CDC

Change Data Capture (CDC) streams row-level inserts, updates, and deletes from the OLTP log (WAL, binlog, oplog) to downstream consumers: search indexes, warehouses, caches, event buses. Tools include Debezium, Maxwell, native logical replication, and managed connectors.

Ordering and schema evolution matter: consumers must handle out-of-order retries idempotently; additive schema changes are safer than destructive ones; tombstones signal deletes to Kafka compacted topics.

					OLTP → WAL → CDC connector → queue → search / warehouse / cache projector
				

On interviews: contrast CDC with nightly batch ETL for freshness; explain idempotent projection into Elasticsearch; mention PII filtering before the stream leaves trust boundary.

Common pitfalls: full table scan disguised as sync; no delete propagation; consumer lag unbounded; breaking rename without compatibility layer; exposing raw CDC topic without ACL.

The trade-off is near-real-time downstream freshness versus pipeline complexity, ordering guarantees, and operational monitoring of lag and poison messages.

Checklist:

• Define CDC source: WAL/binlog/oplog.
• Guarantee idempotent downstream apply.
• Plan schema evolution and delete events.
• Monitor lag and dead-letter handling.
• Filter PII at capture or consume boundary.