Build and ship both halves from one commit
Check / check (push) Successful in 2m29s
Check / guardrails (push) Failing after 29s
Check / bundle (push) Successful in 1m6s

The audit's other "do it differently" (F76). v1's backend was an image built by
CI; its frontend was `rsync -azX --delete` from a laptop, run by hand, with the
API's absolute URL compiled into the WASM by an environment variable set in the
same script. Two artefacts, one pipeline, and nothing keeping them in step.

Here there is one image with the server binary at `/usr/local/bin` and the
built frontend at `/srv/dist`, and the frontend asks for `/api/...` relative to
wherever it is served -- so the artefact is the same in every environment and
the two halves cannot be deployed apart. Caddy still serves the static files
from a directory, because that is what the reverse proxy in front of everything
already does; the updater lifts them out of the image rather than out of a
build on somebody's machine. It stages them and moves the directory into place,
since `index.html` names hashed files and a browser that fetches new HTML with
old JavaScript gets a blank page.

The image is 97 MB and holds no build tooling. Migrations are compiled into the
binary and run at startup, and the database is created if it is missing, so
there is no `sqlx-cli`, no entrypoint script, and nothing that can decide to
carry on after a failed migration -- v1's start.sh ran migrations with
`|| echo "Migration failed but continuing..."`. TLS is rustls with its roots
compiled in, so there is no OpenSSL to keep patched. There is no dummy-source
dance for dependency caching either; a BuildKit cache mount does what that
trick was inventing, and the binary is copied out of the mount because a cache
mount is not part of the layer.

Deployment is a timer on the server rather than CI reaching into it. Nothing in
the workflow holds a credential for the machine it deploys to, and a bad build
cannot take the site down on its own; the cost is a few minutes between push
and deploy, and `deploy/runway-update` for when that is too long.

The rest of this is the CI that was promised at M1 and never written.

"Rules that only live in a doc get forgotten, so these are lints and CI checks."
The lints landed. The CI half did not exist, which meant for twenty-seven
milestones the forgetting was still perfectly possible -- it was just mine
rather than the repository's. `cargo fmt`, `clippy -D warnings` and the whole
test suite now run on every push, alongside the three guardrails that were only
ever configuration:

`cargo machete` passes. `deny.toml`, written at M1, did not: three permissive
licences its allow-list had not anticipated, and two `unmaintained` advisories
arriving through Leptos's macros. Both are now allowed by name with a date and
a reason rather than by widening a category. MPL-2.0 came out of the list,
because nothing uses it and an allow-list should say what is actually there.
That file had been quietly wrong since the day it was written, which is the
argument for CI in one line.

The bundle budget is 1.8 MB against today's 1.31 MB and v1's 2.5 MB, printed on
every run. A bundle grows one convenient dependency at a time.

The print rule is deliberately not in CI: it already exists as a test, it
covers the CalDAV client too, and a shell grep cannot tell a call from a comment
about a call -- the first draft of that step failed on the paragraph in
`observability.rs` that explains the rule.

Two things the image found that reading could not.

`/db` is created in the image now. SQLite creates the database file if it is
missing but not the directory holding it, so the container started only when
something happened to be mounted there and otherwise died with "unable to open
database file", which says nothing about what is wrong.

