提高了PostgreSQL查询性能，并且已经连接了1亿个数据
我使用的是Postgresql-9.2 version ， Windows 7 64 bit ， RAM 6GB 。 这是一个Java企业项目。
我必须在我的页面中显示订单相关信息。 有三个表通过左连接汇集在一起。
表：
- TV_HD（389772行）
- TV_SNAPSHOT（1564756行）
- TD_MAKKA（419298行）
在离开加入3个表后，查询给出了487252 。 它也会日益增加。
表关系：
- TV_HD包含与TV_SNAPSHOT的“一对多”关系
- TV_HD与TD_MAKKA包含“一对多”关系
为了更好地理解，我现在给出一个带有sql查询的图片视图
SELECT * FROM tv_hd其中urino = 1630799 
SELECT * FROM tv_snapshot其中urino = 1630799 
SELECT * FROM td_makka其中urino = 1630799 
此查询大约在90秒内运行。 如何提高查询性能？
我也想过索引。 但据我所知，当我们想从表中获得2％-4％的数据时，实际使用了索引。但在我的情况下，我需要来自这3个表的所有数据。
这是查询：
SELECT count(*) FROM (SELECT HD.URINO FROM TV_HD HD LEFT JOIN TV_SNAPSHOT T ON (HD.URINO = T.URINO AND HD.TCODE = T.TCODE AND T.DELFLG = 0 AND T.SYUBETSU = 1) LEFT JOIN TV_SNAPSHOT T_SQ ON (HD.URINO = T_SQ.URINO AND HD.SQCODE = T_SQ.TCODE AND T_SQ.DELFLG = 0 AND T_SQ.SYUBETSU = 3) LEFT JOIN (SELECT N.URINO FROM TD_MAKKA N WHERE N.UPDATETIME IN ( SELECT MIN(NMIN.UPDATETIME) FROM TD_MAKKA NMIN WHERE N.URINO = NMIN.URINO AND NMIN.TORIKESHIFLG -1 ) ) NYUMIN ON (HD.URINO = NYUMIN.URINO) LEFT JOIN ( SELECT NSUM.URINO, SUM(COALESCE(NSUM.NYUKIN, 0)) NYUKIN, SUM(COALESCE(NSUM.NYUKIN, 0)) + SUM(COALESCE(NSUM.TESU, 0)) + SUM(COALESCE(NSUM.SOTA, 0)) SUMNYUKIN FROM TD_MAKKA NSUM GROUP BY URINO ) NYUSUM ON (HD.URINO = NYUSUM.URINO) LEFT JOIN ( SELECT N.URINO FROM TD_MAKKA N WHERE UPDATETIME = ( SELECT MAX(UPDATETIME) FROM TD_MAKKA NMAX WHERE N.URINO = NMAX.URINO AND NMAX.TORIKESHIFLG -1 ) ) NYUMAX ON (HD.URINO = NYUMAX.URINO) WHERE ((HD.URIBRUI '1') OR (HD.URIBRUI = '1' AND T_SQ.NYUKOBEFLG = '-1')) ORDER BY HD.URINO DESC ) COUNT_
这是EXPLAIN ANALYZE的结果
Aggregate (cost=7246861.21..7246861.22 rows=1 width=0) (actual time=69549.159..69549.159 rows=1 loops=1) -> Merge Left Join (cost=7240188.92..7242117.36 rows=379508 width=6) (actual time=68602.689..69510.563 rows=487252 loops=1) Merge Cond: (hd.urino = n.urino) -> Sort (cost=3727299.33..3728248.10 rows=379508 width=6) (actual time=62160.072..62557.132 rows=420036 loops=1) Sort Key: hd.urino Sort Method: external merge Disk: 6984kB -> Hash Right Join (cost=169264.26..3686940.26 rows=379508 width=6) (actual time=54796.930..60172.248 rows=420036 loops=1) Hash Cond: (n.urino = hd.urino) -> Seq Scan on td_makka n (cost=0.00..3511201.36 rows=209673 width=6) (actual time=24.326..4640.020 rows=419143 loops=1) Filter: (SubPlan 1) Rows Removed by Filter: 155 SubPlan 1 -> Aggregate (cost=8.33..8.34 rows=1 width=23) (actual time=0.009..0.009 rows=1 loops=419298) -> Index Scan using idx_td_makka on td_makka nmin (cost=0.00..8.33 rows=1 width=23) (actual time=0.006..0.007 rows=1 loops=419298) Index Cond: (n.urino = urino) Filter: (torikeshiflg (-1)::numeric) Rows Removed by Filter: 0 -> Hash (cost=163037.41..163037.41 rows=379508 