Merge pull request 'fix(ci): share one nginx-vhost renderer; drop unused deploy.sh' (#5) from fix/render-drift-remove-deploy-sh into main
All checks were successful
deploy / Build prerendered web (push) Successful in 8m43s
deploy / Deploy web to oolon (push) Successful in 17s
deploy / Build api + worker (static musl) (push) Successful in 6m12s
deploy / Deploy moments-worker to frootmig (push) Successful in 17s
deploy / Deploy moments-api to nikola (push) Successful in 29s
refresh / Rebuild prerendered web (push) Successful in 8m23s
refresh / Deploy refreshed web to oolon (push) Successful in 29s

This commit was merged in pull request #5.
This commit is contained in:
2026-07-26 12:54:48 +00:00
8 changed files with 105 additions and 602 deletions

View File

@@ -6,10 +6,11 @@ name: deploy
# - web (prerendered static SPA + nginx vhost) -> oolon
#
# This workflow is the source of infra truth (hosts, ports, paths live in `env`
# below — there is no manifest.yml in this model). It replaces script/deploy.sh:
# instead of an operator running it from a workstation with `pass`, a Gitea
# Actions runner deploys as the dedicated `gitea_ci` user over SSH, with secrets
# from the repo settings and scoped sudo (see asset/sudoers.d/ + script/infra-setup.sh).
# below — there is no manifest.yml in this model). A Gitea Actions runner deploys
# as the dedicated `gitea_ci` user over SSH, with secrets from the repo settings
# and scoped sudo (see asset/sudoers.d/ + script/infra-setup.sh). The nginx vhost
# is rendered by script/render-site-conf.py, shared with refresh.yml so the two
# never drift on what they substitute.
#
# The api/worker binaries are pure-rustls (no openssl), so they build as fully
# static musl — a runner newer than the target host can't produce an unloadable
@@ -338,22 +339,10 @@ jobs:
gitea_ci@"${WEB_HOST}" 'hostname -f'
- name: Render nginx vhost
run: |
mkdir -p rendered
python3 - <<'PY'
import os
t = open("asset/nginx/site.conf.tmpl").read()
subs = {
"SERVER_NAME": os.environ["SERVER_NAME"],
"DOCROOT": os.environ["WEB_ROOT"],
"WEB_LISTEN": os.environ["WEB_LISTEN"],
"API_UPSTREAM_SCHEME": os.environ["API_UPSTREAM_SCHEME"],
"API_UPSTREAM_ADDR": os.environ["API_UPSTREAM_ADDR"],
}
for k, v in subs.items():
t = t.replace("{{%s}}" % k, v)
open("rendered/site.conf", "w").write(t)
PY
# Shared with refresh.yml so the two pipelines can't drift on what they
# substitute; fails loudly if any {{PLACEHOLDER}} lacks an env value or
# survives substitution, so a broken vhost never reaches oolon.
run: python3 script/render-site-conf.py rendered/site.conf
- name: Sync static site (prerendered)
run: |

View File

@@ -23,6 +23,11 @@ env:
WEB_HOST: oolon.kosherinata.internal
SERVER_NAME: rob.tn
WEB_ROOT: /var/www/rob.tn
# TCP 443 on oolon is owned by the stream SNI router; this vhost sits behind it
# on 127.0.0.1:14443 with PROXY protocol. Must match deploy.yml's WEB_LISTEN —
# render-site-conf.py fails the build if it is missing rather than shipping a
# literal placeholder (see the render step below and deploy.yml for the history).
WEB_LISTEN: 127.0.0.1:14443 ssl proxy_protocol
API_PORT: "42424"
API_UPSTREAM_SCHEME: http
API_UPSTREAM_ADDR: nikola.kosherinata.internal:42424
@@ -67,16 +72,13 @@ jobs:
gitea_ci@"${WEB_HOST}" 'hostname -f'
- name: Render nginx vhost
run: |
mkdir -p rendered
python3 - <<'PY'
import os
t = open("asset/nginx/site.conf.tmpl").read()
for k in ("SERVER_NAME", "API_UPSTREAM_SCHEME", "API_UPSTREAM_ADDR"):
t = t.replace("{{%s}}" % k, os.environ[k])
t = t.replace("{{DOCROOT}}", os.environ["WEB_ROOT"])
open("rendered/site.conf", "w").write(t)
PY
# Shared with deploy.yml so the two pipelines can't drift on what they
# substitute; fails loudly if any {{PLACEHOLDER}} lacks an env value or
# survives substitution, so a broken vhost never reaches oolon. (This
# step is exactly where the drift bit: WEB_LISTEN was added to the
# template and deploy.yml but not here, and the nightly refresh shipped
# a literal `listen {{WEB_LISTEN}};` that froze every reload on oolon.)
run: python3 script/render-site-conf.py rendered/site.conf
- name: Sync static site (prerendered)
run: |