And the graceful shutdown only listened for Ctrl-C. A container runtime stops a
service with SIGTERM, waits ten seconds, and sends SIGKILL -- so `podman stop`
took ten seconds and killed the process outright, and the handler written to
stop a restart dropping a CalDAV write half-way through worked everywhere
except deployment, which is the one place restarts happen. It listens for both
now, and the container stops in two.
This commit is contained in:
2026-08-28 12:47:53 -04:00
parent 907a0f2b01
commit b0da3afafc
11 changed files with 558 additions and 5 deletions
+9
View File
@@ -0,0 +1,9 @@
target
node_modules
crates/runway-web/dist
crates/runway-web/styles/generated.css
e2e/screenshots
.git
*.db
*.db-shm
*.db-wal
+121
View File
@@ -0,0 +1,121 @@
# The checks the audit said would be machine-enforced.
#
# They were written down at M1 as "lints and CI checks" on the grounds that
# rules living only in a document get forgotten. The lints landed; the CI half
# did not exist until now, which means for twenty-seven milestones the
# forgetting was still possible -- it was just my forgetting rather than the
# repository's.
name: Check
on:
push:
pull_request:
jobs:
check:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
# The toolchain, its components and the wasm target all come from
# rust-toolchain.toml, so there is one place that says which Rust this is.
- name: Install the pinned toolchain
run: |
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs \
| sh -s -- -y --default-toolchain none
echo "$HOME/.cargo/bin" >> "$GITHUB_PATH"
- uses: actions/cache@v4
with:
path: |
~/.cargo/registry
~/.cargo/git
target
key: cargo-${{ runner.os }}-${{ hashFiles('Cargo.lock') }}
restore-keys: cargo-${{ runner.os }}-
- name: Formatting
run: cargo fmt --all --check
- name: Clippy
run: cargo clippy --workspace --all-targets -- -D warnings
# The CalDAV integration tests skip themselves when there is no server to
# talk to, which is what happens here. They are exercised against a real
# Baikal by `crates/runway-caldav/tests/baikal/run.sh`.
- name: Tests
run: cargo test --workspace
- name: The frontend still builds for the browser
run: cargo check -p runway-web --target wasm32-unknown-unknown
guardrails:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- name: Install the pinned toolchain
run: |
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs \
| sh -s -- -y --default-toolchain none
echo "$HOME/.cargo/bin" >> "$GITHUB_PATH"
# v1 shipped reqwest, ical, regex and indexed_db_futures into a WASM
# bundle that used none of them.
- name: No unused dependencies
run: |
cargo install cargo-machete --locked
cargo machete
- name: Licences and advisories
run: |
cargo install cargo-deny --locked
cargo deny check
# v1 had 157 of these in its backend, one of which logged the length of a
# password. Tests may print; the server may not.
# The print rule is not here on purpose. It already exists as a test --
# `no_println_or_dbg_in_the_server_or_the_caldav_client` -- so `cargo
# test` above enforces it, and it covers the CalDAV client too. Doing it
# again here would mean two rules to keep in step, and the CI copy would
# be the worse of the two: grep from a shell cannot tell a call from a
# comment about a call, and this file's first draft failed on the
# paragraph in `observability.rs` explaining the rule.
bundle:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- name: Install the pinned toolchain
run: |
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs \
| sh -s -- -y --default-toolchain none
echo "$HOME/.cargo/bin" >> "$GITHUB_PATH"
- name: Install Trunk
run: |
curl -fsSL https://github.com/trunk-rs/trunk/releases/download/v0.21.14/trunk-x86_64-unknown-linux-gnu.tar.gz \
| tar -xzC /usr/local/bin trunk
- run: npm ci
- name: Build the frontend
working-directory: crates/runway-web
run: trunk build --release
# Printed on every run and enforced, because a bundle grows one
# convenient dependency at a time and nobody notices until it is 2.5 MB.
# That was v1's. This budget is comfortably above today's and well below
# that, so it is a ratchet rather than a rubber stamp.
- name: WASM bundle size
run: |
wasm=$(find crates/runway-web/dist -name '*.wasm' -print -quit)
size=$(stat -c %s "$wasm")
budget=1800000
echo "$wasm is $size bytes (budget $budget, v1 shipped 2500000)"
