Compare commits

...
3 Commits
Author SHA1 Message Date
Bendik Aagaard LynghaugandClaude Fable 5 8858336ecc Semver releases cut with cargo-release; publish on v* tags only
Test / test (push) Successful in 23s
Publish release / publish (push) Successful in 1m8s
cargo release <level> bumps, tags v<semver>, and pushes; publish.yml
reacts to the tag and attaches the artifact to that release. Plain
main pushes now run tests (test.yml) instead of publishing build-<sha>
artifacts on every push.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-08-24 21:44:31 +02:00
Bendik Aagaard LynghaugandClaude Fable 5 2b593ba53f Deploy becomes publish: portal ships as a versioned release artifact
Publish release / publish (push) Successful in 1m7s
The portal repo no longer deploys instances. CI builds once, packages
portal + question_lint + site/ into a tarball, and publishes it as a
Gitea release tagged build-<shortsha> using the run's ephemeral token.
Content repos (uhhm/questions, redoal/questions) now own their instance
deploys - env, unit restart, Caddy route - pinned to a release tag.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-08-24 21:21:06 +02:00
Bendik Aagaard LynghaugandClaude Fable 5 e0094f2e97 Trigger deploy: redoal OAuth2 credentials now configured
Deploy / deploy (push) Successful in 1m8s
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-08-23 20:39:16 +02:00
4 changed files with 136 additions and 187 deletions
-187
View File
@@ -1,187 +0,0 @@
name: Deploy
on:
push:
branches: [main]
jobs:
deploy:
runs-on: bare
env:
# The bare runner's own systemd service intentionally has a minimal
# PATH/HOME (no rustup default toolchain in reach) - point it at the
# shared toolchain install directly rather than assuming an ambient
# dev shell environment.
CARGO_HOME: /var/local/cargo
RUSTUP_HOME: /var/local/rustup
PATH: /var/local/cargo/bin:/usr/local/sbin:/usr/local/bin:/usr/bin
# Shared with interactive dev builds (see ~/.config/fish/config.fish)
# so a crate compiled once, by either a manual build or CI, is
# cached for the other - real cache hits, not just a warm toolchain.
# SCCACHE_SERVER_PORT deliberately differs from the interactive
# dev shell's (4227): sccache's server is discovered by a fixed
# TCP port shared by every local user, so if both contexts used
# the same port, whichever one's server happened to be running
# would silently "win" and serve build requests it doesn't have
# filesystem permission for. Separate ports keep each context's
# own server answering its own requests; the cache directory
# (not the server process) is what's actually shared.
SCCACHE_DIR: /var/local/sccache
SCCACHE_SERVER_PORT: "4228"
CARGO_TARGET_DIR: /var/local/cargo-target
# cargo-leptos's own downloaded tools (wasm-bindgen, wasm-opt) cache
# under $XDG_CACHE_HOME - left at its default that resolves inside
# this unit's systemd StateDirectory, files it creates there end up
# owned by the kernel's overflow "nobody" uid instead of the
# runner's own dynamic uid (a StateDirectory quirk, not something
# in our control), so a later run can't execute what an earlier run
# downloaded. Redirecting it to our own known-good shared dir avoids
# that entirely.
XDG_CACHE_HOME: /var/local/leptos-cache
steps:
