# S3 Batch Operations

> The dedicated console for large S3 copies with AWS S3 Batch Operations, from readiness to completion report.

For large S3 moves, Douklar wraps AWS **S3 Batch Operations** with readiness
checks, manifests and evidence.

![S3 Batch console with job costs and a list of running, failed, complete and held jobs](console:batch)

## Readiness

A read-only inspection of what a Batch job needs: the role, the operations
bucket, the source bucket policy, and S3 Inventory. Missing source-side
changes are handed over as exact statements for the source bucket's owner to
apply.

## Prepare

**Prepare** is the only action in Douklar that creates resources. In the
destination account it creates a support bucket, the Batch role and its
policy, after you confirm the listed writes. Prepare lists each write with its
exact document and writes nothing if the list changed in the meantime.
Prepare runs only when you start it; submitting a job does not run it.

## Manifests

| Mode | What it is |
|---|---|
| Generated CSV | Version-pinned list of only the missing or changed objects |
| S3 Inventory | Built from an S3 Inventory report |
| AWS-generated | Same Region, empty destination; the job is held until you press Start |

If nothing is missing, no AWS job is created.

## Jobs

Submit, monitor (status refreshes every 8 seconds), cancel, preview the
completion report, and see a cost estimate per job. Jobs are tagged
`ManagedBy=douklar`; Douklar adds no tags to your objects.

## Retry unfinished objects

A Batch job that failed, was cancelled, or completed with failed tasks can be
retried from the Jobs drawer with **Retry unfinished objects**. The retry is
a new AWS job, with a new ID and a new job fee.

| Parent job | Rows the retry considers |
|---|---|
| `Complete` with failed tasks | The failed rows of the completion report |
| `Failed` or `Cancelled`, tasks ran | The parent manifest minus the report's succeeded rows |
| No task ran | The whole parent manifest |
| No complete report, or the report or manifest changed | None: refused, use **New job** |

Before submitting, Douklar checks the destination for every remaining row and
leaves out the rows it already holds. Rows that cannot succeed are left out
and listed: archived objects, objects over 5 GiB, and objects deleted from
the source. The retry lists no source bucket.

## Limits

- 5 GiB per object for Copy (an AWS limit, checked in preflight).
- A chosen KMS key is not yet supported for Batch runs.
- Batch builds `arn:aws:` ARNs, so other AWS partitions are not supported.
- Archived objects are refused with restore guidance.

## Sync or Batch?

Douklar recommends a path from configurable, advisory thresholds: buckets
over 5 GiB or 100,000 objects lean towards Batch.

Source: https://douklar.app/docs/douklar/s3-batch-operations (Douklar docs, last checked 2026-10-23)