[Question] SQL server cdc exactly once to iceberg sink

Im trying to run a job from SQL server to iceberg (multiple tables and multiple sinks).

On average tables have a total of ~200M row between them.
The cluster consists of 3 nodes 16gb ram workers + 1 16gb master.

Currently even with concurrency set to 3 this will oom eventually. Using s3 imap.

The job will run for a while but ultimately memory keeps growing and it dies before it ever finishes the initial snapshot. If job is cancelled, memory never gets returned until the workers are restarted.

What's a good approach to minimize memory and load. Ultimately need to sync whole database with 600+ tables to iceberg.

The issue names no source files, tests, or entry points. Start by reviewing the SQL Server CDC-to-Iceberg job configuration, including concurrency and S3 IMAP usage, and observe worker memory during the initial snapshot; done means a documented approach that syncs the database without eventual OOM.

Written by the indexing model from the issue text.