[DRILL-4010] In HBase reader, create child vectors for referenced HBase columns to avoid spurious schema changes

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Component/s: Execution - Data Types, Storage - HBase
• Assignee: Daniel Barclay
• Reporter: Daniel Barclay
• Created: 03/Nov/15 00:26
• Updated: 19/Dec/15 19:59
• Resolved: 19/Dec/15 19:59

Description

HBaseRecordReader needs to create child vectors for all referenced/requested columns.

Currently, if a fragment reads only HBase rows that don't have a particular referenced column (within a given column family), downstream code adds a dummy column of type NullableIntVector (as a child in the MapVector for the containing HBase column family).

If any other fragment reads an HBase row that does contain the referenced column, that fragment's reader will create a child NullableVarBinaryVector for the referenced column.

When the data from those two fragments comes together, Drill detects a schema change, even though logically there isn't really any schema change.

Issue Links

• is part of: DRILL-2288 ScanBatch violates IterOutcome protocol for zero-row sources [was: missing JDBC metadata (schema) for 0-row results...]