4
.gitignore vendored
View File

@@ -14,3 +14,7 @@
# rendered configs (templates committed, rendered output never)
/asset/config/*.toml
!/asset/config/*.toml.tmpl
# python bytecode (script/render-site-conf.py)
__pycache__/
*.pyc

View File

@@ -107,6 +107,11 @@ Actions, so the worker poller's tokens use the `QUERY_` prefix).
Nginx reverse-proxies `/api/` to the API host and serves the per-route static
files via `try_files $uri $uri/ /index.html`.
`./script/deploy.sh` is the legacy operator-driven path (workstation + `pass`);
it still works and the Gitea workflow supersedes it. Remove it once the workflow
is validated on the live hosts.
Both workflows render the nginx vhost through the shared `script/render-site-conf.py`
rather than an inline substitution per workflow. It requires every `{{PLACEHOLDER}}`
in `asset/nginx/site.conf.tmpl` to have a matching env var and refuses to emit a
file with any placeholder left unrendered — so a variable added to the template
but forgotten in one workflow's `env:` fails that build instead of shipping a
broken vhost to the edge. (The former per-workflow renderers drifted exactly this
way once: `WEB_LISTEN` reached the template and `deploy.yml` but not `refresh.yml`,
and the nightly refresh froze every reload on oolon.)

View File

@@ -15,7 +15,7 @@ server {
ssl_certificate_key /etc/letsencrypt/live/{{SERVER_NAME}}/privkey.pem;
ssl_protocols TLSv1.2 TLSv1.3;
root {{DOCROOT}};
root {{WEB_ROOT}};
index index.html;
# Compress text responses on the wire. text/html is always compressed when

View File

@@ -16,7 +16,7 @@ crates/
ui/ # vite + react + swc + typescript frontend
asset/ # systemd, nginx, firewalld, manifest.yml
script/
deploy.sh # manifest-driven deploy to prod
render-site-conf.py # render the nginx vhost from env (shared by both workflows)
hg-ingest.sh # one-shot local hg clone + psql ingest
certify.sh # letsencrypt cert management
teardown.sh # service removal
@@ -83,14 +83,14 @@ migrations live in `crates/moments-data/migrations/` and run automatically on wo
## deployment
```sh
./script/deploy.sh <env> all # api + worker + web
./script/deploy.sh <env> api worker # subset
./script/deploy.sh <env> default # api + web only (worker untouched)
./script/deploy.sh <env> all --dry-run
```
deployment is driven by Gitea Actions, not an operator workstation:
concrete hosts, ports, and the site's `server_name` live in `asset/manifest.yml`. the shape of the deployment:
- `.gitea/workflows/deploy.yml` — on push to `main` (or manual dispatch): lint/test gate, build the api + worker as static musl binaries and the prerendered web bundle, then deploy each component over SSH as the `gitea_ci` user with scoped sudo (`asset/sudoers.d/`).
- `.gitea/workflows/refresh.yml` — daily `schedule:` (or manual): rebuilds and redeploys only the web tier, re-baking the prerendered crawler snapshot without bouncing the api/worker.
both workflows carry the infra truth (hosts, ports, paths) in their `env:` blocks and render the nginx vhost through the shared `script/render-site-conf.py`, which fails the build if any template placeholder is unset rather than shipping it. one-time per-host provisioning (the `gitea_ci` user, its `authorized_keys`, the scoped sudoers drop-in) is `script/infra-setup.sh`.
the shape of the deployment:
| component | notes |
|-----------|-------|
@@ -101,7 +101,7 @@ concrete hosts, ports, and the site's `server_name` live in `asset/manifest.yml`
postgres roles `moments_rw` and `moments_ro` must exist on the primary, with `pg_ident.conf.d/<host>.conf` mapping the api host's fqdn to `moments_ro` and the worker host's fqdn to `moments_rw`. see `asset/sql/bootstrap-moments.sql`, `asset/postgres/ident.conf.tmpl`, and `script/db-perms.sh`.
secrets are resolved at deploy time via `pass`. the mapping of env-var name to pass-store path lives under `worker.secrets` in `manifest.yml`; `deploy.sh` iterates the map, fetches each secret, and substitutes the matching `{{NAME}}` placeholder in `worker.env.tmpl`.
the worker's poller tokens are Gitea repo Actions secrets (`QUERY_GITHUB_TOKEN`, `QUERY_GITEA_TOKEN` — the bare `GITHUB_TOKEN`/`GITEA_TOKEN` names are reserved by Actions). `deploy.yml`'s deploy-worker job substitutes them into the matching `{{NAME}}` placeholders in `worker.env.tmpl` at deploy time; secrets come from the runner environment and never touch a command line.
## environment variables