width=6) (actual time=54771.078..54771.078 rows=386428 loops=1) Buckets: 4096 Batches: 16 Memory Usage: 737kB -> Hash Right Join (cost=75799.55..163037.41 rows=379508 width=6) (actual time=51599.167..54605.901 rows=386428 loops=1) Hash Cond: ((t_sq.urino = hd.urino) AND (t_sq.tcode = hd.sqcode)) Filter: ((hd.uribrui '1'::bpchar) OR ((hd.uribrui = '1'::bpchar) AND (t_sq.nyukobeflg = (-1)::numeric))) Rows Removed by Filter: 3344 -> Seq Scan on tv_snapshot t_sq (cost=0.00..73705.42 rows=385577 width=15) (actual time=0.053..2002.953 rows=389983 loops=1) Filter: ((delflg = 0::numeric) AND (syubetsu = 3::numeric)) Rows Removed by Filter: 1174773 -> Hash (cost=68048.99..68048.99 rows=389771 width=14) (actual time=51596.055..51596.055 rows=389772 loops=1) Buckets: 4096 Batches: 16 Memory Usage: 960kB -> Hash Right Join (cost=21125.85..68048.99 rows=389771 width=14) (actual time=579.405..51348.270 rows=389772 loops=1) Hash Cond: (nyusum.urino = hd.urino) -> Subquery Scan on nyusum (cost=0.00..35839.52 rows=365638 width=6) (actual time=17.435..49996.674 rows=385537 loops=1) -> GroupAggregate (cost=0.00..32183.14 rows=365638 width=34) (actual time=17.430..49871.702 rows=385537 loops=1) -> Index Scan using idx_td_makka on td_makka nsum (cost=0.00..21456.76 rows=419345 width=34) (actual time=0.017..48357.702 rows=419298 loops=1) -> Hash (cost=13969.71..13969.71 rows=389771 width=20) (actual time=491.549..491.549 rows=389772 loops=1) Buckets: 4096 Batches: 32 Memory Usage: 567kB -> Seq Scan on tv_hd hd (cost=0.00..13969.71 rows=389771 width=20) (actual time=0.052..242.415 rows=389772 loops=1) -> Sort (cost=3512889.60..3512894.84 rows=2097 width=6) (actual time=6442.600..6541.728 rows=486359 loops=1) Sort Key: n.urino Sort Method: external sort Disk: 8600kB -> Seq Scan on td_makka n (cost=0.00..3512773.90 rows=2097 width=6) (actual time=0.135..4053.116 rows=419143 loops=1) Filter: ((updatetime)::text = (SubPlan 2)) Rows Removed by Filter: 155 SubPlan 2 -> Aggregate (cost=8.33..8.34 rows=1 width=23) (actual time=0.008..0.008 rows=1 loops=419298) -> Index Scan using idx_td_makka on td_makka nmax (cost=0.00..8.33 rows=1 width=23) (actual time=0.005..0.006 rows=1 loops=419298) Index Cond: (n.urino = urino) Filter: (torikeshiflg (-1)::numeric) Rows Removed by Filter: 0 Total runtime: 69575.139 ms
这是解释分析结果的详细信息：
http://explain.depesz.com/s/23Fg
第一步：您可以删除选择查询中不需要的更多列，因为您只需计算总行数。 例如：
