Travaux, étapes et modèle JobRepository

Many real-world workloads are not request/response: nightly invoice generation, CSV imports, report exports, data migrations. These run as batch jobs processing large volumes of records.

• They must be restartable after a crash.
• They must track progress so you know what already ran.
• They must handle millions of rows without loading everything into memory.

Spring Batch gives you a structured model for exactly this. The three core abstractions you must understand first are the Job, the Step, and the JobRepository.

A Job is the top-level container for an entire batch process. It has a name and is composed of one or more ordered Step instances.

• One Job definition can be executed many times — each run is a JobInstance.
• Each attempt at running a JobInstance is a JobExecution (status, start time, exit code).

In Spring Boot 4 you build a Job with a JobBuilder, supplying a name and a JobRepository.

This distinction is the heart of restartability. Suppose you run importInvoicesJob for the date 2026-06-10.

• The JobInstance is identified by the job name plus its identifying job parameters (here, the date). Re-running with the same date refers to the same instance.
• Each run produces a new JobExecution. If the first attempt fails and you restart, you get a second JobExecution for the same JobInstance.

A completed JobInstance cannot be run again with the same identifying parameters — Spring Batch throws JobInstanceAlreadyCompleteException. This prevents accidental double-processing.

A Step is an independent, sequential phase of a Job. A Job typically chains several steps: validate input, then process records, then send a summary email.

There are two flavors of step:

• Chunk-oriented: read–process–write in configurable chunks. Ideal for large data sets.
• Tasklet: a single arbitrary action (delete a file, run a stored procedure, ping a service).

Each Step run is tracked by a StepExecution, which records read/write counts, commit count, and status.

A chunk-oriented step ties together an ItemReader, an optional ItemProcessor, and an ItemWriter. The chunk(n, transactionManager) call sets the commit interval: after every n items are read and processed, the writer flushes and the transaction commits.

Smaller chunks mean more frequent commits (safer restart point, more overhead); larger chunks mean fewer commits (faster, but more work lost on rollback).

When a phase is a single action rather than a stream of items, use a Tasklet. The execute method returns RepeatStatus.FINISHED when done, or CONTINUABLE to be called again.

Tasklets are perfect for setup/cleanup steps such as archiving a processed file or truncating a staging table.

Steps run in the order you declare them. Use start(...) for the first step and next(...) to chain the rest. By default, the Job stops if a Step ends with status FAILED.

Here the job validates, imports, then cleans up. If the import step fails, cleanupStep does not run, and on restart the job resumes from the failed step.

The JobRepository is the component that persists all batch metadata: which JobInstances exist, their JobExecutions, each StepExecution, and the execution contexts.

• It is how Spring Batch knows a job already completed.
• It is how a restart figures out where to resume.
• It stores read/write/commit counts for monitoring.

By default it writes to a relational database using a set of BATCH_* tables. Without a working JobRepository there is no restartability and no execution history — it is not optional.

The JobRepository persists into a fixed schema. The key tables are:

• BATCH_JOB_INSTANCE — one row per JobInstance (name + identity hash).
• BATCH_JOB_EXECUTION — one row per run attempt, with status and exit code.
• BATCH_STEP_EXECUTION — per-step counters (read/write/skip counts).
• BATCH_JOB_EXECUTION_CONTEXT / BATCH_STEP_EXECUTION_CONTEXT — serialized state used to resume.
• BATCH_*_SEQ — sequences for primary keys.

Spring Boot ships the DDL and can create these automatically. The property below initializes the schema on startup.

You execute a Job through a JobLauncher, passing JobParameters. Identifying parameters define the JobInstance; here the run date makes each day a distinct instance.

The launcher returns a JobExecution whose status (COMPLETED, FAILED, STOPPED) reflects the run. All of this is recorded in the JobRepository.

Put the pieces together. When a Job fails mid-run:

• The failed JobExecution is marked FAILED, but its JobInstance is not marked complete.
• Each completed Step is recorded; a chunk-oriented step also saved how many items it processed in its ExecutionContext.
• Re-launching with the same identifying parameters creates a new JobExecution on the same instance. Spring Batch skips already-completed steps and resumes the failed step from the last committed chunk.

This is why identifying parameters and the JobRepository must be stable: they are the coordinates of resumption.

Test your understanding of the job/instance/execution model.

You now have the structural model of Spring Batch:

• Job — the top-level process, built with JobBuilder, composed of ordered Steps.
• JobInstance vs JobExecution — an instance is identified by name + identifying parameters; each run attempt is a separate execution.
• Step — a phase, either chunk-oriented (read/process/write with a commit interval) or a tasklet (single action); tracked by a StepExecution.
• JobRepository — persistent metadata in the BATCH_* tables that powers restartability, resume logic, and monitoring.

Master these and the rest of Spring Batch — readers, writers, listeners, partitioning — slots neatly on top.