0

我对 SQL 还很陌生，我正在尝试研究如何在 postgres 中加速复杂的 SQL 查询，也许是通过改进我对索引的使用。这是查询：

SELECT
    (SELECT ev.code FROM classification_item ci, enumeration_value ev
        WHERE ev.key_id = :ak_0
        AND ci.entry_id = t.id AND ci.value_id = ev.id) AS axis_0,
    (SELECT ev.code FROM classification_item ci, enumeration_value ev
        WHERE ev.key_id = :ak_1
        AND ci.entry_id = t.id AND ci.value_id = ev.id) AS axis_1,
    SUM(t.amount) as amount,
    (SELECT ev.code FROM classification_item ci, enumeration_value ev
        WHERE ev.key_id = :key_time_id
        AND ci.entry_id = t.id AND ci.value_id = ev.id) AS time
FROM "entry" t
WHERE t.dataset_id = :dataset_id
AND t.id IN (SELECT ci.entry_id FROM classification_item ci, enumeration_value ev
    WHERE ev.key_id = :k_0
    AND ev.code = :v_0 AND ci.value_id = ev.id)
GROUP BY time, axis_0, axis_1

这基本上是数据库模式（在 Pylons 中定义）：

table_dataset = Table('dataset', meta.metadata,
    Column('id', Integer, primary_key=True),
    )
table_entry = Table('entry', meta.metadata,
    Column('id', Integer, primary_key=True),
    Column('dataset_id', Integer, ForeignKey('dataset.id')),
    Column('amount', Float()),
    )
table_classification_item = Table('classification_item', meta.metadata,
    Column('id', Integer, primary_key=True),
    Column('entry_id', Integer, ForeignKey('entry.id'), index=True),
    Column('value_id', Integer, ForeignKey('enumeration_value.id'), index=True)
)
table_enumeration_value = Table('enumeration_value', meta.metadata,
    Column('id', Integer, primary_key=True),
    Column('key_id', Integer, ForeignKey('key.id'), index=True),
    Column('code', UnicodeText(), index=True),
    )

它的索引如下：

"dataset_pkey" PRIMARY KEY, btree (id)
"entry_pkey" PRIMARY KEY, btree (id)
"classification_item_pkey" PRIMARY KEY, btree (id)
"ix_classification_item_entry_id" btree (entry_id)
"ix_classification_item_value_id" btree (value_id)
"enumeration_value_pkey" PRIMARY KEY, btree (id)
"ix_enumeration_value_code" btree (code)
"ix_enumeration_value_key_id" btree (key_id)

我是否缺少任何可以加快查询速度的明显索引？尤其是：

  • 我应该使用“聚集”索引吗？
  • 我也应该amount对进行索引entry，还是对 没有影响SUM(t.amount) as amount？

谢谢你的帮助。我知道这是一个非常复杂的问题，所以请告诉我是否可以做些什么来改进它。

- - - 更新 - - - - - - -

EXPLAIN ANALYZE对上述查询的输出。

4

2 回答 2

1

如果enumeration_value表很小，我想您可以通过将axis_1and axis_0as 加入并添加一个额外的索引来获得一些改进。

像这样的东西（未经测试）

CREATE INDEX idx_ci_vi_ei ON classification_item(value_id, entry_id);

CREATE INDEX idx_id_ki ON enumeration_value(id, key_id);

SELECT
    ci_0.code AS axis_0,
    ci_1.code AS axis_1,
    SUM(t.amount) as amount,
    ci_t.code AS time
FROM 
   "entry" t,
   (SELECT ev.code FROM classification_item ci, enumeration_value ev
        WHERE ev.key_id = :ak_0 AND ci.value_id = ev.id) ci_0,
   (SELECT ev.code FROM classification_item ci, enumeration_value ev
        WHERE ev.key_id = :ak_1 AND ci.value_id = ev.id) ci_1,
   (SELECT ev.code FROM classification_item ci, enumeration_value ev
        WHERE ev.key_id = :key_time_id AND ci.value_id = ev.id) ci_t
WHERE t.dataset_id = :dataset_id 
AND t.id IN (SELECT ci.entry_id FROM classification_item ci, enumeration_value ev
    WHERE ev.key_id = :k_0
    AND ev.code = :v_0 AND ci.value_id = ev.id)
AND t.id = ci_0.entry_id AND t.id = ci_1.entry_id AND t.id = ci_t.entry_id
GROUP BY time, axis_0, axis_1
于 2010-11-03T13:57:56.233 回答
0

EXPLAIN ANALYZE对查询计划有什么看法？

于 2010-11-03T13:47:11.277 回答