[SPARK-9611] UnsafeFixedWidthAggregationMap.destructAndCreateExternalSorter will add an empty entry to if the map is empty.

• Type: Sub-task
• Status: Resolved
• Priority: Blocker
• Resolution: Fixed
• Fix Version/s: 1.5.0
• Component/s: SQL
• Target Version/s: 1.5.0
• Sprint: Spark 1.5 release
• Assignee: Yin Huai
• Reporter: Yin Huai
• Shepherd: Josh Rosen
• Created: 04/Aug/15 20:26
• Updated: 06/Aug/15 02:19
• Resolved: 06/Aug/15 02:19

Description

There are two corner cases related to the destructAndCreateExternalSorter (class UnsafeKVExternalSorter) returned by UnsafeFixedWidthAggregationMap.

1. The constructor of UnsafeKVExternalSorter tries to first create a UnsafeInMemorySorter based on the BytesToBytesMap of UnsafeFixedWidthAggregationMap. However, when there is no entry in the map, UnsafeInMemorySorter will throw an AssertionError because we are using the size of map (0 at here) as the initialSize of UnsafeInMemorySorter, which is not allowed.

2. Once we fixes the first problem, when we use UnsafeKVExternalSorter's KVSorterIterator loads data back, you can find there is one extra records, which is an empty record.

Issue Links

• links to: [Github] Pull Request #7948 (yhuai)