select count(*) from ( SELECT HD.URINO FROM TV_HD HD LEFT JOIN TV_SNAPSHOT T ON (HD.URINO = T.URINO AND HD.TCODE = T.TCODE AND T.DELFLG = 0 AND T.SYUBETSU = 1) LEFT JOIN TV_SNAPSHOT T_SQ ON (HD.URINO = T_SQ.URINO AND HD.SQCODE = T_SQ.TCODE AND T_SQ.DELFLG = 0 AND T_SQ.SYUBETSU = 3) LEFT JOIN (SELECT N.URINO FROM TD_MAKKA N WHERE N.UPDATETIME IN ( SELECT MIN (NMIN.UPDATETIME) FROM TD_MAKKA NMIN WHERE N.URINO = NMIN.URINO AND NMIN.TORIKESHIFLG <> -1 ) ) NYUMIN ON (HD.URINO = NYUMIN.URINO) LEFT JOIN ( SELECT NSUM.URINO ,SUM (COALESCE(NSUM.NYUKIN ,0)) NYUKIN ,SUM (COALESCE(NSUM.NYUKIN ,0)) + SUM (COALESCE(NSUM.TESU ,0)) + SUM (COALESCE(NSUM.SOTA ,0)) SUMNYUKIN FROM TD_MAKKA NSUM GROUP BY URINO ) NYUSUM ON (HD.URINO = NYUSUM.URINO) LEFT JOIN ( SELECT N.URINO FROM TD_MAKKA N WHERE UPDATETIME = ( SELECT MAX (UPDATETIME) FROM TD_MAKKA NMAX WHERE N.URINO = NMAX.URINO AND NMAX.TORIKESHIFLG <> -1 ) ) NYUMAX ON (HD.URINO = NYUMAX.URINO) WHERE ( (HD.URIBRUI <> '1') OR ( HD.URIBRUI = '1' AND T_SQ.NYUKOBEFLG = '-1' ) ) ORDER BY HD.URINO DESC ) COUNT_
第二步：您可以避免左连接，这对于获取行计数没有意义。 例如：
select count(*) from ( SELECT HD.URINO FROM TV_HD HD LEFT JOIN TV_SNAPSHOT T ON (HD.URINO = T.URINO AND HD.TCODE = T.TCODE AND T.DELFLG = 0 AND T.SYUBETSU = 1) LEFT JOIN TV_SNAPSHOT T_SQ ON (HD.URINO = T_SQ.URINO AND HD.SQCODE = T_SQ.TCODE AND T_SQ.DELFLG = 0 AND T_SQ.SYUBETSU = 3) LEFT JOIN (SELECT N.URINO FROM TD_MAKKA N WHERE N.UPDATETIME IN ( SELECT MIN (NMIN.UPDATETIME) FROM TD_MAKKA NMIN WHERE N.URINO = NMIN.URINO AND NMIN.TORIKESHIFLG <> -1 ) ) NYUMIN ON (HD.URINO = NYUMIN.URINO) LEFT JOIN ( SELECT N.URINO FROM TD_MAKKA N WHERE UPDATETIME = ( SELECT MAX (UPDATETIME) FROM TD_MAKKA NMAX WHERE N.URINO = NMAX.URINO AND NMAX.TORIKESHIFLG <> -1 ) ) NYUMAX ON (HD.URINO = NYUMAX.URINO) WHERE ( (HD.URIBRUI <> '1') OR ( HD.URIBRUI = '1' AND T_SQ.NYUKOBEFLG = '-1' ) ) ) COUNT_
第三步：您可以使用PgAdmin图形解释计划来分析查询并避免其他不必要的执行开销。
根据查询：
这里的实际需求是计算从内部sql找到的所有记录。
统计所有记录的优化理论：
- 删除SELECT查询中不必要的字段
- 删除ORDER BY ASC / DES部分（节省7％ – 10％）
- 删除聚合函数（平均值，总和，计数等）
- 使用标准VACCUUM回收死元组占用的存储空间。
- 研究来自http://explain.depesz.com/的“ EXPLAIN ANALYZE [your_query_here] ”结果
解释1：删除SELECT查询中不必要的字段
select count(*) from ( SELECT HD.URINO /*HD.URIBRUI, HD.TCODE, HD.SQCODE*/ FROM TV_HD HD)
解释2：删除ORDER BY ASC / DES部分（节省7％ – 10％）
select count(*) from ( SELECT HD.URINO FROM TV_HD HD /*ORDER BY HD.URINO DESC*/)
解释3：删除聚合函数（平均值，总和，计数等）
select count(*) from ( SELECT name /*MAX(salary), AVG(salary)*/ FROM Emp)
解释4：使用标准VACCUUM回收死元组占用的存储空间。
VACUUM (VERBOSE, ANALYZE) your_table;
在正常的PostgreSQL操作中，更新删除或废弃的元组不会从其表中物理删除; 它们一直存在，直到VACUUM完成。 因此，有必要periodically执行VACUUM， especially在频繁更新的表上。
VACUUM有两种变体： standard VACUUM和VACUUM FULL 。
VACUUM FULL可以回收更多的磁盘空间，但运行速度要慢得多。 此外，VACUUM的标准forms可以与生产数据库操作并行运行。 （SELECT，INSERT，UPDATE和DELETE等命令将继续正常运行，但在使用ALTER TABLE时，您将无法使用ALTER TABLE等命令修改表的定义。）VACUUM FULL需要独占锁定它正在处理的表，因此不能与表的其他使用并行完成。
因此，一般情况下， 管理员应努力使用standard VACUUM并avoid VACUUM FULL 。
详情如下：
- http://www.postgresql.org/docs/9.1/static/sql-vacuum.html
- http://www.postgresql.org/docs/9.1/static/routine-vacuuming.html
谢谢你的时间。