.public.t_autovacuum_db1 : index scans: 0
pages: 0 removed, 38185 remain, 0 skipped due to pins, 0 skipped frozen
tuples: 500585 removed, 500000 remain, 0 are dead but not yet removable, oldest xmin: 3464
buffer usage: 76407 hits, 1 misses, 15391 dirtied
avg read rate: 0.001 MB/s, avg write rate: 19.978 MB/s
system usage: CPU: user: 0.66 s, system: 0.03 s, elapsed: 6.01 s ,,,,,,,,,
2019-12-11 15:38:26.593 CST,,,2094,,5df09ceb.82e,2,,2019-12-11 15:38:19 CST,6/87,3464,LOG,00000, automatic analyze of table db1.public.t_autovacuum_db1 system usage: CPU: user: 0.35 s, system: 0.00 s, elapsed: 0.75 s ,,,,,,,,,
2019-12-11 15:38:36.546 CST,,,2107,,5df09cfa.83b,1,,2019-12-11 15:38:34 CST,6/91,0,LOG,00000, automatic vacuum of table db3.public.t_autovacuum_db3 : index scans: 0
pages: 0 removed, 41365 remain, 0 skipped due to pins, 0 skipped frozen
tuples: 500758 removed, 500000 remain, 0 are dead but not yet removable, oldest xmin: 3465
buffer usage: 82768 hits, 1 misses, 2 dirtied
avg read rate: 0.005 MB/s, avg write rate: 0.010 MB/s
system usage: CPU: user: 0.58 s, system: 0.03 s, elapsed: 1.63 s ,,,,,,,,,
2019-12-11 15:38:37.276 CST,,,2107,,5df09cfa.83b,2,,2019-12-11 15:38:34 CST,6/92,3465,LOG,00000, automatic analyze of table db3.public.t_autovacuum_db3 system usage: CPU: user: 0.37 s, system: 0.00 s, elapsed: 0.72 s ,,,,,,,,,
從日志可見,PG 認為目前有 4 個數據庫(60s/4)需要處理,每個 15s 調度一個數據庫。
感謝各位的閱讀,以上就是“PostgreSQL 中如何監控 VACUUM 的處理過程”的內容了,經過本文的學習后,相信大家對 PostgreSQL 中如何監控 VACUUM 的處理過程這一問題有了更深刻的體會,具體使用情況還需要大家實踐驗證。這里是丸趣 TV,丸趣 TV 小編將為大家推送更多相關知識點的文章,歡迎關注!