View File

@@ -1,559 +0,0 @@
#!/usr/bin/env bash
#
# moments deployment script.
#
# ./script/deploy.sh <environment> [component...]
# ./script/deploy.sh prod api worker web
# ./script/deploy.sh prod all
#
# Builds artifacts locally, resolves secrets from `pass`, renders config
# templates, rsyncs everything to the target hosts, and reloads systemd /
# nginx / firewalld / SELinux state idempotently.
set -euo pipefail
shopt -s nullglob
repo_root="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
manifest="${repo_root}/asset/manifest.yml"
dry_run=0
usage() {
cat <<EOF >&2
usage: $(basename "$0") <environment> [component...] [--dry-run]
$(basename "$0") prod api worker web
$(basename "$0") prod all
$(basename "$0") prod default # api + web (worker isn't restarted unless asked)
EOF
exit 2
}
log() { printf '\033[1;34m[deploy]\033[0m %s\n' "$*" >&2; }
warn() { printf '\033[1;33m[deploy]\033[0m %s\n' "$*" >&2; }
die() { printf '\033[1;31m[deploy]\033[0m %s\n' "$*" >&2; exit 1; }
run() {
if (( dry_run )); then
printf '\033[2m[dry-run]\033[0m %s\n' "$*" >&2
else
"$@"
fi
}
ssh_run() {
local host="$1"; shift
if (( dry_run )); then
printf '\033[2m[dry-run]\033[0m ssh %s -- %s\n' "$host" "$*" >&2
else
ssh -o BatchMode=yes "$host" "$@"
fi
}
# Ensure /tmp on the remote is world-writable + sticky (mode 1777). Some
# hosts in this fleet have had /tmp reset to root-owned 0755 by an
# unrelated configuration step, which silently breaks the rsync of the
# deploy stage dir under our unprivileged user. Check the mode first so a
# correctly-configured host doesn't incur a needless sudo call.
ensure_tmp_writable() {
local host="$1"
if (( dry_run )); then
printf '\033[2m[dry-run]\033[0m ssh %s -- stat /tmp; chmod 1777 if needed\n' "$host" >&2
return 0
fi
local mode
mode="$(ssh -o BatchMode=yes "$host" 'stat -c %a /tmp')" || {
warn "could not stat /tmp on $host"
return 1
}
if [[ "$mode" != "1777" ]]; then
warn "/tmp on $host is mode $mode; fixing to 1777"
ssh -o BatchMode=yes "$host" 'sudo chmod 1777 /tmp' || {
warn "failed to chmod /tmp on $host"
return 1
}
fi
}
[[ $# -ge 1 ]] || usage
environment="$1"; shift
components=()
while [[ $# -gt 0 ]]; do
case "$1" in
--dry-run) dry_run=1 ;;
*) components+=("$1") ;;
esac
shift
done
[[ -f "$manifest" ]] || die "manifest not found: $manifest"
command -v yq >/dev/null 2>&1 || die "yq is required"
command -v pass >/dev/null 2>&1 || die "pass is required"
command -v rsync >/dev/null 2>&1 || die "rsync is required"
command -v cargo >/dev/null 2>&1 || die "cargo is required"
command -v podman >/dev/null 2>&1 || die "podman is required (used for the deploy build container)"
# Rust binaries are built inside a Debian container so the resulting ELF
# links against an older glibc than this workstation's. Building natively
# on f44 (glibc 2.43) produces binaries that won't load on f42 / f43
# servers — the dynamic loader refuses them outright. Debian bookworm's
# glibc 2.36 is older than every Fedora release we deploy to, so its
# binaries are forward-compatible.
#
# The artifacts land in target/deploy/release/ so a native `cargo build`
# in this checkout (for tests, clippy, dev runs) doesn't compete with
# the container for incremental state, and vice-versa.
rust_build_image="docker.io/library/rust:1-bookworm"
rust_target_dir="${repo_root}/target/deploy"
# Resolve component list ----------------------------------------------------
env_path=".environments.${environment}"
yq --exit-status "${env_path}" "$manifest" >/dev/null \
|| die "environment '$environment' not found in manifest"
