Large-scale transaction processing in I/O-bound services can lead to performance bottlenecks and database instability, particularly on resource-constrained servers. An inappropriate concurrency strategy may cause excessive memory consumption and system failure under high workloads. This study compares the efficiency and stability of three concurrency models in Go: Sequential, Unlimited Goroutine, and Worker Pool with a Backpressure mechanism for processing 11 million transaction records against a PostgreSQL database. Testing was conducted in an isolated environment using Docker containers running on a Colima virtual machine configured with 4 CPU cores, 8 GB RAM, and a maximum of 50 database connections. Data were retrieved iteratively using Dynamic Keyset Filtering, while the batch size was calibrated through a preliminary tuning experiment. The results indicate that a batch size of 16,000 provided a balance between I/O efficiency and memory stability. The Unlimited Goroutine model failed to complete the workload due to an out-of-memory (OOM) condition. The Sequential model processed all records in 632 seconds with a throughput of 17,405 transactions per second (Tx/s). The 4-worker Worker Pool achieved the highest throughput at 21,917 Tx/s in 502 seconds, while the 8-worker Worker Pool achieved 19,755 Tx/s in 557 seconds with a 0% error rate and more consistent performance across the tested conditions. Based on the consistency of its performance, the 8-worker Worker Pool was selected as the preferred configuration in this study. These findings indicate that combining a Worker Pool with Backpressure and an appropriately calibrated batch size can improve the performance and stability of large-scale, Go-based transaction processing under the tested conditions.