Files
stack/docs/plans/2026-10-20_multi-instance.md

71 lines
5.0 KiB
Markdown

# Brief: several Mosaic Stack instances on one host
Sage, 2026-10-20, for Mos's order mos-0974 (MVP, DECISIONS Q176-Q178, Vikunja #414). Gitea #1688. T1 item 1. Owner
Darkwing, reviewer Filbert, gate owner Sage. Base: refactor at the brief commit. This row blocks both MVP
cases: Case 1 (Jason's software factory) and Case 2 (Melanie's creative writing hub) run as two instances
on one homelab host, beside the current mosaic-stack test instance.
Design first. The first deliverable is `agents/darkwing/work/queue-190/DESIGN.md`. Filbert reviews it
before any code, as Q175 did. Then the build follows the design, with tests and mutants.
## The goal
Two or more instances run side by side on dragon-lin. Each one has its own config file, data root (store,
broker.sock, host.lock, state, runs, sessions, workspaces), businesses, bus unit, Console unit and port,
tester port, Discord unit, Claude session env file, stack-active pointer and activation log. No instance's
commands, units or tools address another instance's state by default or by mistake. On one uid this is
addressing only, not a security boundary: a same-uid process can still read another instance's files.
Isolation against that is row 190b (#1693, mos-0986), one Linux user per instance. The current instance keeps working unchanged: its
config path, data root, unit names and ports stay as they are, and no re-pin is needed for it.
## Known single-instance points (GAP-MAP.md, measured at 9ae8c73ac)
The design starts with a full inventory. Grep every script and package, not only these:
- `packages/cli/systemd/mosaic-bus.service.in`: `mosaic-bus@<business>` sets no `MOSAIC_CONFIG` and reads the
fixed `EnvironmentFile=-%h/.config/mosaic-dev/secrets/claude-session.env`.
- `packages/webui/systemd/mosaic-webui.service.in` and `scripts/webui-service.sh`: one `mosaic-webui.service`.
- `packages/discord/systemd/[email protected]` and `scripts/discord-service.sh`.
- `scripts/stack-copy.sh:8` hardcodes `BUSINESS=mosaic-stack`, and `:95` reads `mosaic-webui.service`.
- `scripts/stack-release.sh`: unit names, the Console port and the tester port default, `COPIES` (:28), the
pointer under `$DATA/state` (:1474), and `rehearse`.
- `scripts/mosaic-config.mjs`, `scripts/bootstrap.sh`, `scripts/reset.sh`, `scripts/land-work.sh`,
`packages/webui/src/cli.mjs`, `packages/business/src/{business,vars}.mjs`, and
`packages/conversation/src/guard.mjs` each name the default config path or data root.
## What the design must settle
1. Instance identity. How a name maps to its config file and how every tool picks the instance (an
`--instance NAME` flag, `MOSAIC_CONFIG`, or both), and what happens with neither (the current default,
unchanged). Name rules, and refusal of a name that collides.
2. Config canon. AGENTS.md invariant 2 says `~/.config/mosaic-dev/config.json` is the sole system config,
created only by `scripts/bootstrap.sh`. Several instances mean several config files. The design proposes
the exact wording change and how bootstrap creates an instance config without overwriting one. Mos rules
on the canon change before the build.
3. Unit names and templates. The bus, Console and Discord units per instance, how `MOSAIC_CONFIG` and the
Claude env file reach each unit, and how the current unit names stay valid for the default instance.
4. Ports. Each instance's Console and tester ports, with a refusal when two instances claim the same port.
5. Stack copies and the pointer. Whether copies are shared between instances (read-only trees, so sharing
may be fine) and how the pointer, activation log, backups and `flags_drift` stay per instance.
`stack-release.sh` and `stack-copy.sh` take the instance.
6. Secrets. The per-instance Claude session env file and the secrets slot root, resolved from the instance's
config dir (the Q175 design resolves its slot the same way: `<configDir>/secrets/mosaic-stack/<slot>`).
7. Isolation proof. What the tests show: two instances started side by side on stand-in units and scratch
roots, each reads only its own config and data root, a command aimed at one never touches the other, and
the default instance's paths and unit names are byte-identical to today.
## Boundaries
- No test touches `~/.mosaic-dev`, `~/.config/mosaic-dev`, the live units or a slot file. Use the stand-in
systemctl, units and scratch roots that `packages/cli/tests/stack-release.test.mjs` already builds.
- Standing up a real second instance on dragon-lin is a later window under Q173, with a backup first and a
STATE report after. Exposing a new Console through the tunnel or Traefik changes who can reach it from
outside, so it waits for Jason. SSO (Q175, #412) comes first for any instance a second user can reach.
- Merge-chaining (Q177): rows that need instances (human users, kill switch, lifecycle) may build on this
row's candidate. This row keeps its own review and gate.
## Gate
Filbert approves DESIGN.md, then the manifest. The cli suite and every `scripts/test-*.sh` are green on
Sage's gate, and the design's mutants are killed.