Large-scale database UPDATE and DELETE operations executed as single transactions risk replication lag, high load, and UNDO log bloat. Mercari's DBRE team built a general-purpose Go tool called data-updater that accepts SQL-like configuration and automatically splits operations into safe, cursor-based batches. Key features include: three operation modes (UPDATE, DELETE, NULL/archive), replica routing for SELECT queries, pessimistic locking, JOIN support, before-SQL hooks for archiving, runtime control via Unix socket (pause, resume, batch size, interval), health-check-based hibernation, auto-interval adjustment based on hibernation ratio, automatic resume via status files, and GCS-based PK source files. The tool draws inspiration from gh-ost's philosophy of runtime operational control. Source code is not open-sourced, but the full README is published so teams can build their own implementation, optionally with generative AI assistance.
