[IMPALA-219] Bad performance on tpch q4 on 10 node cluster

• Type: Bug
• Status: Resolved
• Priority: Blocker
• Resolution: Fixed
• Affects Version/s: Impala 0.7
• Fix Version/s: Impala 0.7
• Assignee: Nong Li
• Reporter: Nong Li
• Created: 10/Apr/13 01:16
• Updated: 12/Apr/13 23:39
• Resolved: 12/Apr/13 23:39

Description

Running the tpchq4 query on the 10 node cluster results in terrible performance (it is fine on the 17 node so something about the key distribution).

     HASH_JOIN_NODE (id=2):(24m27s 88.96%)
         - BuildBuckets: 1.02K (1024)                   <--- Few build buckets
         - BuildRows: 573.38K (573377)                  <--- Lots of keys, indicating they have all collided on the same bucket
         - BuildTime: 73.118ms
         - MemoryUsed: 0.00 
         - ProbeRows: 37.94M (37935647)
         - ProbeTime: 22m16s                            <--- Ridiculous amount of time on the probe side, indicating we are spending a lot of time looking through a long chained bucket.
         - RowsReturned: 1.45M (1449806)
         - RowsReturnedRate: 987.00 /sec