if [ "$size" -gt "$budget" ]; then
echo "over budget by $((size - budget)) bytes" >&2
exit 1
fi
+41
View File
@@ -0,0 +1,41 @@
# Builds the image and pushes it. It does not deploy.
#
# What picks it up is a timer on the server (see `deploy/`), so nothing here
# holds a credential for the machine it runs on and a bad build cannot take the
# site down by itself. The trade is that a deploy happens a few minutes after
# the push rather than at the moment of it.
name: Image
on:
push:
branches:
- master
jobs:
image:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- uses: docker/setup-buildx-action@v3
- uses: docker/login-action@v3
with:
registry: ${{ vars.REGISTRY }}
username: ${{ vars.USERNAME }}
password: ${{ secrets.DOCKER_PASSWORD }}
# Tagged with the commit as well as `latest`, so that "what is actually
# running" has an answer, and so a rollback is a tag rather than a revert
# and a rebuild.
- uses: docker/build-push-action@v5
with:
context: .
file: ./Dockerfile
push: true
tags: |
${{ vars.REGISTRY }}/connor/runway:latest
${{ vars.REGISTRY }}/connor/runway:${{ github.sha }}
cache-from: type=gha
cache-to: type=gha,mode=max
+105
View File
@@ -0,0 +1,105 @@
# syntax=docker/dockerfile:1
# One image, holding both halves of the app.
#
# v1 shipped the backend as an image and the frontend as an `rsync` of `dist/`
# over SSH, run by hand from a laptop -- so the two could be, and were, out of
# step with each other, and the audit marked that as the one thing to do
# differently (F76). Here they are built from the same commit into the same
# image and can only be deployed together. The frontend is not served from
# here: it is lifted out of this image onto the directory Caddy already serves,
# which is why it sits at a known path.
#
# The frontend addresses the API as `/api/...`, relative, so this artefact is
# the same everywhere and nothing is baked in per environment. v1 compiled the
# API's absolute URL into the WASM, which is the other half of why its frontend
# deploy was a separate manual step.
# ---------------------------------------------------------------- frontend --
FROM rust:1.98-slim-bookworm AS web
RUN apt-get update \
&& apt-get install -y --no-install-recommends ca-certificates curl \
&& rm -rf /var/lib/apt/lists/*
# Node from the official image rather than Debian's, which ships npm 9.
# Tailwind v4's `oxide` is a native binary chosen by an optional dependency
# keyed on platform *and* libc, and npm 9 does not understand the `libc` field
# -- so it silently installs no binding at all and the build dies on
# `Cannot find module '@tailwindcss/oxide-linux-x64-gnu'`. Copied from a pinned
# image rather than piped from a setup script into a shell, so which Node this
# is stays a line in this file.
COPY --from=node:22-bookworm-slim /usr/local/bin/node /usr/local/bin/node
COPY --from=node:22-bookworm-slim /usr/local/lib/node_modules /usr/local/lib/node_modules
RUN ln -s ../lib/node_modules/npm/bin/npm-cli.js /usr/local/bin/npm \
&& ln -s ../lib/node_modules/npm/bin/npx-cli.js /usr/local/bin/npx
# Trunk as a released binary rather than `cargo install trunk`, which builds it
# from source and takes longer than everything else here put together.
ARG TRUNK_VERSION=0.21.14
RUN curl -fsSL "https://github.com/trunk-rs/trunk/releases/download/v${TRUNK_VERSION}/trunk-x86_64-unknown-linux-gnu.tar.gz" \
| tar -xzC /usr/local/bin trunk
RUN rustup target add wasm32-unknown-unknown
WORKDIR /app
# Tailwind runs from Trunk's pre-build hook as `npx --prefix ../..`, so the
# packages have to be at the workspace root. Copied before the source so a
# change to the Rust does not reinstall them.
COPY package.json package-lock.json ./
RUN npm ci
COPY . .
WORKDIR /app/crates/runway-web
RUN --mount=type=cache,target=/usr/local/cargo/registry \
--mount=type=cache,target=/app/target \
trunk build --release
# ----------------------------------------------------------------- backend --
FROM rust:1.98-slim-bookworm AS server
WORKDIR /app
COPY . .
# The binary is copied out of the cache mount because a cache mount is not part
# of the layer: whatever is written there is gone by the time the next stage
# looks. This is also why there is no dummy-source dance -- the cache does what
# that trick was inventing.
RUN --mount=type=cache,target=/usr/local/cargo/registry \
