如何优化 SQL SELECT 查询以生成用户的新闻源？

Question

I'm currently trying to build a feature to generate a user's newsfeed using the following query from a table of posts . This is the SQL statement we are using:

SELECT *
FROM "posts" AS "post"
WHERE "post"."sourceId" IN (...)
ORDER BY "post"."createdAt" DESC, "post"."timestamp" DESC
LIMIT 10;

The posts table currently has roughly 200K+ rows and likely to grow much larger. My skills in DB performance isn't the strongest, but is there anyway to optimise this query to make it run as fast as possible? I'm assuming it's not enough to add an index on the sourceId column but instead would need a multi column index to also take into account the ORDER BY column.

Answer 1

For this query:

SELECT p.*
FROM posts p
WHERE p.sourceId IN (...)
ORDER BY p.createdAt DESC, p.timestamp DESC
LIMIT 10;

The only index that can really help is an index on posts(sourceId) .

Note that I removed the " . Do not escape table and column names when you define them. Then you don't need to escape them when you use them.

However, the query still has to sort all the data. And that can be time-consuming. A more complicated query is easier for Postgres to optimize:

select p.*
from ((select p.*
       from posts p
       where sourceId = $si_1
       order by p.createdAt desc, p.timestamp desc
       limit 10
      ) union all
      (select p.*
       from posts p
       where sourceId = $si_2
       order by p.createdAt desc, p.timestamp desc
       limit 10
      ) union all
      . . .
    ) p
order by p.createdAt desc, p.timestamp desc;

This query can use an index on posts(sourceId, createdAt desc, timestamp desc) for the inner selects. That should be fast. the outer order by will still need sorting, but the volume of data should be much smaller.

For instance, if a typical source has 10,000 rows and you are only looking for 3 of them, then your version of the query needs to sort 30,000 rows to fetch 10. This version fetches 30 rows uses the index and then sorts them to get the final 10.

That would be a big difference in performance.

Answer 2

You may find that just an index on sourceId is sufficient:

CREATE INDEX src_idx ON posts (sourceId);

Postgres would then manually have to sort the records which make it past the WHERE clause. Further adding the columns in the ORDER BY clause might also help:

CREATE INDEX idx ON posts (sourceId, createdAt DESC, timestamp DESC);

This might speed up the sorting operation by letting Postgres sort the matching groups of sourceId records at once.