- uses: actions/checkout@v4
# SITE_NAME is read via option_env! (src/app.rs) - compile-time,
# not a runtime env var - so it has to be set here, not just in
# the "Write service env" step below.
- name: Build
run: SITE_NAME=${{ vars.PORTAL_SITE_NAME }} cargo leptos build --release
# cargo-leptos only builds the leptos bin-target ("portal") - the
# question-lint utility binary needs its own plain cargo build.
- name: Build question-lint
run: cargo build --release --bin question_lint --features ssr
- name: Ship release
run: |
set -euo pipefail
rel="/srv/app/uhhm-portal/releases/${{ github.sha }}"
mkdir -p "$rel"
cp "$CARGO_TARGET_DIR/release/portal" "$rel/uhhm-portal"
# questions' own CI execs this directly by path (same bare-metal
# runner/host as this job, no artifact download needed) instead
# of running its own hand-maintained yq/jq subset of these rules.
cp "$CARGO_TARGET_DIR/release/question_lint" "$rel/question_lint"
# site-root ("target/site" in Cargo.toml) is project-relative,
# not affected by CARGO_TARGET_DIR - only the plain `cargo build`
# outputs (release/, front/) move with that override.
cp -r target/site "$rel/site"
ln -sfn "$rel" /srv/app/uhhm-portal/current
# Sourced from this repo's own Settings -> Actions Variables/Secrets,
# not typed onto the host by hand - see the infrastructure repo's
# deploy-runner plan for the exact names/values to configure once.
- name: Write service env
run: |
cat > /etc/app/uhhm-portal.env <<EOF
NATS_URL=${{ secrets.PORTAL_NATS_URL }}
KANIDM_URL=${{ vars.PORTAL_KANIDM_URL }}
OAUTH2_CLIENT_ID=${{ vars.PORTAL_OAUTH2_CLIENT_ID }}
OAUTH2_CLIENT_SECRET=${{ secrets.PORTAL_OAUTH2_CLIENT_SECRET }}
PUBLIC_URL=${{ vars.PORTAL_PUBLIC_URL }}
COOKIE_SECURE=true
CONTENT_REPO=https://project.uhhm.no/uhhm/questions
CONTENT_BRANCH=main
SITE_NAME=${{ vars.PORTAL_SITE_NAME }}
LEPTOS_SITE_ADDR=0.0.0.0:3010
# Cargo.toml's site-root ("target/site") is a build-time path;
# the deploy layout copies the bundle to releases/<sha>/site
# (no target/ prefix) - override so the running binary looks
# in the right place for /pkg/*.
LEPTOS_SITE_ROOT=site
AUTOMATION_READ_TOKEN=${{ secrets.PORTAL_AUTOMATION_READ_TOKEN }}
# Read-only (read:user,read:repository,read:organization),
# used only by the GiteaStarred/GiteaOrgRepos resource sources
# (src/resource.rs) - the repo-contents/repo-info calls
# content loading already makes stay anonymous.
GITEA_API_TOKEN=${{ secrets.PORTAL_GITEA_API_TOKEN }}
EOF
# No sudo: the runner's own unit sets NoNewPrivileges=yes, which
# blocks setuid escalation outright (sudo can't work at all under
# it, regardless of sudoers config) - systemctl talks to PID1 over
# D-Bus instead, authorized by a polkit rule scoped to deploy-runner
# + this unit pattern (see /etc/polkit-1/rules.d/10-deploy-runner.rules
# on the host).
- name: Restart service
run: systemctl restart app@uhhm-portal.service
# No sudo here either - the runner is already in the `docker` group,
# so it can talk to the Docker socket directly.
# Apex and www are separate cookie scopes (no shared Domain
# attribute on the session cookie), but Kanidm's redirect_uri is
# fixed to PUBLIC_URL - a login started on the other host set its
# session cookie there, then landed on PUBLIC_URL's callback with
# an empty session ("no login in progress"). Redirecting www to
# the naked domain keeps every visit on one canonical host
# instead - PUBLIC_URL (repo variable) is set to https://{$DOMAIN}
# to match.
- name: Update Caddy routing
run: |
cat > /etc/caddy/services.d/uhhm-portal.caddy <<'EOF'
