[IMPALA-7692] Profile averages not computed for Exchange InactiveTotalTime TIME_NS counter

• Type: Bug
• Status: Resolved
• Priority: Minor
• Resolution: Duplicate
• Affects Version/s: Impala 2.6.1
• Component/s: Backend
• Labels: observability
• Epic Color: ghx-label-6
• Assignee: Unassigned
• Reporter: Paul Rogers
• Created: 10/Oct/18 18:45
• Updated: 14/May/20 19:19
• Resolved: 14/May/20 19:19

Description

Ran a unit test against a query profile created from the following Impala version:

Impala Version: impalad version 2.6.0-cdh5.8.4 RELEASE (build 207450616f75adbe082a4c2e1145a2384da83fa6)

Then, did the math to compare each TCounter in the Fragment Average Fxx subtree with the average of the details in the Fragment Fxx/Instance xxx subtree. All matched except a few using the TUnit.TIME_NS counters. The average was always (sometimes?) zero. For example:

   EXCHANGE  [HASH(...)]: Incorrect average for InactiveTotalTime (TIME_NS), expected = 11605329272, actual = 0                           
  2     EXCHANGE  [BROADCAST]: Incorrect average for InactiveTotalTime (TIME_NS), expected = 11043724227, actual = 0

Have not checked this with the latest version.

Issue Links

• duplicates: IMPALA-2794 Exchange Inactive time in the averaged query profile is always zero