Number of input and output link

Number of input link	:	One
Number of output link	:	One

Buffer Assembly () processes grouped records and produces one or more output records per group.
A Buffer Assembly must always follow a Group By Assembly, because it operates on already grouped and sorted data.

Unlike an Aggregator, which reduces a group into a single summary record, a Buffer has access to all records within a group at once and can emit:

• One record
• Multiple records
• Transformed records based on group context

Key Characteristics

• Requires Group By before execution
• Operates on all records of a group
• Can generate multiple output records per group
• Supports ranking, windowing, routing, and accumulation
• Executes after the shuffle and sort phase

Predefined Buffer Functions

1. Ranking
2. Top N Sum
3. Top N Count
4. Top N Average
5. Top N Max
6. Top N Min
7. Bottom N Sum
8. Bottom N Count
9. Bottom N average
10. Bottom N Max
11. Bottom N Min
12. Accumulation
13. Accumulations
14. Route
15. Running
16. N-Tile
17. NTileAggBucket
18. WindowPartition

Important Notes

• Buffer functions are more memory-intensive than aggregators
• Use Buffer only when:
  • Ranking is required
  • Multiple outputs per group are needed
  • Window or cumulative logic is required
• For simple summaries, prefer Aggregator or Global Combiner

Typical Use Cases

• Top/Bottom N analysis
• Ranking and leaderboard generation
• Running totals and cumulative metrics
• Window-based analytics

Supported Engines

Map Reduce	Spark 1.x RDD	Spark 2.x RDD	Spark 2.x Dataset	Apache Beam 2.29.0 Batch	Apache Beam 2.29.0 Streaming	Apache Flink 1.12.0 Batch	Apache Flink 1.12.0 Streaming	Apache Tez 0.9.1	PySpark