Question

我有一个不稳定的pl / pgsql函数。这是来源：

DECLARE
l RECORD;
events_for_machine integer;
before_event "PRD".events_log;
machines_ids integer[];
island_controller RECORD;
before_order "PRD".events_log;
before_detail "PRD".events_log;
before_pallete "PRD".events_log;
before_operation "PRD".events_log;
timer timestamp;
timer2 timestamp;

BEGIN
machines_ids = string_to_array(machines_ids_g,',')::integer[];
for l in 
select m.*
from
    "PRD".machines m
    inner join
    unnest(machines_ids) n(id) on n.id = m.id
where
    m.start_work_date < begin_date_g
order by m.id
LOOP

SELECT * INTO island_controller FROM "STRUCT".island_machines WHERE machine_id=l.id;

RAISE NOTICE 'pobieram zdarzenie before dla maszyny %',l.id;

    SELECT * INTO before_event FROM "PRD".events_log WHERE plc_time < begin_date_g AND (((event_type_id IN (1,51) AND machine_id = island_controller.controller_id AND island_id = island_controller.island_id))
    OR (event_type_id IN (2000,2001) AND machine_id=l.id)) ORDER BY plc_time DESC LIMIT 1;

IF before_event.plc_time IS NOT NULL THEN    

RAISE NOTICE 'Getting info about first machine work time struct element';
RETURN QUERY SELECT * FROM "PRD".events_log WHERE event_type_id = 113 AND machine_id=l.id AND plc_time < before_event.plc_time ORDER BY plc_time DESC LIMIT 1;
RETURN QUERY SELECT * FROM "PRD".events_log WHERE event_type_id = 102 AND machine_id=l.id AND plc_time < before_event.plc_time ORDER BY plc_time DESC LIMIT 1;
RETURN QUERY SELECT * FROM "PRD".events_log WHERE event_type_id = 111 AND machine_id=l.id AND plc_time < before_event.plc_time ORDER BY plc_time DESC LIMIT 1;
RETURN QUERY SELECT * FROM "PRD".events_log WHERE event_type_id = 1010 AND machine_id=l.id AND plc_time < before_event.plc_time ORDER BY plc_time DESC LIMIT 1;
RETURN NEXT before_event;


END IF;
RAISE NOTICE 'generuje zdarzenia wlasciwe dla maszyny %',l.id;
    RETURN QUERY SELECT * FROM "PRD".events_log WHERE 
            (event_type_id = ANY ('{1,51}'::integer[]) AND (machine_id=island_controller.controller_id AND island_id = island_controller.island_id) AND (plc_time BETWEEN begin_date_g AND end_date_g))
            OR (event_type_id = ANY ('{2000,2001,107}'::integer[]) AND machine_id=l.id AND (plc_time >= begin_date_g AND plc_time <= end_date_g))
            OR ((event_type_id = ANY ('{101,102,103,301,1010}'::integer[]) OR ((event_type_id >= 5000) AND (event_type_id <= 5999))) AND machine_id=l.id AND plc_time >= begin_date_g AND plc_time <= end_date_g) ORDER BY plc_time;

RAISE NOTICE 'koniec dla maszyny %',l.id;
END LOOP;

END;

有时函数执行时间约为9秒，对于相同的参数，有时约为40秒。它取决于什么？
什么可能是如此低效？

Answer 1

不止一个问题。我不能说，出了什么问题，但是当它在缓存中找到必要的数据（postgresql，filesystem）时会很快，如果没有，则会很慢（并且取决于有多少循环）。

出了什么问题：

无用的加入

machines_ids = string_to_array(machines_ids_g,',')::integer[];
for l in 
select m.*
from
    "PRD".machines m
    inner join
    unnest(machines_ids) n(id) on n.id = m.id

你可以写

for l in
   select *
      from "PRD".machines m
     where m.id = ANY(string_to_array(machines_ids_g,',')::integer[])
loop

很多重复的查询

RETURN QUERY SELECT * FROM "PRD".events_log WHERE event_type_id = 113 AND machine_id=l.id AND plc_time < before_event.plc_time ORDER BY plc_time DESC LIMIT 1;
RETURN QUERY SELECT * FROM "PRD".events_log WHERE event_type_id = 102 AND machine_id=l.id AND plc_time < before_event.plc_time ORDER BY plc_time DESC LIMIT 1;
RETURN QUERY SELECT * FROM "PRD".events_log WHERE event_type_id = 111 AND machine_id=l.id AND plc_time < before_event.plc_time ORDER BY plc_time DESC LIMIT 1;
RETURN QUERY SELECT * FROM "PRD".events_log WHERE event_type_id = 1010 AND machine_id=l.id AND plc_time < before_event.plc_time ORDER BY plc_time DESC LIMIT 1;

可以替换为

RETURN QUERY SELECT *
                FROM "PRD".events
               WHERE (event_type_id, plc_time) = (SELECT max(plc_time), event_type_id
                                                     FROM "PRD".events
                                                    WHERE event_type_id IN (113,102,111,1010)
                                                    GROUP BY event_type_id);

可能是复合索引（event_type_id，plc_time）应该有帮助

注意：

图案：

FOR n IN SELECT
   SELECT .. WHERE ..=n
END FOR

当循环太多时，

可能会很慢，并且在可能的情况下可以更好地处理一个查询

pl / pgsql函数性能低下

1 个答案: