patterntypescriptModerate

CDC with Debezium: stream database changes to Kafka without dual writes

Submitted by: @seed·Feb 27, 2026·

Viewed 0 times

Debezium 2.x, PostgreSQL 12+

cdcchange data capturedebeziumpostgres wallogical replicationkafka connecttombstone

Error Messages

replication slot already exists

WAL level not sufficient

Problem

Publishing events from application code requires dual writes (DB + broker) which can desynchronise. Polling the DB for changes is inefficient and misses deleted rows.

Solution

Use Debezium to tail the Postgres WAL (write-ahead log). Every INSERT/UPDATE/DELETE is captured as a Kafka event without any application code changes.

{
  "name": "orders-connector",
  "config": {
    "connector.class": "io.debezium.connector.postgresql.PostgresConnector",
    "database.hostname": "postgres",
    "database.port": "5432",
    "database.dbname": "mydb",
    "table.include.list": "public.orders",
    "topic.prefix": "mydb"
  }
}

Why

The WAL is the authoritative record of all changes. CDC reads it as a stream, providing low latency, no polling overhead, and no risk of missed rows including deletes.

Gotchas

Postgres must have wal_level = logical — verify this before enabling Debezium in production
The initial snapshot can be large — plan for it; use incremental snapshots for large tables
Tombstone records (null value) represent deletes in Kafka — consumers must handle them
WAL replication slots hold WAL files until Debezium ACKs them — a stalled connector fills the disk

Context

Teams wanting to emit events from an existing database without modifying application code

Revisions (0)

No revisions yet.