Postgresql Postgres引擎在jsonb字段属性上的异常行为

嗨，我有这个关于postgres 12的问题

select
   count(*)
from
    modulo730.ods_t_record as r
left join(
    select
        *
    from
        modulo730.ods_t_ire as oti
    where
        cod_azienda = '080239'
        and not cod_scenario = 'CUPDT' ) B on
    r.cod_evento = B.cod_evento
where
  (r.uid_task = '41b7c8ea-4783-4f60-b3f2-980777b86693' )

 and  r.cod_azienda = '080239'
   and r.rec_aggiornato ->> 'annoDocumento' = '2020'

查询的第一个版本的行为与我预期的一样，它使用字段分区

cod\u azienda

它在

uid\u任务

字段和作为json一部分的

annoDocumento

字段上执行位图索引扫描 计数需要350毫秒 现在，如果我在json上添加另一个条件，那么我会发现最后一部分是这样的

  ...
        and r.rec_aggiornato ->> 'annoDocumento' = '2020'
        and ( r.rec_aggiornato ->> 'tipo_documento' = 'FT' )

db变得无法执行查询，甚至不能执行简单的返回。也就是说，只有一个条件（它也可以通过反转条件来工作），如果我将它们组合在一起，它会正确地执行它。。。繁荣

有什么建议吗？ 谢谢

这就是计划

Gather  (cost=34356.37..34356.48 rows=1 width=8)
          Workers Planned: 1
          Single Copy: true
          ->  Aggregate  (cost=33356.37..33356.38 rows=1 width=8)
                ->  Nested Loop Left Join  (cost=1410.54..33356.37 rows=1 width=0)
                      ->  Append  (cost=1410.54..18597.25 rows=1 width=18)
                            ->  Bitmap Heap Scan on ods_t_record_mch (cost=1410.54..18597.24 rows=1 width=18)
                                  Recheck Cond: (((cod_azienda)::text = '080239'::text) AND ((uid_task)::text = '41b7c8ea-4783-4f60-b3f2-980777b86693'::text))
                                  Filter: (((rec_aggiornato ->> 'annoDocumento'::text) = '2020'::text) AND ((rec_aggiornato ->> 'tipo_documento'::text) = 'FT'::text))
                                  ->  Bitmap Index Scan on ods_t_record_mch_cod_azienda_uid_task_idx  (cost=0.00..1410.54 rows=14612 width=0)
                                        Index Cond: (((cod_azienda)::text = '080239'::text) AND ((uid_task)::text = '41b7c8ea-4783-4f60-b3f2-980777b86693'::text))
                      ->  Append  (cost=0.00..14759.11 rows=1 width=18)
                            ->  Seq Scan on ods_t_ire_mch oti  (cost=0.00..14759.10 rows=1 width=18)
                                  Filter: (((cod_scenario)::text <> 'CUPDT'::text) AND ((cod_azienda)::text = '080239'::text) AND ((r.cod_evento)::text = (cod_evento)::text))

第二张桌子

CREATE TABLE modulo730.ods_t_ire (
    uid varchar(36) NOT NULL,
    inserter varchar(128) NULL,
    insert_time timestamp NULL,
    modifier varchar(128) NULL,
    modify_time timestamp NULL,
    uid_task varchar(36) NULL,
    cod_evento varchar(128) NOT NULL,
    cod_proprietario varchar(128) NULL,
    cod_azienda varchar(128) NOT NULL,
    tip_evento varchar(128) NOT NULL,
    cod_scenario varchar(128) NOT NULL,
    data_invio timestamp NULL,
    flg_inviato varchar(1) NULL,
    flg_ritorno varchar(1) NULL,
    rec_inviato_ori jsonb NULL,
    rec_inviato text NULL,
    rec_ritorno text NULL,
    uid_sent_file varchar(36) NULL,
    flow_state varchar(50) NULL,
    cod_error varchar NULL,
    cod_evento_inviato varchar(128) NULL,
    dat_evento timestamp NULL,
    cod_evento_collegato varchar NULL,
    dat_evento_collegato timestamp NULL,
    ticket varchar NULL,
    flg_cancellazione varchar NULL,
    ticket_cancellazione varchar NULL,
    cod_destinatario varchar NOT NULL,
    CONSTRAINT ods_t_ire_pk PRIMARY KEY (cod_azienda, tip_evento, cod_scenario, cod_destinatario, cod_evento)
)
PARTITION BY LIST (cod_azienda);
CREATE UNIQUE INDEX ods_t_ire_cod_azienda_idx ON ONLY modulo730.ods_t_ire USING btree (cod_azienda, uid);

下面进一步指出，仅考虑json属性的一个条件的普通属性是基数较低的属性。 我不是专家，但我看到的是，在更快的情况下，它执行某种并行计算

Finalize Aggregate  (cost=33423.14..33423.15 rows=1 width=8)
  ->  Gather  (cost=33422.93..33423.14 rows=2 width=8)
        Workers Planned: 2
        ->  Partial Aggregate  (cost=32422.93..32422.94 rows=1 width=8)
              ->  Parallel Hash Left Join  (cost=15479.31..32422.85 rows=30 width=0)
                    Hash Cond: ((r.cod_evento)::text = (oti.cod_evento)::text)
                    ->  Parallel Append  (cost=1410.56..18353.88 rows=30 width=18)
                          ->  Parallel Bitmap Heap Scan on ods_t_record_mch r  (cost=1410.56..18353.73 rows=30 width=18)
                                Recheck Cond: (((cod_azienda)::text = '080239'::text) AND ((uid_task)::text = '41b7c8ea-4783-4f60-b3f2-980777b86693'::text))
                                Filter: ((rec_aggiornato ->> 'tipo_documento'::text) = 'FT'::text)
                                ->  Bitmap Index Scan on ods_t_record_mch_cod_azienda_uid_task_idx  (cost=0.00..1410.54 rows=14612 width=0)
                                      Index Cond: (((cod_azienda)::text = '080239'::text) AND ((uid_task)::text = '41b7c8ea-4783-4f60-b3f2-980777b86693'::text))
                    ->  Parallel Hash  (cost=13160.38..13160.38 rows=72669 width=18)
                          ->  Parallel Append  (cost=0.00..13160.38 rows=72669 width=18)
                                ->  Parallel Seq Scan on ods_t_ire_mch oti  (cost=0.00..12797.04 rows=72669 width=18)
                                      Filter: (((cod_scenario)::text <> 'CUPDT'::text) AND ((cod_azienda)::text = '080239'::text))

最终确定聚合（成本=33423.14..33423.15行=1宽度=8）
->聚集（成本=33422.93..33423.14行=2宽度=8）
计划人数：2人
->部分骨料（成本=32422.93..32422.94行=1宽=8）
->并行哈希左连接（成本=15479.31..32422.85行=30宽度=0）
哈希条件：（（r.cod\u evento）：：text=（oti.cod\u evento）：：text）
->并行追加（成本=1410.56..18353.88行=30宽度=18）
->ods_t_记录上的并行位图堆扫描（成本=1410.56..18353.73行=30宽度=18）
重新检查条件：（（cod_azienda:：text='080239'：：text）和（（uid_任务）：：text='41b7c8ea-4783-4f60-b3f2-9807777b86693'：：text））
筛选器：（（rec_aggiornato->>'tipo_documento'：：text）='FT'：：text）
->ods上的位图索引扫描\u t\u记录\u mch\u cod\u azienda\u uid\u任务\u idx（成本=0.00..1410.54行=14612宽度=0）
索引条件：（（cod_azienda:：text='080239'：：text）和（（uid_任务）：：text='41b7c8ea-4783-4f60-b3f2-9807777b86693'：：text））
->并行散列（成本=13160.38..13160.38行=72669宽度=18）
->并行追加（成本=0.00..13160.38行=72669宽度=18）
->ods上的并行顺序扫描（成本=0.00..12797.04行=72669宽度=18）
筛选器：（（cod_场景）：：text'CUPDT'：：text）和（（cod_azienda:：text='080239'：：text））

回答问题并提供答案。添加表的

CREATE

语句及其索引的

CREATE

语句。将示例数据添加为

INSERT

语句。添加预期结果。并标记您正在使用的Postgres版本。

Finalize Aggregate  (cost=33423.14..33423.15 rows=1 width=8)
  ->  Gather  (cost=33422.93..33423.14 rows=2 width=8)
        Workers Planned: 2
        ->  Partial Aggregate  (cost=32422.93..32422.94 rows=1 width=8)
              ->  Parallel Hash Left Join  (cost=15479.31..32422.85 rows=30 width=0)
                    Hash Cond: ((r.cod_evento)::text = (oti.cod_evento)::text)
                    ->  Parallel Append  (cost=1410.56..18353.88 rows=30 width=18)
                          ->  Parallel Bitmap Heap Scan on ods_t_record_mch r  (cost=1410.56..18353.73 rows=30 width=18)
                                Recheck Cond: (((cod_azienda)::text = '080239'::text) AND ((uid_task)::text = '41b7c8ea-4783-4f60-b3f2-980777b86693'::text))
                                Filter: ((rec_aggiornato ->> 'tipo_documento'::text) = 'FT'::text)
                                ->  Bitmap Index Scan on ods_t_record_mch_cod_azienda_uid_task_idx  (cost=0.00..1410.54 rows=14612 width=0)
                                      Index Cond: (((cod_azienda)::text = '080239'::text) AND ((uid_task)::text = '41b7c8ea-4783-4f60-b3f2-980777b86693'::text))
                    ->  Parallel Hash  (cost=13160.38..13160.38 rows=72669 width=18)
                          ->  Parallel Append  (cost=0.00..13160.38 rows=72669 width=18)
                                ->  Parallel Seq Scan on ods_t_ire_mch oti  (cost=0.00..12797.04 rows=72669 width=18)
                                      Filter: (((cod_scenario)::text <> 'CUPDT'::text) AND ((cod_azienda)::text = '080239'::text))