[IMPALA-4618] # hosts is inaccurate in exec summary when mt_dop > 0

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: Impala 2.8.0
• Fix Version/s: Impala 3.4.0
• Component/s: Backend
• Labels: multithreading ramp-up supportability
• Assignee: Norbert Luksa
• Reporter: Tim Armstrong
• Created: 07/Dec/16 22:13
• Updated: 19/May/20 19:35
• Resolved: 05/Dec/19 12:53

Description

I'm seeing this on a 3-node minicluster:

Operator       #Hosts   Avg Time   Max Time   #Rows  Est. #Rows   Peak Mem  Est. Peak Mem  Detail            
-------------------------------------------------------------------------------------------------------------
06:AGGREGATE        1  212.433us  212.433us       0           1    64.00 B        -1.00 B  FINALIZE          
05:EXCHANGE         1    0.000ns    0.000ns       0           1          0        -1.00 B  UNPARTITIONED     
02:AGGREGATE        9  131.439us  167.821us       0           1    64.00 B       10.00 MB                    
04:AGGREGATE        9    0.000ns    0.000ns       0          -1  183.42 MB      128.00 MB                    
03:EXCHANGE         9  707.426ms  823.681ms  20.97M          -1          0              0  HASH(`_c0`)       
01:AGGREGATE        9    5s222ms    5s355ms  22.89M          -1    3.83 MB      128.00 MB  STREAMING         
00:SCAN HDFS        9    2s623ms    2s896ms  23.33M          -1  492.02 KB       88.00 MB  default.big_uuids

It seems like it's reporting the # of fragment instances instead of the number of hosts.

We should probably report both #Hosts and #Instances if mt_dop is enabled (and maybe in all cases)

Issue Links

• is duplicated by: IMPALA-8993 Profile reports incorrect host number with mt_dop on joins