[IMPALA-3678] Preconditions check in SortNode when planning query that has union operands with order by + limit

• Type: Bug
• Status: Resolved
• Priority: Blocker
• Resolution: Fixed
• Affects Version/s: Impala 2.5.0
• Fix Version/s: Impala 2.7.0, Impala 2.6.2
• Component/s: Frontend
• Target Version: Impala 2.7.0, Impala 2.6.2, Impala 2.5.4
• Assignee: Alexander Behm
• Reporter: fishing
• Created: 05/Jun/16 12:07
• Updated: 16/Sep/16 15:53
• Resolved: 15/Jul/16 15:21

Description

Predicates from enclosing query blocks are not properly pushed into union operands that have an order by + limit.
This problem manifests itself as a failed Preconditions check because the SortNode incorrectly picks up such predicates.

Repro:

select * from
  ((select * from functional.alltypes order by id limit 10)
    union all
   (select * from functional.alltypes order by id limit 10)) v
where v.id < 10
ERROR: IllegalStateException: null

The predicate should be placed into a SelectNode in the union branche(s) that have an order by + limit.

Original JIRA description
SELECT xxxxxx.brand_id, cn_name FROM ( (SELECT brand_id FROM (SELECT brand_id, sum(pv) AS g FROM flow_minute_pv GROUP BY brand_id) xx ORDER BY g DESC LIMIT 15) UNION (SELECT brand_id FROM (SELECT brand_id, sum(CASE WHEN gmv IS NULL THEN 0 ELSE gmv END) AS g FROM flow_minute_sku_gmv GROUP BY brand_id) xx ORDER BY g DESC LIMIT 15)) xxxxxx LEFT JOIN brands b ON xxxxxx.brand_id = b.brand_id WHERE xxxxxx.brand_id > 0

ERROR: IllegalStateException: null in SortNode.java 93 行

with out left join query is normal
like this :

(SELECT brand_id
FROM
(SELECT brand_id,
sum(pv) AS g
FROM flow_minute_pv
GROUP BY brand_id) xx
ORDER BY g DESC LIMIT 15 )
UNION
( SELECT brand_id
FROM
(SELECT brand_id,
sum(CASE WHEN gmv IS NULL THEN 0 ELSE gmv END) AS g
FROM flow_minute_sku_gmv
GROUP BY brand_id) xx
ORDER BY g DESC LIMIT 15 )