java.lang.OutOfMemoryError: Java heap space

Root cause

Kafka Connect worker JVM ran out of heap. Common triggers: large initial snapshot (millions of rows buffered), very large individual transactions, or the wal2json plugin (PostgreSQL) which serializes an entire transaction into memory before streaming it.

How to fix

1. Increase Kafka Connect heap:
   export KAFKA_HEAP_OPTS="-Xms1g -Xmx4g"
2. PostgreSQL: use plugin.name=pgoutput (recommended, built-in since PG 10, streams row-by-row) instead of decoderbufs which buffers the whole transaction in memory.
3. Reduce snapshot batch size with snapshot.fetch.size.
4. For large tables, use incremental snapshots instead of full snapshots.
5. Limit captured tables with table.include.list to reduce scope.