www.{$DOMAIN} {
redir https://{$DOMAIN}{uri} permanent
}
{$DOMAIN} {
reverse_proxy host.docker.internal:3010
log {
output file /var/log/caddy/www.log
}
}
EOF
docker exec caddy caddy reload --config /etc/caddy/Caddyfile --adapter caddyfile
# ── redoal.com - second instance, same build ─────────────────
# One binary serves both faces: branding is content-driven
# (site.yaml in each CONTENT_REPO), so only the env differs.
# The compile-time SITE_NAME baked above is just this instance's
# fallback - redoal/questions' site.yaml overrides it.
- name: Ship redoal release
run: |
set -euo pipefail
rel="/srv/app/redoal-portal/releases/${{ github.sha }}"
mkdir -p "$rel"
cp "$CARGO_TARGET_DIR/release/portal" "$rel/redoal-portal"
cp -r target/site "$rel/site"
ln -sfn "$rel" /srv/app/redoal-portal/current
- name: Write redoal service env
run: |
cat > /etc/app/redoal-portal.env <<EOF
NATS_URL=${{ secrets.PORTAL_NATS_URL }}
KANIDM_URL=${{ vars.PORTAL_KANIDM_URL }}
OAUTH2_CLIENT_ID=${{ vars.REDOAL_OAUTH2_CLIENT_ID }}
OAUTH2_CLIENT_SECRET=${{ secrets.REDOAL_OAUTH2_CLIENT_SECRET }}
PUBLIC_URL=https://redoal.com
COOKIE_SECURE=true
CONTENT_REPO=https://project.uhhm.no/redoal/questions
CONTENT_BRANCH=main
LEPTOS_SITE_ADDR=0.0.0.0:3020
LEPTOS_SITE_ROOT=site
# Same read-only token as uhhm's instance - here it also
# backs the gitea_releases source for the private
# redoal/redoal repo, so the owning user needs read access
# there.
GITEA_API_TOKEN=${{ secrets.PORTAL_GITEA_API_TOKEN }}
EOF
- name: Restart redoal service
run: systemctl restart app@redoal-portal.service
- name: Update redoal Caddy routing
run: |
cat > /etc/caddy/services.d/redoal-portal.caddy <<'EOF'
www.redoal.com {
redir https://redoal.com{uri} permanent
}
redoal.com {
reverse_proxy host.docker.internal:3020
log {
output file /var/log/caddy/redoal.log
}
}
EOF
docker exec caddy caddy reload --config /etc/caddy/Caddyfile --adapter caddyfile
+101
View File
@@ -0,0 +1,101 @@
name: Publish release
# Portal does not deploy itself. Cutting a version is deliberate:
# `cargo release <level>` bumps Cargo.toml, commits, tags v<semver>,
# and pushes; this workflow reacts to the tag and publishes the
# artifact as a Gitea release. Each content repo (uhhm/questions,
# redoal/questions) pins PORTAL_RELEASE to one of these tags in its
# own deploy workflow - bumping the pin there is what rolls a version
# out to a site. Plain main pushes only run test.yml.
on:
push:
tags: ["v*"]
jobs:
publish:
runs-on: bare
env:
# The bare runner's own systemd service intentionally has a minimal
# PATH/HOME (no rustup default toolchain in reach) - point it at the
# shared toolchain install directly rather than assuming an ambient
# dev shell environment.
CARGO_HOME: /var/local/cargo
RUSTUP_HOME: /var/local/rustup
PATH: /var/local/cargo/bin:/usr/local/sbin:/usr/local/bin:/usr/bin
# Shared with interactive dev builds (see ~/.config/fish/config.fish)
# so a crate compiled once, by either a manual build or CI, is
# cached for the other - real cache hits, not just a warm toolchain.
# SCCACHE_SERVER_PORT deliberately differs from the interactive
# dev shell's (4227): sccache's server is discovered by a fixed
# TCP port shared by every local user, so if both contexts used
# the same port, whichever one's server happened to be running
# would silently "win" and serve build requests it doesn't have
# filesystem permission for. Separate ports keep each context's
# own server answering its own requests; the cache directory
# (not the server process) is what's actually shared.
SCCACHE_DIR: /var/local/sccache
SCCACHE_SERVER_PORT: "4228"
