PG数据库表膨胀不能清理问题分析处理

PostgreSQL数据库工作学习随笔 2022-07-09

1522

我们知道默认情况下表在条件达到autovacuum_vacuum_scale_factor和autovacuum_vacuum_threshold的限制时会触发autovacuum，比如基于1000W的表，需要有20W以上数据进行更改时才触发autovacuum，但是同时对于有锁的事务比如长事务、大量频繁的更新操作，autovacuum会跳过，由于时间限制不能完成表清理。

那么我们怎么判断是哪些原因导致表不能清理呢？我们做一个测试说明长事务导致表不能清理的情况：

为了方便测试我们把所有autovacuum执行过程写入日志。
postgres=# show log_autovacuum_min_duration ;
 log_autovacuum_min_duration
-----------------------------
 -1
(1 row)


postgres=# alter system set log_autovacuum_min_duration to 0;
ALTER SYSTEM
postgres=# select pg_reload_conf();
 pg_reload_conf
----------------
 t
(1 row)


postgres=# show log_autovacuum_min_duration ;
 log_autovacuum_min_duration
-----------------------------
 0
(1 row)


postgres=#

创建测试表和测试数据，为了方便测试我把表tbl_test_autovacuum的autovacuum触发条件改为autovacuum_vacuum_scale_factor=0, autovacuum_vacuum_threshold=1
postgres=# create table tbl_test_autovacuum(id int ,name varchar);
CREATE TABLE
postgres=# alter table tbl_test_autovacuum set (autovacuum_vacuum_scale_factor=0, autovacuum_vacuum_threshold=1);
ALTER TABLE
postgres=# insert into tbl_test_autovacuum values (1,'hl'),(2,'hl'),(3,'hl'),(4,'hl'),(5,'hl');
INSERT 0 5
postgres=#


查看数据
postgres=# select relname,relpages,reltuples,pg_stat_get_live_tuples(oid),pg_stat_get_dead_tuples(oid) from pg_class where relname='tbl_test_autovacuum';
       relname       | relpages | reltuples | pg_stat_get_live_tuples | pg_stat_get_dead_tuples
---------------------+----------+-----------+-------------------------+-------------------------
 tbl_test_autovacuum |        1 |         5 |                       5 |                       0
(1 row)


postgres=#

session1 开启一个事务，更新一条数据，不提交

postgres=# begin;
BEGIN
postgres=*# update tbl_test_autovacuum set name='test' where id=1;
UPDATE 1
postgres=*#

session2 删除部分数据触发autovacuum

postgres=# delete from tbl_test_autovacuum where id > 2;
DELETE 3
postgres=#

再次查看数据分布情况，产生了三个死元组
postgres=# select relname,relpages,reltuples,pg_stat_get_live_tuples(oid),pg_stat_get_dead_tuples(oid) from pg_class where relname='tbl_test_autovacuum';
       relname       | relpages | reltuples | pg_stat_get_live_tuples | pg_stat_get_dead_tuples
---------------------+----------+-----------+-------------------------+-------------------------
 tbl_test_autovacuum |        1 |         5 |                       2 |                       3
(1 row)


postgres=#

查看数据库日志

autovacuum 触发了，但是有些行无法移除，在查询中说明涉及到的行是会被保护的，提示oldest xmin: 1703。该 xmin 是造成不能 vacuum 的事务 id，根据提示信息，有尚未提交的事务，导致autovacuum 进程不能 vacuum。
2022-06-24 02:30:42.575 EDT [9425] LOG:  automatic vacuum of table "postgres.public.tbl_test_autovacuum": index scans: 0
        pages: 0 removed, 1 remain, 0 skipped due to pins, 0 skipped frozen
        tuples: 0 removed, 5 remain, 3 are dead but not yet removable, oldest xmin: 1703
index scan not needed: 0 pages from table (0.00% of total) had 0 dead item identifiers removed
        avg read rate: 0.000 MB/s, avg write rate: 0.000 MB/s
        buffer usage: 32 hits, 0 misses, 0 dirtied
        WAL usage: 0 records, 0 full page images, 0 bytes
system usage: CPU: user: 0.00 s, system: 0.00 s, elapsed: 0.00 s

