死元组未被 VACUUM FULL 命令收集

Question

抱歉，这很长 post 和问题 & postgres 新手

想知道 postgres VACCUM 和 VACCUM FULL 是如何工作的，这对于 700 GB Table 无需停机归档。

VACCUM 完整案例：

在这种情况下，在执行 vacuum full 命令后，内存似乎被释放到 OS，但 n_dead_tuples 保持相同大小。

这个n_dead_tup会在下一次插入/更新/删除中重复使用吗？

步骤：

CREATE TABLE Foo (
  id SERIAL PRIMARY KEY,
  x  INTEGER NOT NULL
);

// CREATE TABLE

INSERT INTO Foo (x) VALUES (generate_series(1,1000000));

// INSERT 0 1000000

SELECT pg_size_pretty(pg_relation_size('foo'));

// 35 MB

SELECT n_dead_tup FROM pg_stat_user_tables where relname = 'foo';

// 0

UPDATE Foo SET x = x + 1;

// UPDATE 1000000

SELECT pg_size_pretty(pg_relation_size('foo'));

// 69 MB

SELECT n_dead_tup FROM pg_stat_user_tables where relname = 'foo';

// 1000000

VACUUM FULL VERBOSE foo;

// INFO:  vacuuming "public.foo"
// INFO:  "foo": found 1000000 removable, 1000000 nonremovable row versions in 8850 pages
// DETAIL:  0 dead row versions cannot be removed yet.
// CPU 0.07s/0.67u sec elapsed 0.76 sec.
// VACUUM

 SELECT pg_size_pretty(pg_relation_size('foo'));

// 35 MB

SELECT n_dead_tup FROM pg_stat_user_tables where relname = 'foo';

// 1000000

VACCUM 案例

步骤：

与上述步骤相同，除了 VACCUM FULL，需要运行下面的命令，在这种情况下看起来死元组被删除但大小没有释放到 OS.

VACUUM VERBOSE foo;

// INFO:  vacuuming "public.foo"
// INFO:  scanned index "foo_pkey" to remove 1000000 row versions
// DETAIL:  CPU 0.01s/0.27u sec elapsed 0.34 sec.
// INFO:  "foo": removed 1000000 row versions in 4425 pages
// DETAIL:  CPU 0.00s/0.02u sec elapsed 0.02 sec.
// INFO:  index "foo_pkey" now contains 1000000 row versions in 8237 pages
// DETAIL:  1000000 index row versions were removed.
// 0 index pages have been deleted, 0 are currently reusable.
// CPU 0.00s/0.00u sec elapsed 0.00 sec.
// INFO:  "foo": found 1000000 removable, 1000000 nonremovable row versions in 8850 out of 8850 pages
// DETAIL:  0 dead row versions cannot be removed yet.
// There were 0 unused item pointers.
// Skipped 0 pages due to buffer pins.
// 0 pages are entirely empty.
// CPU 0.03s/0.52u sec elapsed 0.61 sec.
// VACUUM

SELECT pg_size_pretty(pg_relation_size('foo'));

// 69 MB

SELECT n_dead_tup FROM pg_stat_user_tables where relname = 'foo';

// 0

AUTO VACUUM 等同于 VACUUM ?

提前致谢。

Answer 1

一个简单的 VACUUM 通常不会像 explained in the documentation:

那样释放磁盘 space

[...] However, extra space is not returned to the operating system (in most cases); it's just kept available for re-use within the same table. [...]

对于 VACUUM FULL 它是不同的，因为它基本上将整个 table 数据重写为 explained in the documentation:

[...] This method also requires extra disk space, since it writes a new copy of the table and doesn't release the old copy until the operation is complete. [...]

通过重写，数据在物理上紧密地写入，没有旧行版本引起的漏洞。所需的不断增长的磁盘 space 是 PostgreSQL 实施的 MVCC 的缺点之一，需要仔细观察和调整大型和更新繁重的 tables。

自动吸尘器将永远不会执行VACUUM FULL，因为它会阻塞它正在工作的table/relation上的任何其他activity对任何连接的应用程序产生不可预测的table后果。

如果您运行遇到这里的问题，您应该至少开始考虑以维护 VACUUM FULL 的影响保持较低的方式对 table 进行分区（因为分区体积小）甚至更好，使您能够只删除一个分区而不需要重写。但这实际上取决于您存储和更新的数据的语义。

死元组未被 VACUUM FULL 命令收集

Dead Tuples are not being collected by VACCUM FULL Command

sql

postgresql

vacuum

postgresql-9.4