Root cause
The worker exhausted JVM heap. Large records/transactions, connector queues, converters and snapshots can contribute; inspect heap and runtime metrics.
How to fix
- Measure heap, GC, queue sizes, record sizes and concurrent tasks before choosing a heap budget.
- Review connector-specific queue and snapshot limits. snapshot.fetch.size is not a universal cap on transaction or converter memory.
- Check the installed output plugin and connector behavior; do not assume pgoutput or decoderbufs always buffers an entire transaction or always streams one row at a time.
Recovery guidance reviewed 2026-10-27 against v3.7.0.Final.
Official Debezium documentation ↗