mapfile -t all_components < <(yq --raw-output "${env_path}.components | keys | .[]" "$manifest")
if [[ ${#components[@]} -eq 0 ]]; then
usage
fi
case "${components[0]:-}" in
all) components=("${all_components[@]}") ;;
default) components=(api web) ;;
esac
# Build artifacts -----------------------------------------------------------
needs_rust=0
needs_web=0
for c in "${components[@]}"; do
case "$c" in
api|worker) needs_rust=1 ;;
web) needs_web=1 ;;
esac
done
if (( needs_rust )); then
log "cargo build --release in ${rust_build_image} (api, worker)"
install --directory "$rust_target_dir"
# Named volumes cache the cargo registry and git index across runs so
# subsequent builds don't re-fetch every crate. CARGO_TARGET_DIR
# redirects build output into the host-mounted target/deploy.
# :Z relabels the bind mount for SELinux on Fedora hosts.
run podman run --rm \
--volume "${repo_root}:/workspace:Z" \
--volume moments-deploy-cargo-registry:/usr/local/cargo/registry \
--volume moments-deploy-cargo-git:/usr/local/cargo/git \
--workdir /workspace \
--env CARGO_TARGET_DIR=/workspace/target/deploy \
"$rust_build_image" \
cargo build --release --bin moments-api --bin moments-worker
fi
if (( needs_web )); then
log "vite build (ui)"
run sh -c "cd '${repo_root}/ui' && pnpm install --frozen-lockfile && pnpm run build"
fi
# Per-component deploy ------------------------------------------------------
deploy_api() {
local host="$1"
log "api -> $host"
local bind
bind="$(yq --raw-output "${env_path}.components.api.config.bind" "$manifest")"
[[ -n "$bind" && "$bind" != "null" ]] || die "api.config.bind missing in manifest"
[[ "$bind" == *:* ]] \
|| die "api.config.bind must be host:port form: '$bind'"
local api_port
api_port="${bind##*:}"
[[ "$api_port" =~ ^[0-9]+$ ]] \
|| die "api.config.bind port is not numeric: '$api_port'"
if (( dry_run )); then
printf '\033[2m[dry-run]\033[0m render api.env (HOSTNAME=%s, BIND=%s) + firewalld svc (port=%s) + units, stage to %s:/tmp/, install via heredoc, run sysusers/restorecon/semanage/systemctl on %s\n' \
"$host" "$bind" "$api_port" "$host" "$host" >&2
return 0
fi
local fqdn="$host"
local stage
stage="$(mktemp --directory)"
trap "rm --recursive --force '$stage'" RETURN
install --directory \
"$stage/etc/moments" \
"$stage/etc/systemd/system" \
"$stage/etc/sysusers.d" \
"$stage/etc/firewalld/services" \
"$stage/usr/local/bin"
local rendered
rendered="$(<"${repo_root}/asset/config/api.env.tmpl")"
rendered=${rendered//'{{HOSTNAME}}'/$fqdn}
rendered=${rendered//'{{BIND}}'/$bind}
printf '%s\n' "$rendered" > "$stage/etc/moments/api.env"
rendered="$(<"${repo_root}/asset/systemd/moments-api-cert.path")"
rendered=${rendered//'{{HOSTNAME}}'/$fqdn}
printf '%s\n' "$rendered" > "$stage/etc/systemd/system/moments-api-cert.path"
rendered="$(<"${repo_root}/asset/firewalld/moments-api.xml.tmpl")"
rendered=${rendered//'{{API_PORT}}'/$api_port}
printf '%s\n' "$rendered" > "$stage/etc/firewalld/services/moments-api.xml"
chmod 0644 "$stage/etc/firewalld/services/moments-api.xml"
install --mode=0644 "${repo_root}/asset/systemd/moments-api.service" "$stage/etc/systemd/system/"
install --mode=0644 "${repo_root}/asset/systemd/moments-api-cert-reload.service" "$stage/etc/systemd/system/"
install --mode=0644 "${repo_root}/asset/systemd/moments.sysusers.conf" "$stage/etc/sysusers.d/moments.conf"
install --mode=0755 "${rust_target_dir}/release/moments-api" "$stage/usr/local/bin/moments-api"
chmod 0640 "$stage/etc/moments/api.env"
