[HIVE-17209] ObjectCacheFactory should return null when tez shared object registry is not setup

• Type: Bug
• Status: Closed
• Priority: Minor
• Resolution: Fixed
• Fix Version/s: 3.0.0
• Hadoop Flags: Reviewed
• Assignee: Rajesh Balamohan Assign to me
• Reporter: Rajesh Balamohan
• Created: 31/Jul/17 01:48
• Updated: 22/May/18 23:59
• Resolved: 02/Aug/17 02:23

Description

HIVE-15269 introduced dynamic min/max bloom filter ("hive.tez.dynamic.semijoin.reduction=true"). This needs to access ObjectCache and in tez, ObjectCache can only be created by TezProcessor.

In the following case AM --> splits --> OrcInputFormat.pickStripes::evaluatePredicateMinMax --> DynamicValue.getLiteral --> objectCache access, AM ends up throwing lots of NPE since AM has not created ObjectCache.

Orc reader catches these exceptions, skips PPD and proceeds further. For e.g, in Q95 it ends up throwing ~30,000 NPE before completing split information.

ObjectCacheFactory should return null when tez shared object registry is not setup.