如果我像这样指定一个键 data->'type',它会削弱查询的动态性。
我了解您在列data上定义了一个gin索引,如下所示:
CREATE INDEX ON documents USING GIN (data);
这个查询可用于索引:
EXPLAIN ANALYSE
SELECT "documents".*
FROM "documents"
WHERE data @> '{"type": "foo"}';
QUERY PLAN
------------------------------------------------------------------------------------------------------------------------------
Bitmap Heap Scan on documents (cost=30.32..857.00 rows=300 width=25) (actual time=0.639..0.640 rows=1 loops=1)
Recheck Cond: (data @> '{"type": "foo"}'::jsonb)
Heap Blocks: exact=1
-> Bitmap Index Scan on documents_data_idx (cost=0.00..30.25 rows=300 width=0) (actual time=0.581..0.581 rows=1 loops=1)
Index Cond: (data @> '{"type": "foo"}'::jsonb)
Planning time: 7.928 ms
Execution time: 0.841 ms
但对于这个不行:
EXPLAIN ANALYSE
SELECT "documents".*
FROM "documents"
WHERE (data->'type' ?| array['foo', 'bar']);
QUERY PLAN
Seq Scan on documents (cost=0.00..6702.98 rows=300 width=25) (actual time=31.895..92.813 rows=2 loops=1)
Filter: ((data -> 'type'::text) ?| '{foo,bar}'::text[])
Rows Removed by Filter: 299997
Planning time: 1.836 ms
Execution time: 92.839 ms
解决方案1:使用操作符
@>
两次,索引将用于两个条件:
EXPLAIN ANALYSE
SELECT "documents".*
FROM "documents"
WHERE data @> '{"type": "foo"}'
OR data @> '{"type": "bar"}';
QUERY PLAN
------------------------------------------------------------------------------------------------------------------------------------
Bitmap Heap Scan on documents (cost=60.80..1408.13 rows=600 width=25) (actual time=0.222..0.233 rows=2 loops=1)
Recheck Cond: ((data @> '{"type": "foo"}'::jsonb) OR (data @> '{"type": "bar"}'::jsonb))
Heap Blocks: exact=2
-> BitmapOr (cost=60.80..60.80 rows=600 width=0) (actual time=0.204..0.204 rows=0 loops=1)
-> Bitmap Index Scan on documents_data_idx (cost=0.00..30.25 rows=300 width=0) (actual time=0.144..0.144 rows=1 loops=1)
Index Cond: (data @> '{"type": "foo"}'::jsonb)
-> Bitmap Index Scan on documents_data_idx (cost=0.00..30.25 rows=300 width=0) (actual time=0.059..0.059 rows=1 loops=1)
Index Cond: (data @> '{"type": "bar"}'::jsonb)
Planning time: 3.170 ms
Execution time: 0.289 ms
解决方案2. 在
(data->'type')
上创建一个额外的索引:
CREATE INDEX ON documents USING GIN ((data->'type'));
EXPLAIN ANALYSE
SELECT "documents".*
FROM "documents"
WHERE (data->'type' ?| array['foo', 'bar']);
QUERY PLAN
------------------------------------------------------------------------------------------------------------------------------
Bitmap Heap Scan on documents (cost=30.32..857.75 rows=300 width=25) (actual time=0.056..0.067 rows=2 loops=1)
Recheck Cond: ((data -> 'type'::text) ?| '{foo,bar}'::text[])
Heap Blocks: exact=2
-> Bitmap Index Scan on documents_expr_idx (cost=0.00..30.25 rows=300 width=0) (actual time=0.035..0.035 rows=2 loops=1)
Index Cond: ((data -> 'type'::text) ?| '{foo,bar}'::text[])
Planning time: 2.951 ms
Execution time: 0.108 ms
解决方案3。实际上这是解决方案1的一个变体,使用了不同格式的条件语句,这可能更方便客户程序使用:
EXPLAIN ANALYSE
SELECT "documents".*
FROM "documents"
WHERE data @> any(array['{"type": "foo"}', '{"type": "bar"}']::jsonb[]);
QUERY PLAN
------------------------------------------------------------------------------------------------------------------------------
Bitmap Heap Scan on documents (cost=60.65..1544.20 rows=600 width=26) (actual time=0.803..0.819 rows=2 loops=1)
Recheck Cond: (data @> ANY ('{"{\"type\": \"foo\"}","{\"type\": \"bar\"}"}'::jsonb[]))
Heap Blocks: exact=2
-> Bitmap Index Scan on documents_data_idx (cost=0.00..60.50 rows=600 width=0) (actual time=0.778..0.778 rows=2 loops=1)
Index Cond: (data @> ANY ('{"{\"type\": \"foo\"}","{\"type\": \"bar\"}"}'::jsonb[]))
Planning time: 2.080 ms
Execution time: 0.304 ms
(7 rows)
阅读文档以了解更多相关IT技术。