Spring Batch restart input: pin the manifest before resuming a cursor

Freeze the source set

The checked two-process fixture keeps four database rows unchanged. A real import may read a CSV replaced at the same path or a query whose result changes while the job is down. Persist a manifest that names an immutable object version or source snapshot, includes a count and digest, and becomes part of the job's business identity. A location string by itself is not a version.

Reject changed replay

Before resuming, recompute the manifest digest or reopen the immutable object version. If it differs, fail for review. Do not let an upsert writer conceal a changed receipt under the same source ID. For mutable database sources, use a snapshot or a bounded query over versioned rows; a new row inserted ahead of the cursor can shift ordering if the reader relies on position.

Separate correction from restart

A corrected source file is a new business input. Give it a new manifest identity and record its relationship to the earlier failed one. Reusing the old identity with new bytes makes the saved checkpoint meaningless. The checked kit now mutates ID 4 after failure and rejects the same-manifest restart before a new Batch execution is created. The digest test remains a gate against change at launch, not an immutable database snapshot.

Boundary sketch

record ImportManifest(String manifestId, String sourceVersion,
                      long sourceCount, String sha256) {}
// Reopen the pinned sourceVersion and compare its digest before restart.

Cost and verification

Immutable input consumes storage and retention time. Digesting a large source is linear in its byte size and may require a separate pass.

Common Mistakes

• Do not treat a pathname as immutable input.
• Do not resume a positional cursor over a reordered query result.
• Do not overwrite a changed receipt because its ID already exists.

Read next

Spring Batch restart across two JVMs after a recorded failed execution, Spring Batch writers: use a stable source key when a chunk is replayed, Spring Batch job identity: a manifest ID defines restart versus a new run, Spring Batch reconciliation: report accepted and rejected receipt IDs.