[SPARK-16686] Dataset.sample with seed: result seems to depend on downstream usage

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.6.2, 2.0.0
• Fix Version/s: 2.0.1, 2.1.0
• Component/s: SQL

Environment

Spark 1.6.2 and Spark 2.0 - RC4
Standalone
Single-worker cluster

• Assignee: L. C. Hsieh
• Reporter: Joseph K. Bradley
• Created: 22/Jul/16 17:59
• Updated: 19/Aug/16 18:19
• Resolved: 26/Jul/16 04:06

Description

Summary to reproduce bug:

• Create a DataFrame DF, and sample it with a fixed seed.
• Collect that DataFrame -> result1
• Call a particular UDF on that DataFrame -> result2

You would expect results 1 and 2 to use the same rows from DF, but they appear not to.
Note: result1 and result2 are both deterministic.

See the attached notebook for details. Cells in the notebook were executed in order.

Issue Links

• is duplicated by: SPARK-15382 monotonicallyIncreasingId doesn't work when data is upsampled
• links to: [Github] Pull Request #14327 (viirya)