[PARQUET-2109] Parquet Cpp Reader Can Loop Forever If Page Values Overstated

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: cpp-7.0.0
• Component/s: parquet-cpp
• Labels: pull-request-available
• Assignee: William Butler
• Reporter: William Butler
• Created: 17/Dec/21 00:24
• Updated: 23/Jun/24 03:32
• Resolved: 13/Jan/22 12:55

Description

If the page header states that there are more values than are actually present in the page, the Parquet CPP can loop forever.  This is because HasNext() will return true but the actual ReadBatch() will have nothing to read and will not change reader state, causing an infinite loop. We first noticed the bug via ScanFileContents(), but this impacts any code that does not check to see if ReadBatch() consumed anything.

Issue Links

• links to: GitHub Pull Request #11984