# Stage to a tmpdir on the remote, then `install` each file at its final
# path via the heredoc. Never rsync into /, since rsync of staged parent
# dirs (etc/, usr/, ...) can leak ownership, ACLs and xattrs onto the
# live system dirs.
local remote_stage="/tmp/moments-deploy.api.${$}.${RANDOM}"
ensure_tmp_writable "$host" || return 1
rsync \
--archive \
--hard-links \
--numeric-ids \
--rsh='ssh -o BatchMode=yes' \
"$stage/" \
"${host}:${remote_stage}/"
ssh_run "$host" "sudo bash -s -- ${remote_stage@Q} ${api_port@Q}" <<'REMOTE_EOF'
set -euo pipefail
remote_stage="$1"
api_port="$2"
trap 'rm --recursive --force "$remote_stage"' EXIT
fqdn="$(hostname --fqdn)"
install --owner=root --group=root --mode=0644 \
"$remote_stage/etc/sysusers.d/moments.conf" \
/etc/sysusers.d/moments.conf
systemd-sysusers /etc/sysusers.d/moments.conf
install --directory --owner=root --group=moments --mode=0750 /etc/moments
install --directory --owner=moments --group=moments --mode=0750 /var/lib/moments
install --owner=root --group=moments --mode=0640 \
"$remote_stage/etc/moments/api.env" \
/etc/moments/api.env
install --owner=root --group=root --mode=0644 \
"$remote_stage/etc/systemd/system/moments-api.service" \
/etc/systemd/system/moments-api.service
install --owner=root --group=root --mode=0644 \
"$remote_stage/etc/systemd/system/moments-api-cert.path" \
/etc/systemd/system/moments-api-cert.path
install --owner=root --group=root --mode=0644 \
"$remote_stage/etc/systemd/system/moments-api-cert-reload.service" \
/etc/systemd/system/moments-api-cert-reload.service
install --owner=root --group=root --mode=0644 \
"$remote_stage/etc/firewalld/services/moments-api.xml" \
/etc/firewalld/services/moments-api.xml
install --owner=root --group=root --mode=0755 \
"$remote_stage/usr/local/bin/moments-api" \
/usr/local/bin/moments-api
# Grant the moments user read access to the host private key for the
# postgres mTLS connection.
setfacl --modify=u:moments:r "/etc/pki/tls/private/${fqdn}.pem" || true
# Idempotent label: --add fails if the port is already labelled (we suppress
# that one stderr line); --modify is then a no-op or fixes a stale type.
semanage port --add --type=http_port_t --proto=tcp "$api_port" 2>/dev/null \
|| semanage port --modify --type=http_port_t --proto=tcp "$api_port"
firewall-cmd --reload
zone="$(firewall-cmd --get-default-zone)"
if ! firewall-cmd --zone="$zone" --query-service=moments-api >/dev/null 2>&1; then
firewall-cmd --permanent --zone="$zone" --add-service=moments-api
firewall-cmd --zone="$zone" --add-service=moments-api
fi
restorecon -Rv /usr/local/bin/moments-api /etc/moments /var/lib/moments
systemctl daemon-reload
systemctl enable --now moments-api-cert.path
systemctl enable --now moments-api.service
systemctl restart moments-api.service
# Quietly retry while the service binds; only show curl's diagnostics if
# every attempt fails. The journalctl tail on failure is the verbose source.
for i in 1 2 3 4 5 6 7 8 9 10; do
if curl --fail --silent "http://${fqdn}:${api_port}/v1/healthz" >/dev/null 2>&1; then
echo "moments-api healthy"
exit 0
fi
sleep 1
done
echo "moments-api did not become healthy" >&2
curl --fail --silent --show-error "http://${fqdn}:${api_port}/v1/healthz" >/dev/null || true
journalctl --unit=moments-api.service --lines=50 --no-pager >&2
exit 1
REMOTE_EOF
}
deploy_worker() {
local host="$1"
log "worker -> $host"
# Manifest entries under `worker.secrets` map env-var name -> pass store path.
# The script fetches each via `pass` and substitutes the matching {{NAME}}
# placeholder in worker.env.tmpl. Adding a new secret is then a manifest +
# template change; no script edit required.
local -a secret_lines secret_keys