--mount=type=cache,target=/app/target \
cargo build --release --locked -p runway-server \
&& cp target/release/runway-server /runway-server
# ----------------------------------------------------------------- runtime --
FROM debian:bookworm-slim AS runtime
# TLS is rustls with its roots compiled in, so there is no OpenSSL here and
# nothing to keep patched. `tzdata` is not optional: this is a calendar, and
# recurrence over a DST boundary is decided by the zone database.
RUN apt-get update \
&& apt-get install -y --no-install-recommends ca-certificates tzdata \
&& rm -rf /var/lib/apt/lists/*
COPY --from=server /runway-server /usr/local/bin/runway-server
# Where the updater looks for the frontend. Nothing serves it from inside this
# container.
COPY --from=web /app/crates/runway-web/dist /srv/dist
# Migrations are compiled into the binary by `sqlx::migrate!` and run on
# startup, and the database is created if it is missing -- so there is no
# entrypoint script, no `sqlx-cli` in the image, and nothing that can decide to
# carry on after a failed migration the way v1's start.sh did.
# The mount point, created here rather than left to the volume. SQLite creates
# the database *file* if it is missing but not the directory holding it, so an
# image without this starts only when something happens to have mounted
# something at /db, and fails with "unable to open database file" when nothing
# has -- which says nothing about what is actually wrong.
RUN mkdir -p /db
ENV RUNWAY_DATABASE_URL=sqlite:///db/runway.db \
RUNWAY_BIND=0.0.0.0:3000
EXPOSE 3000
CMD ["runway-server"]
+14
View File
@@ -105,6 +105,20 @@ cargo run -p runway-cli -- calendars
cargo run -p runway-cli -- list-events --from 2026-08-24 --to 2026-08-31
```
## Deployment
One image holds both halves — the server binary and the built frontend — so
they cannot be deployed out of step with each other. CI builds and pushes it on
every merge to `master`; a timer on the server picks it up. Nothing goes up
from a laptop.
```sh
deploy/runway-update # what the timer runs, if you would rather not wait
```
The full arrangement, and the one-time setup it needs, is in
[`deploy/README.md`](deploy/README.md).
## Principles
Carried over from the audit, and enforced by lints and CI rather than by good intentions:
+37 -4
View File
@@ -53,9 +53,42 @@ async fn run() -> Result<(), Box<dyn std::error::Error>> {
Ok(())
}
/// Lets in-flight requests finish on Ctrl-C, so a restart does not drop a save
/// half-way through a CalDAV write.
/// Lets in-flight requests finish before the process goes, so a restart does
/// not drop a save half-way through a CalDAV write.
///
/// Both signals, because the two ways this process gets stopped send different
/// ones. Ctrl-C at a terminal sends SIGINT; a container runtime stopping a
/// service sends SIGTERM, waits ten seconds, and then sends SIGKILL. Listening
/// only for the first meant the handler worked everywhere except deployment --
/// which is the one place restarts actually happen, and the one place a
/// half-written CalDAV request matters.
async fn shutdown() {
let _ = tokio::signal::ctrl_c().await;
tracing::info!("shutting down");
#[cfg(unix)]
{
use tokio::signal::unix::{SignalKind, signal};
let mut terminate = match signal(SignalKind::terminate()) {
Ok(signal) => signal,
// Nothing to be done about it, and refusing to serve because a
// signal handler could not be installed would be worse.
Err(error) => {
tracing::warn!(%error, "cannot listen for SIGTERM");
let _ = tokio::signal::ctrl_c().await;
tracing::info!(signal = "SIGINT", "shutting down");
return;
}
};
let signal = tokio::select! {
_ = tokio::signal::ctrl_c() => "SIGINT",
_ = terminate.recv() => "SIGTERM",
};
tracing::info!(signal, "shutting down");
}
#[cfg(not(unix))]
{
let _ = tokio::signal::ctrl_c().await;
tracing::info!(signal = "SIGINT", "shutting down");
}
}
+21 -1
View File
@@ -1,5 +1,21 @@
# This file was written at M1 and, until M28 wired it into CI, had never been
# run against the dependency tree it describes. It did not pass. Everything
# below with a date on it is what that first run turned up -- which is the
# argument for CI in one paragraph: a policy nobody executes is a document, and
# this one had been quietly wrong for twenty-seven milestones.
[advisories]
yanked = "deny"
ignore = [