CARGO_TARGET_DIR: /var/local/cargo-target
# cargo-leptos's own downloaded tools (wasm-bindgen, wasm-opt) cache
# under $XDG_CACHE_HOME - left at its default that resolves inside
# this unit's systemd StateDirectory, files it creates there end up
# owned by the kernel's overflow "nobody" uid instead of the
# runner's own dynamic uid (a StateDirectory quirk, not something
# in our control), so a later run can't execute what an earlier run
# downloaded. Redirecting it to our own known-good shared dir avoids
# that entirely.
XDG_CACHE_HOME: /var/local/leptos-cache
steps:
- uses: actions/checkout@v4
# SITE_NAME is read via option_env! (src/app.rs) - compile-time,
# not a runtime env var. It is only the last-resort fallback name:
# each instance's site.yaml (content-driven branding) overrides it,
# so one artifact serves every site.
- name: Build
run: SITE_NAME=${{ vars.PORTAL_SITE_NAME }} cargo leptos build --release
# cargo-leptos only builds the leptos bin-target ("portal") - the
# question-lint utility binary needs its own plain cargo build.
- name: Build question-lint
run: cargo build --release --bin question_lint --features ssr
- name: Package
run: |
set -euo pipefail
tag="${{ github.ref_name }}"
stage=$(mktemp -d)
cp "$CARGO_TARGET_DIR/release/portal" "$stage/portal"
cp "$CARGO_TARGET_DIR/release/question_lint" "$stage/question_lint"
# site-root ("target/site" in Cargo.toml) is project-relative,
# not affected by CARGO_TARGET_DIR - only the plain `cargo build`
# outputs (release/, front/) move with that override.
cp -r target/site "$stage/site"
tar -C "$stage" -czf "portal-$tag.tar.gz" portal question_lint site
rm -rf "$stage"
# The run's own ephemeral token has write access to this repo -
# no long-lived PAT to manage. The tag already exists (cargo
# release pushed it), so the release attaches to it; a re-run
# finds the existing release instead of failing.
- name: Publish release
run: |
set -euo pipefail
tag="${{ github.ref_name }}"
api="${{ github.server_url }}/api/v1/repos/${{ github.repository }}"
auth="Authorization: token ${{ secrets.GITHUB_TOKEN }}"
subject=$(git log -1 --format=%s)
body=$(printf '{"tag_name":"%s","name":"%s"}' \
"$tag" "$tag: $(echo "$subject" | sed 's/"/\\"/g')")
id=$(curl -sf -X POST -H "$auth" -H 'Content-Type: application/json' \
-d "$body" "$api/releases" | jq .id) \
|| id=$(curl -sf -H "$auth" "$api/releases/tags/$tag" | jq .id)
# Replace the asset if a re-run already uploaded one.
for aid in $(curl -sf -H "$auth" "$api/releases/$id/assets" | jq '.[].id'); do
curl -sf -X DELETE -H "$auth" "$api/releases/$id/assets/$aid"
done
curl -sf -X POST -H "$auth" \
-F "attachment=@portal-$tag.tar.gz" \
"$api/releases/$id/assets?name=portal-$tag.tar.gz" > /dev/null
echo "published $tag"
+26
View File
@@ -0,0 +1,26 @@
name: Test
# Publishing only happens on v* tags (publish.yml), so this is what
# keeps plain main pushes honest between releases.
on:
push:
branches: [main]
pull_request:
jobs:
test:
runs-on: bare
env:
# Same shared-toolchain/cache story as publish.yml.
CARGO_HOME: /var/local/cargo
RUSTUP_HOME: /var/local/rustup
PATH: /var/local/cargo/bin:/usr/local/sbin:/usr/local/bin:/usr/bin
SCCACHE_DIR: /var/local/sccache
SCCACHE_SERVER_PORT: "4228"
CARGO_TARGET_DIR: /var/local/cargo-target
steps:
- uses: actions/checkout@v4
- name: Test
run: cargo test --features ssr
+9
View File
@@ -131,3 +131,12 @@ bin-default-features = false
lib-features = ["hydrate"] lib-features = ["hydrate"]
lib-default-features = false lib-default-features = false
lib-profile-release = "wasm-release" lib-profile-release = "wasm-release"
# cargo release <level> is how a portal version is cut: bump, commit,
# tag v{{version}}, push. CI (publish.yml) reacts to the tag and
# publishes the release artifact; nothing on crates.io.
[package.metadata.release]
publish = false
push = true
tag-name = "v{{version}}"
pre-release-commit-message = "Release {{version}}"