Pinterest faced significant challenges with its legacy database ingestion system due to high latency and inefficiency. Updates often took over 24 hours, hindering real-time analytics and machine learning efforts. The reliance on batch jobs led to wasted resources since many tables saw less than 5% daily changes. Additionally, the lack of support for row-level deletion complicated data compliance, while operational complexity arose from multiple independently maintained pipelines. To address these issues, Pinterest developed a unified Change Data Capture (CDC) framework leveraging tools like Debezium, Kafka, and Iceberg. This new system processes database changes in minutes, focuses only on updated records, and supports incremental processing. The architecture includes source databases such as MySQL and TiDB, a CDC layer that captures changes with minimal latency, and a streaming layer that persists data into Iceberg tables. Maintenance jobs ensure ongoing performance optimization. The system differentiates between CDC tables, which quickly capture changes, and base tables that reflect the current state of the data. Updates to the base tables utilize a SparkSQL query that identifies recent changes and applies them efficiently with a `Merge Into` statement. Pinterest standardized on the Merge-on-Read (MOR) approach, avoiding the Copy-on-Write (COW) strategy due to its high storage and compute costs. This shift aims to enhance the overall efficiency of the ingestion process, making it better suited for modern data demands.