查看1703是什么事务，通过pg_stat_activity视图可以看到就是session1执行的未提交事务。

postgres=# \x
Expanded display is on.
postgres=# select * from pg_stat_activity where backend_xid=1703;
-[ RECORD 1 ]----+-------------------------------------------------------
datid            | 13892
datname          | postgres
pid              | 8926
leader_pid       |
usesysid         | 10
usename          | atlasdb
application_name | psql
client_addr      |
client_hostname  |
client_port      | -1
backend_start    | 2022-06-24 02:22:43.805632-04
xact_start       | 2022-06-24 02:27:55.184073-04
query_start      | 2022-06-24 02:27:57.200055-04
state_change     | 2022-06-24 02:27:57.200932-04
wait_event_type  | Client
wait_event       | ClientRead
state            | idle in transaction
backend_xid      | 1703
backend_xmin     |
query_id         | 3783348657976902404
query            | update tbl_test_autovacuum set name='test' where id=1;
backend_type     | client backend


postgres=#

此时提交session1中的事务

postgres=# begin;
BEGIN
postgres=*# update tbl_test_autovacuum set name='test' where id=1;
UPDATE 1
postgres=*# end;
COMMIT
postgres=
查看数据元组情况，由于autovacuum_naptime参数设置不能立刻触发autovacuum
postgres=# select relname,relpages,reltuples,pg_stat_get_live_tuples(oid),pg_stat_get_dead_tuples(oid) from pg_class where relname='tbl_test_autovacuum';
       relname       | relpages | reltuples | pg_stat_get_live_tuples | pg_stat_get_dead_tuples
---------------------+----------+-----------+-------------------------+-------------------------
 tbl_test_autovacuum |        1 |         2 |                       2 |                       4
(1 row)


postgres=#

继续观察数据库日志，可以看到死元组被清理了。

2022-06-24 02:38:42.677 EDT [9841] LOG:  automatic vacuum of table "postgres.public.tbl_test_autovacuum": index scans: 0
        pages: 0 removed, 1 remain, 0 skipped due to pins, 0 skipped frozen
        tuples: 4 removed, 2 remain, 0 are dead but not yet removable, oldest xmin: 1705
index scan not needed: 1 pages from table (100.00% of total) had 3 dead item identifiers removed
        avg read rate: 0.000 MB/s, avg write rate: 0.000 MB/s
        buffer usage: 32 hits, 0 misses, 0 dirtied
        WAL usage: 2 records, 0 full page images, 118 bytes
system usage: CPU: user: 0.00 s, system: 0.00 s, elapsed: 0.00 s

再次查看数据分布情况，已经没有死元组了

postgres=# select relname,relpages,reltuples,pg_stat_get_live_tuples(oid),pg_stat_get_dead_tuples(oid) from pg_class where relname='tbl_test_autovacuum';
       relname       | relpages | reltuples | pg_stat_get_live_tuples | pg_stat_get_dead_tuples
---------------------+----------+-----------+-------------------------+-------------------------
 tbl_test_autovacuum |        1 |         2 |                       2 |                       0
(1 row)


postgres=#

总结：

1. 在存在长事务的时候，触发autovacuum后死元组是不被清理的；

2. 当长事务提交，但是条件不够触发autovacuum时，可以使用vacuum手动进行数据清理；

3. 如果使用vacuum清理的时候如果长事务未提交，也是不能清理死元组的。

事务数据库事务 test stat

文章转载自PostgreSQL数据库工作学习随笔，如果涉嫌侵权，请发送邮件至：contact@modb.pro进行举报，并提供相关证据，一经查实，墨天轮将立刻删除相关内容。

PG数据库表膨胀不能清理问题分析处理

评论