Skip to main content
Version: 1.x

To-do lists and plans

Two files keep unfinished work visible: a to-do list for the steps of one session, and a plan for work that runs over several.

A session's to-do list​

For any task of two or more steps, Hostwarden keeps memory/machines/<hostname>/todo.md and ticks each step the moment it is done, not at the end — so an interrupted session leaves behind what was really done.

- [x] Snapshot guest 101 before the upgrade
- [x] Upgrade Debian 12 → 13 on web1.example.com
- [ ] Check that nginx and the app come back
- [ ] delete snapshot pre-upgrade on pve1.example.com

The next connection to the host shows the open items before new work. An item with a due time waits silently until then:

- [ ] First run of db-dump.timer after rename, due 2026-09-21 03:00

The connection that finds it due checks the result and ticks it. Once everything is done, the file is deleted.

Plans that outlive a session​

Work that spans sessions — a rollout in phases, a migration with decisions still open, a change on many hosts — gets a plan in memory/plans/<slug>.md. It holds what a checklist cannot: the goal, the decisions taken and why, the phases, and what is next. It opens with the hosts it touches and where it stands:

# Syslog collector for all sites
- Hosts: log1.example.com, fw1.example.com,
  pve1.example.com
- Status: phase 2 of 4 — collector runs, forwarders
  pending
- Updated: 2026-09-17

Each host it touches points to it from its memory.md:

- Plan: syslog-collector (memory/plans/syslog-collector.md)
  • The next phase never starts unasked. A session working on one of those hosts reads the plan when its work touches what the plan covers.
  • When it is done, its facts move into the hosts' memory, what you decided and still holds becomes a decision, and the plan and its Plan: lines are deleted.
  • A rollout on several hosts is written down this way until every host is done, so a later session can finish it (One task on many hosts).

Maintenance windows​

A planned downtime is a plan of the same kind, with the window, what it reaches and who has to be told by when above its steps:

# pve1 kernel update
- Hosts: pve1.example.com
- Window: 2026-10-05 22:00–23:30 Europe/Berlin
- Kind: reboot
- Affected: web1.example.com, db1.example.com (radius of 2026-09-24)
- Notify by: 2026-09-28 (shop customers, 7 days)
- Status: planned
- Updated: 2026-09-24

Every affected host gets a Downtime: line, which the recent-activity summary names on each connection until the plan is done (Planning a maintenance window).