[IMPALA-2979] Simple scheduler : Reservoir sampling doesn't provide enough randomization

• Type: Bug
• Status: Resolved
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: Impala 2.5.0
• Fix Version/s: Impala 2.7.0
• Component/s: Backend
• Labels: performance scheduler
• Target Version: Impala 2.7.0
• Assignee: Lars Volker
• Reporter: Mostafa Mokhtar
• Created: 11/Feb/16 02:50
• Updated: 15/Mar/18 19:57
• Resolved: 17/Aug/16 15:27

Description

It appears with random_non_cached_tiebreak enabled the reservoir sampling doesn't provide enough randomization which can result in hot spots.

Issue Links

• is blocked by: IMPALA-2918 Increase unit test coverage of simple-scheduler.cc
• relates to: IMPALA-2924 Add query hints to adjust replica selection policy