如何减少优化的查询执行时间？

Question

您好，当前的查询执行时间过长，应进行哪些更改以减少执行时间？

EXPLAIN ANALYZE
SELECT  date_trunc({{period}}, wb.timestamp),
        CAST(COUNT(DISTINCT(org.id)) AS float) / CAST(COUNT(DISTINCT(wb.anonymous_id)) AS float) AS "marketing-lead conversion rate"
FROM website_prod.pages wb
FULL JOIN core_prod.organizations_organization org ON CAST(wb.timestamp AS DATE) = CAST(org.created_at AS DATE)
WHERE   NOT org.is_internal 
        AND wb.timestamp > date_trunc('month', CURRENT_DATE) - INTERVAL '1 year'
        AND org.created_at > date_trunc('month', CURRENT_DATE) - INTERVAL '1 year'
GROUP BY 1

QUERY PLAN

Sort Key: ((org.created_at)::date)
Sort Method: quicksort Memory: 608kB
-> Seq Scan on organizations_organization org (cost=0.00..259.02 rows=5080 width=19) (actual time=0.017..6.266 rows=5316 loops=1)
Filter: ((NOT is_internal) AND (created_at > (date_trunc('month'::text, (('now'::cstring)::date)::timestamp with time zone) - '1 year'::interval)))
Rows Removed by Filter: 1540
-> Materialize (cost=69882.44..71012.63 rows=226038 width=45) (actual time=541.950..849.952 rows=3917712 loops=1)
-> Sort (cost=69882.44..70447.54 rows=226038 width=45) (actual time=541.945..592.666 rows=232720 loops=1)
Sort Key: ((wb."timestamp")::date)
Sort Method: external merge Disk: 14104kB
-> Seq Scan on pages wb (cost=0.00..42826.15 rows=226038 width=45) (actual time=40.801..358.687 rows=232720 loops=1)
Filter: ("timestamp" > (date_trunc('month'::text, (('now'::cstring)::date)::timestamp with time zone) - '1 year'::interval))
Rows Removed by Filter: 72766
Planning time: 0.214 ms
Execution time: 26614.737 ms

Answer 1

你的加入在我看来非常可疑......

  website_prod.pages wb
FULL JOIN
  core_prod.organizations_organization org
    ON CAST(wb.timestamp AS DATE) = CAST(org.created_at AS DATE)

在这种情况下，您要将wb 中的所有行与org 中的所有行连接起来，前提是它们“用于”同一日期。

因此，例如，如果在特定日期（例如，2021-04-01），有 10 条 wb 记录和 10 条 org 记录，则连接将给出所有 100 种组合。

• 这真的是你想要的吗？？？

我会“期待”一些更像...的东西

  website_prod.pages wb
FULL JOIN
  core_prod.organizations_organization org
    ON  wb.some_key = org.some_key
    AND CAST(wb.timestamp AS DATE) = CAST(org.created_at AS DATE)

此外，为了使联接有效，该计划表明它必须按时间顺序对数据进行排序。

您没有具体说明存在哪些索引，但对于当前查询，您将从正在连接的列上的索引中受益...wb.timestamp 和org.created_at。不过，如果您按上述方式更改连接谓词，情况就会发生变化。