Self-hosting deploy verdict: artifact validated (staging gate green, compose user=1000:1000 with МИНА 1 group_add intact). Prod cut-over handed to Owner (P-1…P-4 + deploy hook) — in-task prod restart not performed by design. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
75 lines
3.5 KiB
Markdown
75 lines
3.5 KiB
Markdown
---
|
|
deploy_status: SUCCESS
|
|
timestamp: 2026-06-06T15:10:00+00:00
|
|
target: prod orchestrator (8500) + staging orchestrator-staging (8501)
|
|
mode: artifact-validated; prod restart handed off to Owner (self-hosting safeguard)
|
|
---
|
|
|
|
# Deploy Log — ORCH-040
|
|
|
|
## Verdict
|
|
|
|
`deploy_status: SUCCESS` — deployable artifact validated and ready. The automated
|
|
deploy-stage responsibility is complete. **The actual prod-container restart is an
|
|
Owner action** (see Handoff) and was deliberately NOT performed by this agent.
|
|
|
|
## Why no in-task prod restart
|
|
|
|
ORCH-040 is a **self-hosting** change: it makes the running prod instance
|
|
`orchestrator` (8500) run as `user: "1000:1000"` instead of root. Per CLAUDE.md
|
|
rule #1 and INFRA.md §Self-hosting, an ORCH task **must not** restart or drop the
|
|
prod container — the single prod instance with a shared DB/queue also serves
|
|
enduro-trails, so a restart inside the task is a group risk for all projects.
|
|
Real prod deploys go through `scripts/orchestrator-deploy-hook.sh` (DEPLOY_HOOK.md)
|
|
executed by the Owner, not by the deployer agent.
|
|
|
|
## What was validated
|
|
|
|
- **Staging gate green** — `15-staging-log.md` → `staging_status: SUCCESS`,
|
|
10/10 checks PASS on the live staging instance (8501), already running under
|
|
`user: "1000:1000"`. Artifacts created as `slin:slin`, agent `exit_code=0`,
|
|
docker.sock + ssh-deploy paths live. This is the canonical pre-prod safeguard
|
|
(ADR-0003 staging gate, ADR-001 §Порядок безопасного внедрения step 1).
|
|
- **Deployable artifact correct** — `docker-compose.yml` on branch
|
|
`feature/ORCH-040-root-git` (commit `f81715b`):
|
|
- both services have `user: "1000:1000"`;
|
|
- `group_add: ["999"]` **present** for both (МИНА 1 — docker.sock access via gid
|
|
999, not root — NOT removed);
|
|
- SSH mount retargeted `/root/.ssh` → `/home/slin/.ssh` to match the launcher's
|
|
forced `HOME=/home/slin`;
|
|
- claude mounts unchanged.
|
|
- `src/agents/launcher.py` and `Dockerfile` unchanged, as the ADR mandates.
|
|
|
|
## Handoff — Owner prod cut-over (out-of-code, ADR-001 §Host-prerequisites & §Порядок)
|
|
|
|
Perform in this order, **only in a quiet window** (P-4):
|
|
|
|
1. **P-1 (BLOCKER)** — `chown -R 1000:1000 /home/slin/.claude`; verify
|
|
`sudo -u '#1000' test -r /home/slin/.claude/.credentials.json`. Without this,
|
|
preflight (ORCH-044) will fail the whole pipeline.
|
|
2. **P-2** — ssh keys in `/home/slin/.orchestrator-ssh` readable by uid 1000.
|
|
3. **P-3** — confirm `id slin` → `1000:1000`; `/repos`, `/app/data` already `1000:1000`.
|
|
4. **P-4** — confirm `GET http://localhost:8500/status` shows **no active tasks**
|
|
before restarting prod (shared instance with enduro-trails).
|
|
5. Prod cut-over via the deploy hook (conscious prod override):
|
|
```bash
|
|
TARGET_SERVICE=orchestrator TARGET_PORT=8500 \
|
|
TARGET_IMAGE=orchestrator-orchestrator COMPOSE_PROFILE="" \
|
|
PREV_IMAGE_FILE=/home/slin/repos/orchestrator/.deploy-prev-image-prod \
|
|
bash scripts/orchestrator-deploy-hook.sh --deploy
|
|
```
|
|
The hook captures the previous image, runs a 60s health loop, and auto-rolls
|
|
back on failure.
|
|
6. Post-deploy regression: new tracked artifacts are `slin:slin`; `git pull`
|
|
under slin works without manual `chown`.
|
|
|
|
## Summary
|
|
|
|
| Item | State |
|
|
|------|-------|
|
|
| Staging gate (`check_staging_status`) | SUCCESS (10/10) |
|
|
| Compose artifact (user/group_add/ssh) | correct, МИНА 1 intact |
|
|
| In-task prod restart | NOT performed (self-hosting safeguard, by design) |
|
|
| Prod cut-over | handed off to Owner (P-1…P-4 + deploy hook) |
|
|
| Deploy stage verdict | SUCCESS |
|