Phase moves to "live in production"; last-commit pointer and open-issues list reflect the deploy pipeline and proxy-headers fix. Session 3 summary added; Session 1 rotates off to the wiki (per the "most recent 3" rule). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
6.1 KiB
Quartermaster (repo instructions)
Database safety rule
The SQLite database at ./quartermaster.db (or whatever path
QUARTERMASTER_DB_URL points at) holds real user data. It is
.gitignored, so losing it means the data is gone.
Before any operation that interacts with the database at the schema or destructive level, run:
./scripts/backup-db.sh <reason>
That includes:
rm,mv, truncate, or otherwise replacequartermaster.db- Ad-hoc
sqlite3sessions that mightDROP,DELETE, orUPDATE - Any script that does data migration or cleanup outside the running application
Alembic runs scripts/backup-db.sh alembic automatically on every
alembic upgrade, alembic downgrade, or alembic revision --autogenerate
via the hook in alembic/env.py, so you do not need to call it
manually for ordinary schema work. The hook is defense in depth, not
a substitute for thinking.
Routine writes through the running web application are NOT covered by this rule. Those are normal application behaviour, not "interactions" in the sense meant here.
Where backups go
Default: <dir-of-db-file>/backups/quartermaster-YYYYMMDD-HHMMSS-{slug}.db.
For the standard ./quartermaster.db that resolves to ./backups/.
Override with QUARTERMASTER_BACKUP_DIR=/some/path.
The backups/ directory is .gitignored. Retention is forever:
backups are small. Prune manually if you need to.
Restoring
A backup is a complete SQLite file. To restore, stop the app, replace
quartermaster.db with the chosen backup (cp backups/... quartermaster.db),
then restart.
Current Project State
- Phase: live in production on home-ctr-onyx at
https://quartermaster.unbiasedgeek.com/. Every merge tomainrolls out automatically via.forgejo/workflows/deploy.yml. - Last worked on: 2026-04-19
- Last commit on main:
ee6eaae— fix(docker): enable uvicorn proxy-headers so url_for works behind Traefik - Open PRs: none
- Open issues: #23 MCP proposal; #31 small cleanups (non-blocking polish); #26 and #27 are landed but weren't closed when their work merged — safe to close.
- Test count: 148 / 148 passing
- Migrations: 5 applied; latest
cc60e7f73a1c - Blocking issues: none
After pulling new work, always:
uv run alembic upgrade head
The backup hook fires before any migration, so this is safe against the live DB.
Session Log
Most recent 3 sessions (full history in the wiki).
Session 3 — 2026-04-19
Deploy-pipeline arc: four PRs (#32 Dockerfile, #33 compose.yml,
#34 Forgejo Actions workflow, #35 post-deploy proxy-headers fix)
took Quartermaster from "deploy prep merged" to live on
https://quartermaster.unbiasedgeek.com/. All three dependency-chained
issues (#28/#29/#30) closed.
Key decision, locked mid-flight: the deploy workflow has no SSH
step. Initial draft used an SSH-from-runner-to-host pattern with
DEPLOY_SSH_KEY + DEPLOY_KNOWN_HOSTS secrets. Jeff pushed back —
the homelab runner lives on home-ctr-onyx itself with the host's
Docker socket mounted, so docker compose pull && up -d runs
directly against the same daemon that hosts production. Dropped
two secrets and the private-key risk surface. Remaining Actions
secrets: REGISTRY_TOKEN (archeious PAT, read:package +
write:package, minted via the Forgejo API using the admin
password exposed by homelab-IaC/bin/load-ops-secrets) and
QUARTERMASTER_SMOKE_PASSWORD (plaintext basic-auth for the
post-deploy /healthz probe).
Other design points: image tag parameterised via
QUARTERMASTER_TAG (workflow writes a per-deploy .env),
COMPOSE_PROJECT_NAME=quartermaster pinned so the runner's
ephemeral workspace path doesn't confuse compose, smoke step does
curl -u admin:… https://quartermaster.unbiasedgeek.com/healthz
to catch TLS + routing + basic-auth regressions in one probe.
One post-deploy bug: the first rolled image rendered unstyled
because uvicorn was started without --proxy-headers, so
Starlette ignored X-Forwarded-Proto from Traefik and url_for()
generated http://<internal>/static/… hrefs, which browsers
blocked as mixed content on the https:// page. Reproduced
locally by curling the pre-fix image with Traefik-style headers;
added --proxy-headers --forwarded-allow-ips='*' to
docker/entrypoint.sh in #35. Safe to trust all forwarded IPs
because compose.yml publishes no host port — only Traefik on
proxy-net can reach port 8000.
Full retro: Session3.
Session 2 — 2026-04-19
Platform contract intake (#25) filled out and accepted; platform team
provisioned DNS (quartermaster.unbiasedgeek.com), Traefik middlewares
(basic-auth + rate-limit), the /mnt/quartermaster/ bind mount, and
basic-auth creds. Two issues landed on main (#26 /healthz, #27
structured JSON logs) via the superpowers brainstorm → spec → plan →
subagent-driven-TDD workflow. Thirteen commits on a single branch
(feat/platform-deploy-prep), rebased onto origin/main so history
stays linear despite the unrelated MCP-doc PR that landed alongside.
Key decisions: single-source-of-truth logconfig.json (not a Python
dict + YAML shim, which would need pyyaml and introduce drift); five
seed app events (month_created, month_closed,
template_entry_updated, posting_added, posting_deleted) placed
after commit + refresh so they only fire on durable success; /healthz
on a dedicated router as a file boundary, with auth living in Traefik
per the platform contract.
Two real misses caught by code review mid-flight: the autouse-fixture
workaround for Task 3's dictConfig state leak was wrong (fix: make
the contaminating test save/restore state itself); the third LogQL
example in README used {{.path}} on month_closed records that
carry year_month, not path. Both landed as fix commits.
Deploy-pipeline work queued as #28 (Dockerfile), #29 (compose.yml), #30 (Forgejo Actions), dependency-chained. Polish umbrella at #31.
Full retro: Session2.