Key Tips for Improving Spring Batch Application Performance

To improve Spring Batch performance, rather than drastically changing the structure with parallel processing from the start, it is more efficient to maintain the current structure while bundling I/O operations. This is because excessive structural changes can incur greater costs when there are larger performance demands in the future.

The main causes of performance degradation can be summarized into two. The first is the Network I/O wait time that occurs when calling APIs one by one in the ItemProcessor, and the second is the frequent Database I/O caused by Dirty Checking in JpaItemWriter. For example, if a task with a 150ms API response time is processed in Chunk units of 1,000, approximately 2.5 minutes of wait time occurs.

The optimization strategies to solve this are as follows.

• Remove Processor and Consolidate into Writer: Improve the structure so that the single-unit tasks performed in the Processor are moved to the Writer, enabling bulk processing.
• Utilize Parallel Processing: Introduce parallel processing to reduce Network I/O wait time and increase API call efficiency.
• Minimize I/O Count: Use Projection to reduce the overhead of the persistence context, and drastically reduce the number of database updates through the In Update method.