标签
PostgreSQL , 10.0 , WARM , 写放大 , 索引写放大
背景
目前，PostgreSQL的MVCC是多版本来实现的，当更新数据时，产生新的版本。(社区正在着手增加基于回滚段的存储引擎)
由于索引存储的是KEY+CTID（行号），当tuple的新版本与旧版本不在同一个数据块(BLOCK)的时候，索引也要随之变化，当新版本在同一个块里面时，则发生HOT UPDATE，索引的值不需要更新，但是因为产生了一条新的记录，所以也需要插入一条索引item，垃圾回收时，将其回收，因此产生了写放大。
（HOT指，旧的tuple头部，CHAIN指向新的TUPLE。）
但是HOT总不能覆盖100%的更新，当tuple新版本不在同一个BLOCK时，即使索引的字段值未发生变化，也需要更新索引，因为行号变化了。
这个问题在UBER的某篇文档中反映过
《为PostgreSQL讨说法 - 浅析《UBER ENGINEERING SWITCHED FROM POSTGRES TO MYSQL》》
PostgreSQL 10.0会将这个问题解决掉，有两个手段。
1. 增加间接索引的功能。
《PostgreSQL 10.0 preview 性能增强 - 间接索引(secondary index)》
2. 增加WARM的特性，也就是本文要说的。
索引放大的问题
使用pageinspect观察page的变化。
postgres=# create extension pageinspect ;
CREATE EXTENSION
创建一个表，3列，每列一个索引。
postgres=# create table tbl(id int primary key, c1 int, c2 int);
CREATE TABLE
postgres=# create index idx1 on tbl (c1);
^[[ACREATE INDEX
postgres=# create index idx2 on tbl (c2);
CREATE INDEX
插入一条记录
postgres=# insert into tbl values (1,1,1);
INSERT 0 1
更新一个字段的值，其他字段的值不变
postgres=# begin;
BEGIN
postgres=# update tbl set c1=2 where id=1 returning ctid,*;
ctid | id | c1 | c2
-------+----+----+----
(0,2) | 1 | 2 | 1
(1 row)
UPDATE 1
ctid=0,2，因此发生了HOT，记录没有出现在其他PAGE。
在另一个会话，观察三个索引的leaf page，可以看到，数据没有变化的索引，也插入了一条ITEM
postgres=# select * from bt_page_items('idx2',1);
itemoffset | ctid | itemlen | nulls | vars | data
------------+-------+---------+-------+------+-------------------------
1 | (0,2) | 16 | f | f | 01 00 00 00 00 00 00 00
2 | (0,1) | 16 | f | f | 01 00 00 00 00 00 00 00
(2 rows)
postgres=# select * from bt_page_items('idx1',1);
itemoffset | ctid | itemlen | nulls | vars | data
------------+-------+---------+-------+------+-------------------------
1 | (0,1) | 16 | f | f | 01 00 00 00 00 00 00 00
2 | (0,2) | 16 | f | f | 02 00 00 00 00 00 00 00
(2 rows)
postgres=# select * from bt_page_items('tbl_pkey',1);
itemoffset | ctid | itemlen | nulls | vars | data
------------+-------+---------+-------+------+-------------------------
1 | (0,2) | 16 | f | f | 01 00 00 00 00 00 00 00
2 | (0,1) | 16 | f | f | 01 00 00 00 00 00 00 00
(2 rows)
使用vacuum回收垃圾（或者等其自动回收），垃圾回收时，如果要删除heap表中的dead tuple，首先要删除索引对应的item。
postgres=# end;
COMMIT
postgres=# vacuum verbose tbl;
INFO: vacuuming "public.tbl"
INFO: scanned index "tbl_pkey" to remove 1 row versions
DETAIL: CPU 0.00s/0.00u sec elapsed 0.00 sec.
INFO: scanned index "idx1" to remove 1 row versions
DETAIL: CPU 0.00s/0.00u sec elapsed 0.00 sec.
INFO: scanned index "idx2" to remove 1 row versions
DETAIL: CPU 0.00s/0.00u sec elapsed 0.00 sec.
INFO: "tbl": removed 1 row versions in 1 pages
DETAIL: CPU 0.00s/0.00u sec elapsed 0.00 sec.
