shibumistack.dev

History and rollback.

Inspect verified deployments and restore the previous retained image.

History and rollback

Read recent history#

›shis history example-com
渋み shis (shibumi-server)
2026-08-12T09:03:07Z webhook succeeded 55a26db5c43a 42182ms
2026-08-12T08:44:19Z rollback succeeded 4e5f3223b71d 38914ms
2026-08-12T08:31:02Z webhook failed a15c9e2d773f health 27546ms
3 recent records
shis history example-com 渋み shis (shibumi-server) success|2026-08-12T09:03:07Z webhook succeeded 55a26db5c43a 42182ms success|2026-08-12T08:44:19Z rollback succeeded 4e5f3223b71d 38914ms warn|2026-08-12T08:31:02Z webhook failed a15c9e2d773f health 27546ms outro|3 recent records

Machine-readable form:

shis history example-com --json

Each app keeps latest 100 records in mode-0600 JSONL. Records contain timestamp, app ID, full commit, operation kind, result, verified GitHub delivery ID, failed stage, and duration when available.

History never stores webhook payloads, signatures, secrets, or request headers.

Restore the previous image#

shis rollback example-com

Shibumi selects the one previous successful image retained for the app for up to 12 hours, retags it under the Compose image name, recreates the service without building, and verifies health. Successful rollback rotates retention, making the replaced image available for the next rollback. Failed startup or health restores the current image.

Use --yes only for confirmed automation:

shis rollback example-com --yes

Receiver pauses during rollback and restarts afterward, preventing webhook deployment from racing operator action.

From a connected project, bun ship --rollback invokes the same retained-image rollback over SSH.

What rollback does not undo#

Rollback restores an image, not database contents or uploaded files. Persistent volumes stay in place. Keep migrations backward-compatible and use an explicit database backup/restore for data recovery.

Service recreation can interrupt requests. Caddy retries within a finite window, but a slow startup or recovery from an unhealthy release can exceed it. Deployment output records the measured restore time and retry-budget headroom or overrun. A successful recovery means the previous release is healthy again; it is not a guarantee of zero failed requests during the replacement.

渋み Create a Shibumi project

Start locally. You need Bun 1.4.4 or newer; install Bun or run bun upgrade. No Docker, domain or VPS is needed yet.

In the directory that will contain your project:

› bun create shibumi@latest my-app

For a first app, choose Bun full-stack app, then Later for deployment. When creation finishes:

› cd my-app && bun dev

Open localhost:9001. Click Count, then reload; the value should remain.

Follow the quick start for blog, static and deployment instructions.