Backup-forge.

With a database per tenant, a backup script that lists databases by name is out of date as soon as a new tenant arrives. Backups have to find every tenant database on their own, and a failure on one must never pass silently.

An EventBridge schedule starts an ECS task. The container lists every non-system database, runs gzip-compressed mongodump jobs in parallel (two at a time by default), retries each database up to three times, and uploads the dumps to S3 under a dated folder, with the run's log beside them.

Any database that still fails after its retries is reported, and the task exits non-zero, so the failure shows up in ECS instead of hiding in a log nobody reads.

A shell script packaged as a Docker image with no host dependencies, pushed to ECR. Credentials come from ECS task secrets, and the task role only needs permission to write to the backup bucket.