[DRILL-3830] Query with aggregate window functions returns possibly wrong results on large scale data

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Invalid
• Affects Version/s: 1.2.0
• Component/s: Execution - Relational Operators

Environment

10 Performance Nodes
DRILL_MAX_DIRECT_MEMORY=100g
DRILL_INIT_HEAP="8g"
DRILL_MAX_HEAP="8g"
planner.memory.query_max_memory_per_node bumped up to 20 GB
TPC-DS SF 1000 dataset (Parquet)

• Assignee: Abdel Hakim Deneche
• Reporter: Abhishek Girish
• Created: 23/Sep/15 21:47
• Updated: 25/Apr/16 22:17
• Resolved: 27/Sep/15 04:20

Description

Results returned by the following two queries slightly differ from those returned by Greenplum DB.

SELECT SUM(ss.ss_net_paid_inc_tax) OVER (PARTITION BY ss.ss_store_sk) FROM store_sales ss LIMIT 1;

SELECT SUM(ss.ss_net_paid_inc_tax) OVER (PARTITION BY ss.ss_store_sk ORDER BY ss.ss_store_sk) FROM store_sales ss LIMIT 2;

Drill:
9.653697131700665E9

Greenplum DB:
9.628946925860903E9

P.S. Both queries return same results

I was unable to reproduce this on smaller scale (tried SF 1). I'll attach plans from both systems.