mapfile -t secret_lines < <(yq --raw-output \
"${env_path}.components.worker.secrets // {} | to_entries | .[] | \"\(.key)=\(.value)\"" \
"$manifest")
local line
for line in "${secret_lines[@]}"; do
[[ -n "$line" ]] && secret_keys+=("${line%%=*}")
done
if (( dry_run )); then
printf '\033[2m[dry-run]\033[0m render worker.env (HOSTNAME=%s, secrets [%s] from pass) + units, stage to %s:/tmp/, install via heredoc, run sysusers/restorecon/systemctl on %s\n' \
"$host" "${secret_keys[*]:-none}" "$host" "$host" >&2
return 0
fi
local fqdn="$host"
local stage
stage="$(mktemp --directory)"
trap "rm --recursive --force '$stage'" RETURN
install --directory \
"$stage/etc/moments" \
"$stage/etc/systemd/system" \
"$stage/etc/sysusers.d" \
"$stage/usr/local/bin"
# Render templates in-memory so secrets never appear on a command line
# (sed would expose them to anything that can read /proc/<pid>/cmdline).
local rendered
rendered="$(<"${repo_root}/asset/config/worker.env.tmpl")"
rendered=${rendered//'{{HOSTNAME}}'/$fqdn}
local key pass_path value
for line in "${secret_lines[@]}"; do
[[ -z "$line" ]] && continue
key="${line%%=*}"
pass_path="${line#*=}"
if pass show "$pass_path" >/dev/null 2>&1; then
value="$(pass show "$pass_path")"
else
warn "no secret in pass at '${pass_path}' for ${key}; worker will run without ${key}"
value=""
fi
rendered=${rendered//"{{${key}}}"/$value}
done
printf '%s\n' "$rendered" > "$stage/etc/moments/worker.env"
rendered="$(<"${repo_root}/asset/systemd/moments-worker-cert.path")"
rendered=${rendered//'{{HOSTNAME}}'/$fqdn}
printf '%s\n' "$rendered" > "$stage/etc/systemd/system/moments-worker-cert.path"
install --mode=0644 "${repo_root}/asset/systemd/moments-worker.service" "$stage/etc/systemd/system/"
install --mode=0644 "${repo_root}/asset/systemd/moments-worker-cert-reload.service" "$stage/etc/systemd/system/"
install --mode=0644 "${repo_root}/asset/systemd/moments.sysusers.conf" "$stage/etc/sysusers.d/moments.conf"
install --mode=0755 "${rust_target_dir}/release/moments-worker" "$stage/usr/local/bin/moments-worker"
chmod 0640 "$stage/etc/moments/worker.env"
# Stage to a tmpdir on the remote, then `install` each file at its final
# path via the heredoc. Never rsync into /.
local remote_stage="/tmp/moments-deploy.worker.${$}.${RANDOM}"
ensure_tmp_writable "$host" || return 1
rsync \
--archive \
--hard-links \
--numeric-ids \
--rsh='ssh -o BatchMode=yes' \
"$stage/" \
"${host}:${remote_stage}/"
ssh_run "$host" "sudo bash -s -- ${remote_stage@Q}" <<'REMOTE_EOF'
set -euo pipefail
remote_stage="$1"
trap 'rm --recursive --force "$remote_stage"' EXIT
fqdn="$(hostname --fqdn)"
install --owner=root --group=root --mode=0644 \
"$remote_stage/etc/sysusers.d/moments.conf" \
/etc/sysusers.d/moments.conf
systemd-sysusers /etc/sysusers.d/moments.conf
install --directory --owner=root --group=moments --mode=0750 /etc/moments
install --directory --owner=moments --group=moments --mode=0750 /var/lib/moments
install --owner=root --group=moments --mode=0640 \
"$remote_stage/etc/moments/worker.env" \
/etc/moments/worker.env
install --owner=root --group=root --mode=0644 \
"$remote_stage/etc/systemd/system/moments-worker.service" \
/etc/systemd/system/moments-worker.service
install --owner=root --group=root --mode=0644 \
"$remote_stage/etc/systemd/system/moments-worker-cert.path" \
/etc/systemd/system/moments-worker-cert.path
install --owner=root --group=root --mode=0644 \
