[IMPALA-4658] Potential race if compiler reorders ReachedLimit() usage

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: Impala 2.8.0
• Fix Version/s: Impala 3.3.0
• Component/s: Backend
• Target Version: Product Backlog
• Assignee: Abhishek Rawat
• Reporter: Matthew Jacobs
• Created: 14/Dec/16 17:52
• Updated: 10/May/19 23:06
• Resolved: 10/May/19 23:06

Description

There is a potential race in our many usages of ExecNode::ReachedLimit() during execution from a different thread than the one updating rows_returned, e.g. scanner threads. The compiler in theory is free to notice that they're not updating rows_returned_ and hoist the load up out of loops, etc.

We should consider AtomicInt64 or NoBarrier_Load(...).

Issue Links

• blocks: IMPALA-5904 Enable ThreadSanitizer for Impala