Seven of ten partitions completed before a job failed.
Restart only the failed partitions without duplicating successful output.
Which recovery approach is most appropriate?
Persist partition-level checkpoints or a manifest and rerun only incomplete work with idempotent output.
Detailed explanation
A committed completion state lets recovery skip successful work and target the failed inputs.
A committed completion state lets recovery skip successful work and target the failed inputs.
Full reruns waste resources and can duplicate existing results.
Full reruns waste resources and can duplicate existing results.
API history does not guarantee that a business partition was committed successfully.
API history does not guarantee that a business partition was committed successfully.
Deletion removes the input needed for investigation and replay.
Deletion removes the input needed for investigation and replay.
Check official documentation
Compare the following points with the official documentation.
AWS Glue job bookmarks、Step Functions Map、checkpoint、idempotent outputの公式説明を確認する。
What to confirm
全再実行と粒度の細かいチェックポイント復旧の違いを説明できる。
Key points
- Checkpoint
- Manifest
- Partition retry
- Idempotency
Notes
- Environment: AWS公式DEA-C01試験ガイドと各サービスの公式ドキュメントの確認
- Confirm current AWS service behavior in the official DEA-C01 documentation.
Foundation review
State trust
A checkpoint must represent committed output, not a partially written file.
Recovery rehearsal
Test failure, restart, duplicate delivery, and late input scenarios regularly.