The snapshot in Debezium will do a historical data load from the source database to the Kafka topics. But generally its not a good practice to this if you have a huge data in your tables. Recently I have published many blog posts to perform this snapshot from Read Replica(with/without GTID, AWS Aurora). One guy commented that, in GCP the MySQL managed service is called CloudSQL. There we don’t have much control to stop replication, perform the modifications that we want. So how can we avoid snapshots in CloudSQL and take debezium snapshots from CloudSQL Read Replica? I have spent some time today and figured out a way to do this.
The Approach:
We can’t enable binlogs on read replica. So we have to setup an
external read replica for this. If the external replica is a VM,
then we can enable the log-slave-updates with GTID.
Then we can …