在 Postgres 中使用公共列进行索引

Question

. Answers to this question are eligible for a +50 reputation bounty. Nitika is looking for an .

Suppose I have 4 attributes in my Postgres db a,b,c,d.

I want to create an index on a and b, these are the two options I can go ahead with

1. indexing a and b separately
2. creating index as (c,a) and (c,b)

Will creating index in the 2nd form optimize the performance of db? And how does indexing work in that case?

Answer 1

An index can only be efficiently used by a query if a prefix of its columns is applied to the query's WHERE -condition. The order of the columns in the CREATE INDEX -statement matters. Suppose we have a table

CREATE TABLE my_table(
   a INT,
   b INT,
   c INT,
   d INT
);

and indexes

CREATE INDEX idx_a ON my_table (a);
CREATE INDEX idx_b ON my_table (b);
CREATE INDEX idx_ca ON my_table (c, a);
CREATE INDEX idx_cb ON my_table (c, b);

then query

SELECT * FROM my_table WHERE a = 42;

will only use idx_a (while all other indexes won't fit to this query),

SELECT * FROM my_table WHERE b = 42;

will only use idx_b (while all other indexes won't fit to this query),

SELECT * FROM my_table WHERE a = 42 and c = 23;

will use idx_ca ( idx_a would fit, too, but it has probably less performance),

SELECT * FROM my_table WHERE b = 42 and c = 23;

will use idx_cb ( idx_b would fit, too, but it has probably less performance) and

SELECT * FROM my_table WHERE c = 23;

can eather use idx_ca or idx_cb (the optimizer will decide by statistical information).

So it depends on your queries. If they always include a condition on c then idx_ca will presumably have better query performance in comparison to idx_a .

Of course you shouldn't create indexes on suspicion that actually aren't needed. In Postgres you can use EXPLAIN ( docs ) to see how the optimizer makes use of the indexes you have defined.

Answer 2

As already said, it really depends ot the queries on the table.

Postgres can in certain situation use several index for a query. See this very good answer: https://dba.stackexchange.com/a/91269/216879

If it is possible You can create the 4 indexes, and check after some times which of your index are not used by querying the pg_stat_user_indexes table.

More litterature about index utilization from the offical documentation: https://www.postgresql.org/docs/current/indexes-bitmap-scans.html#:~:text=Fortunately%2C%20PostgreSQL%20has%20the%20ability,conditions%20across%20several%20index%20scans .