[AVRO-3643] GenericDatumWriter Potential Performance improvement

• Type: Improvement
• Status: Open
• Priority: Minor
• Resolution: Unresolved
• Affects Version/s: 1.11.1
• Component/s: java
• Assignee: Unassigned
• Reporter: Paul Anderson
• Created: 18/Oct/22 15:38
• Updated: 19/Oct/22 13:29

Description

Sorry for the seemingly naïve nature of this observation, however whilst trying to resolve a Kafka client performance issue on Mainframe I noticed a potential opportunity in Avro serialisation. I am using a compiled Avro schema and noticed with my special profiling tool (see attached image) that writeWithoutConversion calls resolveUnion which performs three HashMap.get's.

Seeing this a compiled serialiser is there not an opportunity to eliminate these calls which would reduce the Record serialisation cost to something closer the absolute minimum .writeString (.writePrimitive).

 

Issue Links

• links to: GitHub Pull Request #1917