menu-admin was the odd one out: it built from the working directory with a date
tag and a multi-arch manifest, deployed with `-e tag=`, and kept the Docker-era
container name. The other two apps build from `git archive HEAD` and tag with the
commit, so what runs on the host is exactly what git knows about; build only for
the architecture emily actually runs; and use one slug for the container, its
network and its data directory.
The build is now their three lines: `podman build --platform linux/arm64` from
`git archive HEAD`, tagged `:<commit>` and `:latest`, pushed with `podman push`
rather than a manifest. `deploy.sh` pushes the commit first, then refuses a dirty
tracked tree or a HEAD that is not origin/main — which is what makes the commit
tag trustworthy — takes the same `--force` and `--no-push`, and ends in the same
`-u tanshu ... -e image_tag=<commit>`. Its own lint/typecheck/test gate stays,
but behind `--checks`, so a plain deploy behaves as theirs does.
The registry credential moves into vars/default.yml and the role logs in, pulls
and logs out around it, as their roles do. The image is only pulled at deploy
time, so the unit keeps podman's default `Pull=missing` and a reboot never needs
the registry. This is the same plaintext credential the other two repos carry.
One real bug goes with it: the healthcheck gate ran before the handler that
restarts a changed unit, so a run that only changed the unit checked the stale
container. `meta: flush_handlers` before the gate is the fix the sibling repos
already made.
The container, network and data directory are all menu-<site> now. The dotted
Docker-era name and the NetworkAlias it needed are gone, so `podman logs
menu-<site>` and `systemctl --user status menu-<site>` are the same two commands
across all three apps. The Dockerfile keeps its name for the same reason: podman
reads it as readily as a Containerfile, and the siblings use that name too.
The fleet on emily is rootless Podman with Quadlet units under
~/.config/containers/systemd and state in ~/podman-data, all behind one shared
Caddy unit. This brings the menus in line instead of leaving them the last
Docker deployment. The build moves with them: podman reads a Containerfile as
readily as a Dockerfile, so the name now says which toolchain is real.
The app is one .container and one .network file. Two details carry it.
UserNS=keep-id maps the image's own node account onto the user that owns the
units, so data/ and site.env stay owned by tanshu: plain tar backups, no
subuids. provision-site-env.sh writes site.env with the same mapping rather
than --user, which under rootless Podman would have produced a subuid-owned
file the app could not read. And the Site registers its own network with Caddy
by writing a drop-in beside caddy.container, so Caddy never lists its upstreams
and never changes when a Site is added. The unit keeps the Docker-era container
name for the scripts and muscle memory that look it up, while the Caddyfile
proxies to a dotless NetworkAlias, because a container name with dots in it is
a liability in DNS.
The unit keeps podman's Pull=missing, so a restart or a reboot never depends on
the registry being reachable, while the play pulls explicitly at deploy time.
That is also where a missing `podman login` for the private registry now fails,
with the instruction, rather than as a unit that will not start.
copy-site-data.sh moves a Site's data across, and is not a plain rsync:
/var/lib/menu.* on monoco belongs to uid 1000 while the ssh user there is 1001,
so the read needs sudo; monoco cannot ssh to emily, so the workstation relays
the tar; and the extraction runs as tanshu on emily, which is what makes the
files land owned by the user keep-id maps the app onto. It proves the copy by
comparing a SHA-256 of the whole tree against the source and checking every
file's owner.
Not yet done: the copies themselves and the DNS move, which need a human.
There is no previous system kept serving, so the playbook no longer knows about
one: the retire flags, the temporary web address, the task that removed the old
nginx snippet and the task that stopped the old container are gone. One run
starts the container and points the Site's web address at it, and deploying an
earlier tag is the rollback. The vars lose vhost, retire_nginx_snippet,
retire_nginx_container and legacy_container; the Caddy role keeps only the
snippet upsert.
The deployment readme drops its staged-rollout sections and the glossary of words
that no longer exist, and the plan's sequence and the operating notes describe
the single run instead.
ESLint cannot parse this project's TypeScript: typescript@7 is the native port
with no JavaScript compiler API, which every ESLint TypeScript parser needs, and
typescript-eslint refuses to install with a >=4.8.4 <6.1.0 peer. oxlint parses
TypeScript without the compiler API, so it lints the .ts and the .js together in
about 20ms over the whole repo, and it runs the correctness category as errors.
It found one real thing on the first run: the drag harness built its fetch
options with a body key even for GET. The four opinionated rules it also flagged
are off with their reasoning in the README, since oxlint's config takes no
comments. lint is back in deploy.sh's checks, and the editor has lint tasks and
the oxc extension recommendation. Prettier is unchanged.
The playbook, its roles, its vars and a new inventory now live in ansible/,
beside the app they deploy, the way the other apps on this machine are laid out.
The inventory follows the same shape: one entry per Site, all on the same box,
each naming its own vars file through var_file, so adding a Site is one line and
a file rather than a command-line flag. ansible_host is written as the
resolvable name rather than an ssh alias so the inventory works on any machine.
The image is now built with podman as a multi-arch manifest and pushed to
registry.tanshu.com/menu-admin, mirroring the Makefile used elsewhere on this
machine: build-production, plus build-check and build-check-local for a build
that does not push. deploy.sh runs the checks, builds, pushes, then runs the
playbook for the Site named by SITE, with SKIP_DEPLOY for a push without a
deploy.
ESLint comes out again: this project's typescript@7 has no JavaScript compiler
API, so typescript-eslint cannot install, and linting only the JavaScript was
not worth keeping. Prettier stays, and the README records why ESLint is absent
and what would bring it back.