INFO: index "tbl_pkey" now contains 1 row versions in 2 pages
DETAIL: 1 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: index "idx1" now contains 1 row versions in 2 pages
DETAIL: 1 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: index "idx2" now contains 1 row versions in 2 pages
DETAIL: 1 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: "tbl": found 1 removable, 1 nonremovable row versions in 1 out of 1 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 0.00s/0.00u sec elapsed 0.00 sec.
VACUUM
ostgres=# select * from bt_page_items('idx1',1);
itemoffset | ctid | itemlen | nulls | vars | data
------------+-------+---------+-------+------+-------------------------
1 | (0,2) | 16 | f | f | 02 00 00 00 00 00 00 00
(1 row)
postgres=# select * from bt_page_items('idx2',1);
itemoffset | ctid | itemlen | nulls | vars | data
------------+-------+---------+-------+------+-------------------------
1 | (0,2) | 16 | f | f | 01 00 00 00 00 00 00 00
(1 row)
postgres=# select * from bt_page_items('tbl_pkey',1);
itemoffset | ctid | itemlen | nulls | vars | data
------------+-------+---------+-------+------+-------------------------
1 | (0,2) | 16 | f | f | 01 00 00 00 00 00 00 00
(1 row)
写放大就是这样产生的，所以10.0的两个特性可以解决这样的问题。
间接索引
之前讲过，这里就不重复了
《PostgreSQL 10.0 preview 性能增强 - 间接索引(secondary index)》
WARM
warm使用chain和recheck来解决写放大的问题。
建表与索引如下
CREATE TABLE test (col1 int, col2 int, col3 int, col4 text);
CREATE INDEX testindx_col1 ON test (col1);
CREATE INDEX testindx_col2 ON test (col2);
CREATE INDEX testindx_col3 ON test (col3);
INSERT INTO test VALUES (1, 11, 111, 'foo');
当前的索引条目
testindx_col1: (1) ===> (0,1)
testindx_col2: (11) ===> (0,1)
testindx_col3: (111) ===> (0,1)
WARM上场，更新col1=2，指向0,2，此时没有变更的索引，不需要更新，依旧指向0,1，而发生变化的索引，也指向0,1，通过chain指向0,2。
Now if a transaction T1 UPDATEs the table such as, "UPDATE test SET col1 =
2". This does not satisfy the HOT property since index column 'col1' is
being updated. But as per WARM algorithm, since only testindx_col1's index
key has changed, we insert a new index tuple only in testindx_col1. Say the
new heap tuple has CTID (0,2). So testindx_col1 will have a new index tuple
with key (2), but still pointing to CTID (0,1)
testindx_col1: (1) ==> (0,1)
(2) ===> (0,1)
testindx_col2: (11) ===> (0,1)
testindx_col3: (111) ===> (0,1)
The heap at this point has two tuples, linked by CTID chain.
(0,1)* ---> (0,2)
不管怎么更新，都通过chain来指向，所以不需要修改没有发生变化的索引。
If T3 later updates col2 and T4 further updates col3,
T3: UPDATE test SET col2 = 12;
T4: UPDATE test SET col3 = 112;
The heap will look like:
(0,1)* ---> (0,2) ---> (0,3) ---> (0,4)
And the index pointers will be:
testindx_col1: (1) ===> (0,1)
(2) ===> (0,1)
testindx_col2: (11) ===> (0,1)
(12) ===> (0,1)
testindx_col3: (111) ===> (0,1)
(112) ===> (0,1)
通过FLAG表示当前行是否有chain。
当查询到chain tuple时，顺藤摸瓜，根据事务快照，判断可见性。
为什么需要recheck, 因为warm的引入发生值变化的索引，值与heap root lp中存储的值不一样了，因此需要recheck.
warm的引入，对有较多索引的表的更新，性能提升是非常明显的。
这个patch的讨论，详见邮件组，本文末尾URL。
PostgreSQL社区的作风非常严谨，一个patch可能在邮件组中讨论几个月甚至几年，根据大家的意见反复的修正，patch合并到master已经非常成熟，所以PostgreSQL的稳定性也是远近闻名的。
