[IMPALA-2327] Query completed without error but incorrectly returned no results

• Type: Bug
• Status: Resolved
• Priority: Blocker
• Resolution: Fixed
• Affects Version/s: Impala 2.3.0
• Labels: stress
• Environment: impalad version 2.3.0-cdh5-INTERNAL DEBUG (build acd68caf7dd931fbc3edf8965298b57297a82aa0); Built on Thu, 10 Sep 2015 22:51:08 PST
• Target Version: Impala 2.3.0
• Assignee: Skye Wanderman-Milne
• Reporter: casey
• Created: 12/Sep/15 00:13
• Updated: 20/Dec/15 00:05
• Resolved: 07/Oct/15 00:05

Description

This has come up several times in the stress test. I'm still investigating but I'm pretty sure this is a bug. Some times a query will complete successfully with either (1) no results or (2) an incorrect result set. I suspect these may be the same issue where a partial result set is returned and #1 is a special case of #2.

From the profile archive

Query State FINISHED
Query Status OK

Operator              #Hosts   Avg Time   Max Time  #Rows  Est. #Rows   Peak Mem  Est. Peak Mem  Detail
--------------------------------------------------------------------------------------------------------------------------------
17:MERGING-EXCHANGE        1   12.372ms   12.372ms      0         100          0        -1.00 B  UNPARTITIONED
11:TOP-N                   4    65.30ms  102.316ms      0         100   52.00 KB       15.32 KB
10:SELECT                  4   168.58us  183.383us      0     171.11K   20.00 KB              0
09:ANALYTIC                4  129.650ms  166.328ms      0       1.71M    8.02 MB              0
08:SORT                    4  319.956us  355.990us      0       1.71M   24.01 MB       64.00 MB
16:AGGREGATE               4    9s951ms   13s251ms      0       1.71M    3.26 MB      122.01 MB  FINALIZE
15:EXCHANGE                4   17.852us   18.716us      0       1.71M          0              0  HASH(i_category,i_brand,s_s...
07:AGGREGATE               4   14s260ms   17s713ms      0       1.71M   16.30 MB      252.88 MB
06:HASH JOIN               4  795.488ms    1s016ms      0       1.71M   13.02 MB        1.76 KB  INNER JOIN, BROADCAST
|--14:EXCHANGE             4   19.917us   21.137us      0          32          0              0  BROADCAST
|  03:SCAN HDFS            1   44s726ms   44s726ms      0          32   86.00 KB       48.00 MB  tpcds_3_decimal_parquet.store
05:HASH JOIN               4  847.780ms   942.83ms  2.85K       1.66M   15.01 MB        71.00 B  INNER JOIN, BROADCAST
|--13:EXCHANGE             4  125.696us  127.781us    366           4          0              0  BROADCAST
|  02:SCAN HDFS            1   47s818ms   47s818ms    366           4    1.80 MB       48.00 MB  tpcds_3_decimal_parquet.dat...
04:HASH JOIN               4    1s015ms    2s301ms  2.85K       1.66M    3.11 MB       54.09 KB  INNER JOIN, BROADCAST
|--12:EXCHANGE             4    2.817ms    9.445ms  2.67K         587          0              0  BROADCAST
|  00:SCAN HDFS            1   27s847ms   27s847ms  2.67K         587    1.02 MB       64.00 MB  tpcds_3_decimal_parquet.item
01:SCAN HDFS               4       3m5s         6m  8.17K       1.66M  450.28 KB       48.00 MB  tpcds_3_decimal_parquet.sto...

The query should return 100 rows but returned none.

The query id is 7046b44020e587a1:8075419ca9e03797. A log grep and serialized profile are attached. The full log file was too big to upload (17mb compressed). The cluster is still up. The coordinator is impala-stress-4.vpc.cloudera.com.

Issue Links

• is duplicated by: IMPALA-1933 Scanner sometimes returns wrong error in very low memory
• is related to: IMPALA-1933 Scanner sometimes returns wrong error in very low memory