# Both of these are `unmaintained`, not vulnerabilities, and both arrive
# through Leptos's macros rather than through anything this workspace asks
# for. They run at build time and none of their code ships. Reviewed
# 2026-08-28; the honest options are to ignore them or to stop using
# Leptos, so they are ignored, by id, with this note rather than by
# turning the whole `unmaintained` class off.
"RUSTSEC-2024-0436", # paste, via leptos_macro
"RUSTSEC-2026-0173", # proc-macro-error2, via leptos_macro
]
[licenses]
allow = [
@@ -11,7 +27,11 @@ allow = [
"ISC",
"Unicode-3.0",
"Zlib",
"MPL-2.0",
# Added 2026-08-28, when this was first run. All three are permissive and
# all three arrive transitively:
"CC0-1.0", # base16, via leptos
"CDLA-Permissive-2.0", # webpki-roots -- the TLS root store itself
"BSL-1.0", # xxhash-rust
]
[bans]
+121
View File
@@ -0,0 +1,121 @@
# Deploying Runway
The app is one image holding both halves — the server binary and the built
frontend — pushed by CI on every merge to `master` and picked up by a timer on
the server. Nothing is copied from a laptop, and there is no step that can
deploy one half without the other. That was the audit's one "do it differently"
(F76): v1's frontend went up by `rsync` over SSH, by hand, separately from its
backend's CI.
## What runs where
| | |
|---|---|
| Image | `git.rcjohnstone.com/connor/runway:latest`, and `:<commit>` |
| Built by | `.gitea/workflows/release.yml`, on push to `master` |
| Deployed by | `deploy/runway-update`, from a systemd timer |
| Backend | `runway-backend` in `~/compose.yml`, on the `internal` network |
| Frontend | `~/data/runway/dist`, served by the root Caddy from `/srv/runway` |
| Database | `~/data/runway/db`, a volume on the backend container |
| TLS and auth | The root Caddy, which already fronts `runway.rcjohnstone.com` behind Authelia |
The frontend asks for `/api/...` relative to wherever it is served, so the same
image is correct in every environment and nothing is compiled in per host. v1
baked the API's absolute URL into the WASM, which is why its frontend could not
be deployed by the same pipeline as its backend.
## Setting it up
Five things, of which four are one-time.
**1. The image.** Set on the Gitea repository, under Settings → Actions:
- variable `REGISTRY` = `git.rcjohnstone.com`
- variable `USERNAME` = `connor`
- secret `DOCKER_PASSWORD` = a token with package write
These are the names v1's workflow used, so they may already exist at the
organisation level.
**2. The signing key.** Generate it once, on the server, and keep it:
```sh
podman run --rm git.rcjohnstone.com/connor/runway:latest runway-server genkey
```
Put the output in `~/.env` as `RUNWAY_SECRET_KEY`. It encrypts the stored
CalDAV passwords, so **if it changes, every saved credential becomes
unreadable** and everybody has to sign in again. It is not derived from
anything and cannot be recovered; back it up with the rest of `~/.env`.
**3. The compose service.** Replace the existing `runway-backend` block:
```yaml
runway-backend:
image: git.rcjohnstone.com/connor/runway:latest
restart: unless-stopped
networks:
- internal
environment:
# Absent on purpose: RUNWAY_INSECURE_COOKIES. Caddy terminates TLS, so
# the session cookie is Secure, and setting this would quietly stop it
# being so.
- RUNWAY_SECRET_KEY=${RUNWAY_SECRET_KEY}
- RUNWAY_DATABASE_URL=sqlite:///db/runway.db
- TZ=America/Louisville
volumes:
- ./data/runway/db:/db
```
`~/data/runway/db` is root-owned from v1 and the container still runs as root,
so it keeps working untouched. To run it unprivileged instead, `chown` that
directory and add a `user:` line.
The database is `runway.db`, not v1's `calendar.db`, so this starts empty —
which is what you want. There is nothing to migrate across: the schemas share
no ancestry, and v1's stored passwords were encrypted with a key this app does
not have.
**4. Caddy.** No change. The existing `runway.rcjohnstone.com` block already
proxies `/api/*` to `runway-backend:3000` and serves `/srv/runway` with an
`index.html` fallback, which is exactly what this needs.
**5. The timer.**
```sh
mkdir -p ~/.config/systemd/user
ln -sf ~/docs/projects/runway/deploy/runway-update.service ~/.config/systemd/user/
ln -sf ~/docs/projects/runway/deploy/runway-update.timer ~/.config/systemd/user/
systemctl --user daemon-reload
