Skip to content

v1.8.0 — PiHerder

Choose a tag to compare

View all tags
@github-actions github-actions released this 01 Oct 08:46
· 106 commits to main since this release
v1.8.0
f1e2351

PiHerder v1.8.0

1 October 2026. Tag v1.8.0. Package 1.8.0. Shipped.

An agent can start the jobs the herder already runs: host reboot, compose stack actions, and template deploy. Backups can copy on to Google Drive after they land on the herder. Home Assistant plugin 0.4.4 is one card, and it can start, stop, restart, or update one container. Leftover console sessions can be listed and killed. A Move that dies while the destination is starting can be put back. Retention, the herder’s own backup, and host facts keep running if the web process restarts.

Image: bjorngluck/piherder 1.8.0 · 1.8 · latest (amd64 + arm64). Pins 1.7.0 / 1.7 stay valid. The public demo is the 1.7.0 image.

Operator how-to: Agents (MCP) · Home Assistant · PiHerder backup · Jobs · Move a service. Technical record: PLAN_v1.8.0. Maintainer QA: QA_v1.8.0.

What’s new

Agents can run the jobs you already run by hand

Hosted POST /mcp accepts the job types the bearer API already accepted in 1.7, plus host reboot, compose check, deploy, stop, start, and restart, and template deploy and redeploy. The air-gapped adapter is piherder-mcp 0.2.0 (uvx piherder-mcp). A client that already points at hosted /mcp does not install it.

A stack or template call uses the compose project path the jobs page already takes. Restart host names the machine. It is refused while an OS patch, a container patch, or a backup is already running. If a job is already active, the tool returns 409. Poll that job. Do not start another.

Still refused: Move, undo, nmap, the console, docker compose down, remove, and the four one-service buttons on the Home Assistant card (container_start, container_stop, container_restart, container_redeploy). Whether those four join the agent tool is a question for the next release.

Wiki: Agents (MCP)

A backup can copy on to Google Drive

Host backups still rsync on to a directory on the herder. That directory stays the default, and restore still uses it. Settings → PiHerder backup can then copy the whole backup drive to a folder in your Google account.

The account is a Google sign-in from this PiHerder, not a service-account key. The dialog lists the Cloud steps: Branding, Audience (Testing, plus you as a test user), the Drive scope, and the redirect URL. The client secret is not shown again. A blank secret keeps a saved one. Test checks that the folder is in that account’s My Drive. It does not copy files.

Folders are a tree. Tick a folder and the children stay ticked. Open it and untick a child to leave that child out. There is no typed exclude list.

Copy now, the schedule, and the optional follow-up after a host backup enqueue a Drive copy job. A failed copy does not change the host backup time. A copy that runs for hours stays one job. It is not cut off at two hours. It can run for up to seven days. If the worker stops cleanly, the row fails and Copy now can run again. Files already on Drive stay. A hard kill of the worker can leave the row running, and Copy now then returns that row until someone marks it failed.

OneDrive and a LAN NAS are in the service list and cannot be selected. The public demo does not upload. The token API and MCP cannot start this copy.

Wiki: Backups — copy to Google Drive. That page has the Cloud steps and the pictures of the card.

Home Assistant is one card

Install bjorngluck/piherder-ha 0.4.4 with HACS, then restart Home Assistant. It is not inside the PiHerder image. Set the dashboard resource to /local/piherder-dashboard-card.js?v=0.4.4 as a JavaScript module, and hard-refresh. The Plugin sensor reads 0.4.4. Delete an older ?v=0.3.0, ?v=0.4.0, ?v=0.4.1, ?v=0.4.2, or ?v=0.4.3 resource if it is still there. The wiki pictures were captured on 0.4.3. The card is the same.

One card, three tabs: Fleet, Host, Updates. A host strip picks the machine, with a Raspberry Pi mark and an OS icon. Click memory, disk, or CPU load to open that sensor’s Home Assistant history. A thin sparkline draws when the recorder has points. The card keeps the tab, the selected host, and an open Containers section when it redraws.

Updates writes the counts out: OS updates, container updates, and reboot pending. They are gold when something is due. The row opens that host.

Containers lists the last inventory. A running service has Stop and Restart. A stopped service has Start. A container that needs an image update has a gold name and Update. Each one asks first, then runs docker compose for that service only. Update pulls the image and recreates that service. The other containers stay as they are. A container with no compose directory or service name has no button. A read token shows no job buttons.

A job the plugin was watching fires piherder_job_completed on the Home Assistant bus when it leaves the active set. The first poll after startup does not fire it. There is no new herder route and no webhook.

A 1.7 herder answers 400 for start, stop, restart, and update. Those four need this release’s web and worker.

Wiki: Home Assistant

Leftover console sessions

On a host’s SSH access page, PiHerder lists leftover ph-u* tmux or screen sessions for that host and can kill one. Hide still detaches. Closing the console still kills that session. There is no automatic reattach. Removing the server does not kill every Unix user’s sessions. Console mux stays off unless you turn it on, and it stays off on HAOS and the public demo. Kill ends any ph-u* session on that host, including another operator’s tab.

A Move that dies while the destination is starting

If a Move dies during destination start, you can inspect the destination, then stop it and start the source again. DNS and Nginx Proxy Manager stay as they were. A Move that finished still has no Undo. Destination down -v is not offered. Move stays off until you set PIHERDER_SERVICE_MIGRATE.

Retention, self-backup, and host facts survive a web restart

retention, the herder’s own backup, and host facts run on the Celery worker. Recycling web does not fail them. Recycling the worker while one is running fails it. It is not resumed mid-flight. Host facts stays one job per host. The herder backup and retention do not take a host slot.

Defaults (opt-in stays off)

Default
Move a service off (PIHERDER_SERVICE_MIGRATE)
Console mux off per host (and never on HAOS or the demo)
Web SSH console off (PIHERDER_SSH_CONSOLE)
Host Files (real SFTP) off (PIHERDER_HOST_FILES)
Google Drive copy off until you connect an account
MCP off until you create a token. No second container
Catalog in the nav on

Upgrade from 1.7

Alembic 047 (backup_destination) runs when web starts. Last shipped revision on 1.7.0 is 046.

  1. Full DR self-backup. Keep PIHERDER_MASTER_KEY.
  2. Pull bjorngluck/piherder:1.8.0 (or git checkout v1.8.0).
  3. docker compose pull && docker compose up -d — recreate web and celery-worker. App code is not bind-mounted.
  4. Confirm the database revision includes 047.
  5. Move stays off.
  6. HACS: update piherder-ha to 0.4.4, restart Home Assistant, and set the card resource query to v=0.4.4.
  7. If you use uvx piherder-mcp, that package is already 0.2.0. Hosted /mcp updates with this image. It does not gain the four one-service jobs.

Honest limits
MCP Bearer token only. No OAuth. The four one-service container jobs stay on the Home Assistant card. Down, remove, Move, undo, nmap, and the console stay off the tool.
Google Drive The scope is full Drive, because a folder you created in the Drive UI is invisible to the narrower scope. A hard kill of the copy worker can leave the job running.
Home Assistant No Move, Files, console, webhooks, or whole-project stop. The card pictures were captured on plugin 0.4.3. HACS current is 0.4.4. The Google Drive card is in the wiki pictures.
Move Stays off. A finished Move still has no Undo.
Not this release OneDrive, a LAN NAS, per-host grants, and putting the one-service jobs on MCP.
Assets 2
Loading