Prettier owns formatting: .prettierrc.json, format and format:check scripts, and
a one-off pass over the repo so the check is green from the start. Embedded
language formatting is off deliberately - the renderers build HTML in html
tagged templates and letting Prettier reflow those would change the markup
customers receive. docs/ is excluded because it is hand-wrapped prose, and shell
scripts have no Prettier parser.
ESLint covers the JavaScript: the two browser scripts, the build and check
scripts, and its own config. no-var is scoped out for the browser scripts on
purpose; they are plain ES5-shaped IIFEs and a hoisting difference would only
show up in a browser.
ESLint cannot lint the TypeScript here: typescript@7 is the native port with no
JavaScript compiler API, and typescript-eslint requires a 4.8-6.1 peer, so it
refuses to install and forcing it through would leave a parser that cannot read
the compiler. tsc --noEmit stays the TypeScript gate; README records the two
ways to change that.
lint and format:check now run in deploy.sh, and VS Code formats on save with
Prettier and surfaces ESLint findings.
Categories were baked into config/<site>.site.json, so adding one meant a
rebuild and a redeploy. They are now data: data/sections.json, seeded from the
config the first time, and managed in the admin (add, rename, reorder, remove).
A category's web address is fixed when it is created because that is what a QR
code points at, and a category cannot be removed while it still holds menus.
The landing and the top bar were the ugliest parts. The landing was a column of
hairlines with tiny meta — a broadsheet, and the paper that makes this app worth
looking at was nowhere on it. It is now a shelf: each category is a row whose
right edge is a sliver of that menu's first page, bleeding off the screen, with
the shelf edge in brass and the name in heavy tight Archivo. The bar was a row
of teal pills that would not survive more categories: it now names the menu you
are reading and opens a sheet listing the rest, each with its own paper sliver,
and the current one is the only teal thing on screen. The sheet is a link to the
landing page, so navigating still works without the script.
Also fixes the category manager's own layout, which I shipped unstyled, and the
markup contract the route tests described.
Two defects, both found by driving the real editor in Firefox:
1. The editor drew grab handles with a grab cursor on sets it refuses to
reorder, so the live Food menu looked draggable and did nothing. Handles and
the move buttons are now rendered only when reordering is possible, and the
notice says what can still be done.
2. Reordering was saved only from dragend. Browsers suppress mouse events during
a drag and dragend is not guaranteed to arrive, which left the rows visibly
reordered while the server kept the old order — the drag looked like it
worked and then reverted on reload. The order now commits from dragover after
a pause in the drag, with drop/dragend/mouse events as faster paths, deduped
against the last confirmed order and reported in the page.
To make (1) coherent the Current set may now be reordered while its Pages stay
fixed: order is presentation, media URLs are content-hashed and pages are
rendered per request, so nothing stale can be served (ADR 0006). Archives stay
frozen. Admin pages are no-store and assets cache for 5 minutes, so a fix like
this one is not masked by a stale admin.js.
scripts/check-dnd.mjs drives the real editor and asserts the drag persists; it
is the only automated check that reaches this path.
npm run dev now runs scripts/dev-env.ts first, which creates .dev/site.env with
a known dev password and a random session secret, so a fresh clone serves a
working admin without anyone hand-building a hash. Adds .vscode launch
configurations and tasks for the dev server, one test file, and check-site, and
brings scripts/ under typecheck, which it had been missing.
A file past the size limit makes the multipart stream itself throw, which
escaped the per-file catch and reached the error handler: the phone that just
sent a 35 MB photo got 'Something went wrong'. Catch the limit errors around the
part loop, say which limit was hit, and report it once rather than twice. Also
asserts that derivatives carry no EXIF while the Original keeps the uploaded
bytes.
Every mutating form now carries the CSRF token as a hidden field, and the
upload route accepts it as a multipart field as well as a header. Driving the
real UI in Firefox found what the header-only tests could not: Create Draft,
Save, Publish, Delete and the up/down buttons all posted without a token and
were rejected. Adds npm run check-site, which verifies a running Site from the
outside and exits non-zero so it can gate a cutover.
The playbook pulls the tag from the registry, creates the data root owned by the
app user, runs the container with its config baked into the image, waits for
health, then verifies the data directory as the app user so a permissions
mistake surfaces at deploy time. Caddy is pointed at the app with a stable
marker, so a temporary hostname can be checked beside the live site and the
cutover is one invocation with the retire flags set.
The admin signs in with the shared password, creates a Menu Set, drops photos
into it (shrunk in the browser first), reorders Pages by dragging or with
up/down buttons, previews the Draft exactly as customers would see it, and
publishes it over the previous set. Restore and delete close the loop. Mutations
need a session, a same-origin request and a CSRF token derived from the session;
a Draft's media is served only to a session holder.
Landing page, Section pages and archived sets rendered per request from the
store, with the approved design: hand-written CSS, self-hosted Archivo, a
copper pour line in the Section rail, an admin-only Thumbnail and a text page
index. Media is keyed by Page id so URLs survive publishing, and a Draft's
media needs a session. Includes the scrypt admin auth and the login throttle
the media rule needs.
The store keeps a Menu Set's state in the directory layout (drafts/ versus
published/) and its metadata in a manifest, with publishing as a single
directory rename. Editing is allowed on Drafts only, so a published menu can
never change under a customer. Stages 2 and 3 of docs/plan.md, with tests.
tanshu