systemctl --user enable --now runway-update.timer
loginctl enable-linger connor # so it runs when you are not logged in
```
## Deploying by hand
The timer runs the same script, so this is what it does, and it is also the
answer to "I do not want to wait ten minutes":
```sh
deploy/runway-update
```
It exits without touching anything if the image has not changed.
## Rolling back
The frontend of the previous deploy is kept:
```sh
mv ~/data/runway/dist ~/data/runway/dist.bad
mv ~/data/runway/dist.previous ~/data/runway/dist
```
The backend is a tag. Pin the service's `image:` to
`git.rcjohnstone.com/connor/runway:<commit>` and bring it up; the timer will
leave a pinned tag alone, because it only ever pulls what the service names.
Rolling back **across a migration** is the case to think about before doing it:
migrations only go forwards, so an older binary meeting a newer database is not
something to try casually. Roll the frontend back first — that is usually the
half that is wrong, and it is reversible in a way the schema is not.
+65
View File
@@ -0,0 +1,65 @@
#!/usr/bin/env bash
# Pull the current image and put both halves of it in place.
#
# Two halves, because Caddy serves the frontend from a directory and the
# backend from a container -- so a deploy is "restart the container" *and*
# "refresh the directory", and they have to come from the same image or the
# thing v1 got wrong (a frontend calling an endpoint the backend does not have)
# comes straight back. Nothing here builds anything; the image is whatever CI
# pushed.
#
# Safe to run when nothing has changed: if the pull brings down no new layers
# the digest is unchanged and it stops.
set -euo pipefail
IMAGE="${RUNWAY_IMAGE:-git.rcjohnstone.com/connor/runway:latest}"
COMPOSE_DIR="${RUNWAY_COMPOSE_DIR:-$HOME}"
SERVICE="${RUNWAY_SERVICE:-runway-backend}"
DIST="${RUNWAY_DIST:-$HOME/data/runway/dist}"
runtime() {
if command -v podman >/dev/null 2>&1; then echo podman
elif command -v docker >/dev/null 2>&1; then echo docker
else echo "need podman or docker" >&2; exit 1
fi
}
rt="$(runtime)"
before="$("$rt" image inspect --format '{{.Id}}' "$IMAGE" 2>/dev/null || true)"
"$rt" pull --quiet "$IMAGE" >/dev/null
after="$("$rt" image inspect --format '{{.Id}}' "$IMAGE")"
if [ "$before" = "$after" ]; then
echo "runway: already on ${after:7:12}, nothing to do"
exit 0
fi
echo "runway: ${before:7:12} -> ${after:7:12}"
# The frontend first. It is copied into place through a staging directory and
# moved, so that Caddy is never serving a half-written bundle -- `index.html`
# names hashed files, and a browser that fetches the new HTML and the old
# JavaScript gets a blank page.
staged="$(mktemp -d "${DIST%/*}/.dist.XXXXXX")"
container="$("$rt" create "$IMAGE")"
trap '"$rt" rm -f "$container" >/dev/null 2>&1 || true; rm -rf "$staged"' EXIT
"$rt" cp "$container:/srv/dist/." "$staged/"
# Kept rather than deleted: rolling the frontend back is then a `mv`, which is
# the kind of thing you want to be easy at the moment you need it.
previous="${DIST%/}.previous"
rm -rf "$previous"
# Spelled as an `if` and not `[ -d ... ] && mv`, because under `set -e` the
# second form exits the script the first time it runs, when there is nothing
# there yet -- and it would have looked like the deploy succeeded.
if [ -d "$DIST" ]; then
mv "$DIST" "$previous"
fi
mv "$staged" "$DIST"
chmod -R a+rX "$DIST"
# Then the backend, which runs the migrations as it starts.
( cd "$COMPOSE_DIR" && "$rt" compose up -d "$SERVICE" )
echo "runway: ${after:7:12} deployed"
+8
View File
@@ -0,0 +1,8 @@
[Unit]
Description=Deploy whatever CI last built of Runway
Documentation=https://git.rcjohnstone.com/connor/runway
After=network-online.target
[Service]
Type=oneshot
ExecStart=%h/docs/projects/runway/deploy/runway-update
+16
View File
@@ -0,0 +1,16 @@
[Unit]
Description=Check for a new Runway image
[Timer]
# Ten minutes after boot, then every ten. The check is a registry HEAD when
# nothing has changed, which is cheap enough that the interval is chosen for
# how long you are willing to wait after a push rather than for load.
OnBootSec=10min
OnUnitActiveSec=10min
# So that two services updating on the same schedule do not both wake at
# exactly the same second for ever.
RandomizedDelaySec=60
Persistent=true
[Install]
WantedBy=timers.target