โฟลว์ตัวอ่าน-ตัวประมวลผล-ตัวเขียนแบบแบ่งเป็นชุด

Spring Batch reads, processes, and writes data in chunks instead of one record at a time. A chunk is a configurable number of items (the commit-interval) handled inside a single transaction.

• Read N items one by one with an ItemReader.
• Process each item with an ItemProcessor.
• Write the whole batch of N at once with an ItemWriter.

Writing in bulk and committing per chunk is what makes batch jobs scale to millions of rows without exhausting memory.

Every chunk step is built from three single-method interfaces. Knowing their contracts is the foundation of the whole pattern.

• ItemReader<I> → I read() returns the next item, or null when the input is exhausted.
• ItemProcessor<I, O> → O process(I item) transforms an input into an output; returning null filters the item out.
• ItemWriter<O> → void write(Chunk<? extends O> chunk) persists the accumulated chunk.

A chunk step flows typed data from reader to processor to writer. Let's model a simple input and output. The reader emits raw Customer records and the writer stores normalized ones.

Using a Java record keeps these immutable value types concise. This snippet is plain Java with no framework, so it runs standalone.

For database input, JdbcCursorItemReader streams rows one at a time so memory stays flat. You give it a DataSource, a SQL query, and a RowMapper to turn each row into a domain object.

Spring Batch calls read() repeatedly until it returns null, advancing the cursor each time.

The processor is where business logic lives: validation, enrichment, transformation, or filtering. Its input and output types may differ.

• Return a transformed object to pass it downstream.
• Return null to skip the item entirely — it never reaches the writer.

Keep processors stateless and idempotent so they behave correctly when chunks are retried.

The writer receives the whole processed chunk at once via a Chunk<O>. Writing in bulk — one batched INSERT per chunk instead of one per row — is the key performance win.

JdbcBatchItemWriter uses a parameterized SQL statement and a bean-property parameter source to map fields automatically.

A Step ties the three components together. The generic types <Customer, Customer> declare the input and output of the chunk, and the integer is the commit-interval.

With a chunk size of 100, Spring Batch reads 100 items, processes each, then writes all survivors in one transaction before committing.

A Job is an ordered set of steps. For a single chunk step, the job simply starts with it. Spring Boot auto-detects the Job bean and runs it on startup.

The JobRepository records execution metadata — status, read/write counts, and the last committed position — enabling restartability.

The commit-interval is a tuning lever, not a magic number.

• Too small (e.g. 1): one transaction per row — high commit overhead, slow.
• Too large (e.g. 100,000): bigger transactions, more memory and rollback cost if a chunk fails.
• Typical sweet spot: 100–1000, tuned by measuring throughput against your database and row size.

Remember: a failed item rolls back the entire chunk, so larger chunks mean more work redone on failure.

Real input is messy. Wrap the step with .faultTolerant() to keep processing despite isolated failures.

• .skip(...) — tolerate up to skipLimit bad items instead of failing the job.
• .retry(...) — re-attempt transient errors (e.g. deadlocks) up to retryLimit before giving up.

On a skip or retry, Spring Batch re-scans the chunk item by item, isolating the offending record.

Putting it together, each chunk follows the same loop inside one transaction:

• 1. Read: call read() repeatedly until chunk-size items are buffered (or null ends input).
• 2. Process: call process() on each item; nulls are filtered out.
• 3. Write: pass the surviving items as one Chunk to write().
• 4. Commit: commit the transaction and record progress in the JobRepository.

Then the loop repeats for the next chunk until the reader is exhausted.

Test your understanding of the chunk pipeline.

You wired a complete chunk-oriented flow in Spring Batch:

• ItemReader streams input one item at a time, returning null at the end.
• ItemProcessor transforms or filters items; null drops an item.
• ItemWriter persists the whole chunk in bulk for performance.
• The Step binds them with a chunk(size, txManager) commit-interval, and the Job orchestrates the steps.
• Tune chunk size for throughput, and add .faultTolerant() with skip/retry for resilient pipelines.

This read-process-write loop, committed per chunk, is the backbone of scalable batch processing.