在子查询中使用 unnest() 时，Postgresql 分区修剪不起作用

Question

在子查询中使用 unnest() 时，Postgresql (13.4) 无法提出使用执行时分区 p运行ing 的查询计划。

鉴于这些 table：

CREATE TABLE users (
    user_id uuid, 
    channel_id uuid, 
    CONSTRAINT user_pk PRIMARY KEY(user_id, channel_id)
) 
PARTITION BY hash(user_id);
CREATE TABLE users_0 PARTITION OF users FOR VALUES WITH (MODULUS 2, REMAINDER 0);
CREATE TABLE users_1 PARTITION OF users FOR VALUES WITH (MODULUS 2, REMAINDER 1);
    
CREATE TABLE channels (
    channel_id uuid, 
    user_ids uuid[],
    CONSTRAINT channel_pk PRIMARY KEY(channel_id)
) PARTITION BY hash(channel_id);

CREATE TABLE channels_0 partition of channels FOR VALUES WITH (MODULUS 2, REMAINDER 0);
CREATE TABLE channels_1 partition of channels FOR VALUES WITH (MODULUS 2, REMAINDER 1);

插入一些数据：

INSERT INTO users(user_id, channel_id) VALUES('0861180b-c972-42fe-9fb3-3b55e652f893', '45205876-7270-4e06-ab8d-b5f669298422');
INSERT INTO channels(channel_id, user_ids) VALUES('45205876-7270-4e06-ab8d-b5f669298422', '{0861180b-c972-42fe-9fb3-3b55e652f893}');

INSERT INTO users 
SELECT 
    gen_random_uuid() as user_id,
    gen_random_uuid() as channel_id
FROM generate_series(1, 100);

INSERT INTO channels
SELECT
    (SELECT max(channel_id::text) FROM (SELECT channel_id FROM users ORDER BY random()*generate_series LIMIT 1) c)::uuid as channel_id,
    (SELECT array_agg(DISTINCT user_id::text) FROM (SELECT user_id FROM users ORDER BY random()*generate_series 
    LIMIT 1) u)::uuid[] as user_ids
FROM (SELECT * FROM generate_series(1, 100)) g
ON conflict DO NOTHING;

以下查询：

EXPLAIN ANALYZE
SELECT * FROM users
WHERE user_id IN (
    SELECT unnest(user_ids) FROM channels WHERE channel_id = '45205876-7270-4e06-ab8d-b5f669298422'
)
AND channel_id = '45205876-7270-4e06-ab8d-b5f669298422'

returns 扫描所有分区的查询计划。

Hash Semi Join  (cost=8.45..37.28 rows=8 width=32) (actual time=0.208..0.387 rows=1 loops=1)
  Hash Cond: (users.user_id = (unnest(channels.user_ids)))
  ->  Append  (cost=0.00..28.71 rows=8 width=32) (actual time=0.037..0.134 rows=1 loops=1)
        ->  Seq Scan on users_0 users_1  (cost=0.00..27.00 rows=7 width=32) (actual time=0.021..0.041 rows=1 loops=1)
              Filter: (channel_id = '45205876-7270-4e06-ab8d-b5f669298422'::uuid)
              Rows Removed by Filter: 45
        ->  Seq Scan on users_1 users_2  (cost=0.00..1.68 rows=1 width=32) (actual time=0.018..0.027 rows=0 loops=1)
              Filter: (channel_id = '45205876-7270-4e06-ab8d-b5f669298422'::uuid)
              Rows Removed by Filter: 54
  ->  Hash  (cost=8.33..8.33 rows=10 width=16) (actual time=0.131..0.172 rows=1 loops=1)
        Buckets: 1024  Batches: 1  Memory Usage: 9kB
        ->  ProjectSet  (cost=0.15..8.23 rows=10 width=16) (actual time=0.060..0.114 rows=1 loops=1)
              ->  Index Scan using channels_0_pkey on channels_0 channels  (cost=0.15..8.17 rows=1 width=32) (actual time=0.040..0.059 rows=1 loops=1)
                    Index Cond: (channel_id = '45205876-7270-4e06-ab8d-b5f669298422'::uuid)
Planning Time: 0.363 ms
Execution Time: 0.515 ms

我希望 Postgresql 运行子查询并查看返回的 user_id 以确定该数据将位于哪些分区。但是，Postgresql 正在为此调查所有分区数据。我试过在频道 table 中使用一行 pr user_id，效果很好。

EXPLAIN ANALYZE
SELECT * FROM users
WHERE user_id IN (
    SELECT user_id FROM channels WHERE channel_id = '45205876-7270-4e06-ab8d-b5f669298422'
)
AND channel_id = '45205876-7270-4e06-ab8d-b5f669298422'

Postgresql 然后不会运行任何无法保存任何数据的分区的步骤。

似乎 unnest() 导致执行时间分区 p运行ing 不起作用。这是为什么？

解决方案：我可以确认 jjanes 的解决方案。通过向 tables 添加 100k 行，使用 unnest() 的查询在执行时进行分区 p运行ing。

Answer 1

您的示例仅显示了它在一个查询中使用的内容，而不是它能够执行的所有操作。

你的 table 小得可笑。再往users里面放10000行table，这样索引其实很重要，看看有什么作用

                                                                   QUERY PLAN                                                                   
------------------------------------------------------------------------------------------------------------------------------------------------
 Nested Loop  (cost=1.85..151.99 rows=2 width=32) (actual time=0.040..0.042 rows=1 loops=1)
   ->  HashAggregate  (cost=1.57..1.67 rows=10 width=16) (actual time=0.022..0.023 rows=1 loops=1)
         Group Key: unnest(channels.user_ids)
         Batches: 1  Memory Usage: 24kB
         ->  ProjectSet  (cost=0.00..1.44 rows=10 width=16) (actual time=0.016..0.019 rows=1 loops=1)
               ->  Seq Scan on channels_0 channels  (cost=0.00..1.39 rows=1 width=37) (actual time=0.013..0.016 rows=1 loops=1)
                     Filter: (channel_id = '45205876-7270-4e06-ab8d-b5f669298422'::uuid)
                     Rows Removed by Filter: 30
   ->  Append  (cost=0.28..15.01 rows=2 width=32) (actual time=0.016..0.017 rows=1 loops=1)
         ->  Index Only Scan using users_0_pkey on users_0 users_1  (cost=0.28..7.50 rows=1 width=32) (actual time=0.014..0.015 rows=1 loops=1)
               Index Cond: ((user_id = (unnest(channels.user_ids))) AND (channel_id = '45205876-7270-4e06-ab8d-b5f669298422'::uuid))
               Heap Fetches: 1
         ->  Index Only Scan using users_1_pkey on users_1 users_2  (cost=0.28..7.50 rows=1 width=32) (never executed)
               Index Cond: ((user_id = (unnest(channels.user_ids))) AND (channel_id = '45205876-7270-4e06-ab8d-b5f669298422'::uuid))
               Heap Fetches: 0
 Planning Time: 0.470 ms
 Execution Time: 0.087 ms

(never executed) 是由于 execution-time 修剪。

在子查询中使用 unnest() 时，Postgresql 分区修剪不起作用

Postgresql partition pruning not working when using unnest() in a subquery

postgresql

subquery

database-partitioning