"$remote_stage/etc/systemd/system/moments-worker-cert-reload.service" \
/etc/systemd/system/moments-worker-cert-reload.service
install --owner=root --group=root --mode=0755 \
"$remote_stage/usr/local/bin/moments-worker" \
/usr/local/bin/moments-worker
setfacl --modify=u:moments:r "/etc/pki/tls/private/${fqdn}.pem" || true
restorecon -Rv /usr/local/bin/moments-worker /etc/moments /var/lib/moments
systemctl daemon-reload
systemctl enable --now moments-worker-cert.path
systemctl enable --now moments-worker.service
systemctl restart moments-worker.service
if ! systemctl is-active --quiet moments-worker.service; then
journalctl --unit=moments-worker.service --lines=50 --no-pager >&2
exit 1
fi
echo "moments-worker active"
REMOTE_EOF
}
deploy_web() {
local host="$1"
log "web -> $host"
local server_name web_root api_upstream web_listen
server_name="$(yq --raw-output "${env_path}.components.web.config.server_name" "$manifest")"
web_root="$(yq --raw-output "${env_path}.components.web.config.root" "$manifest")"
api_upstream="$(yq --raw-output "${env_path}.components.web.config.api_upstream" "$manifest")"
# The edge's stream SNI router owns TCP 443; vhosts sit behind it on 14443.
# Binding 443 here collides with the router — `nginx -t` passes, but every
# later reload aborts while systemctl still reports success. See the
# web.config.listen note in asset/manifest.yml.
web_listen="$(yq --raw-output "${env_path}.components.web.config.listen" "$manifest")"
[[ -n "$web_listen" && "$web_listen" != "null" ]] \
|| web_listen='127.0.0.1:14443 ssl proxy_protocol'
[[ -n "$server_name" && "$server_name" != "null" ]] || die "web.config.server_name missing in manifest"
[[ -n "$web_root" && "$web_root" != "null" ]] || die "web.config.root missing in manifest"
[[ -n "$api_upstream" && "$api_upstream" != "null" ]] || die "web.config.api_upstream missing in manifest"
[[ "$web_root" == /* ]] \
|| die "web.config.root must be an absolute path: '$web_root'"
[[ "$api_upstream" == http://* || "$api_upstream" == https://* ]] \
|| die "web.config.api_upstream must be a http(s) URL: '$api_upstream'"
local api_upstream_scheme api_upstream_addr api_upstream_port
api_upstream_scheme="${api_upstream%%://*}"
api_upstream_addr="${api_upstream#*://}"
[[ "$api_upstream_addr" == *:* ]] \
|| die "web.config.api_upstream must include an explicit port: '$api_upstream'"
api_upstream_port="${api_upstream_addr##*:}"
[[ "$api_upstream_port" =~ ^[0-9]+$ ]] \
|| die "extracted upstream port is not numeric: '$api_upstream_port'"
local site_conf_path="/etc/nginx/conf.d/${server_name}.conf"
if (( dry_run )); then
printf '\033[2m[dry-run]\033[0m render %s (server_name=%s, docroot=%s, upstream=%s://%s) + rsync ui/dist/ to %s:%s/, run nginx -t/reload on %s\n' \
"$site_conf_path" "$server_name" "$web_root" \
"$api_upstream_scheme" "$api_upstream_addr" \
"$host" "$web_root" "$host" >&2
return 0
fi
local stage
stage="$(mktemp --directory)"
trap "rm --recursive --force '$stage'" RETURN
install --directory "${stage}${web_root}" "$stage/etc/nginx/conf.d"
rsync --archive "${repo_root}/ui/dist/" "${stage}${web_root}/"
local rendered
rendered="$(<"${repo_root}/asset/nginx/site.conf.tmpl")"
rendered=${rendered//'{{SERVER_NAME}}'/$server_name}
rendered=${rendered//'{{WEB_LISTEN}}'/$web_listen}
rendered=${rendered//'{{DOCROOT}}'/$web_root}
rendered=${rendered//'{{API_UPSTREAM_SCHEME}}'/$api_upstream_scheme}
rendered=${rendered//'{{API_UPSTREAM_ADDR}}'/$api_upstream_addr}
printf '%s\n' "$rendered" > "${stage}${site_conf_path}"
chmod 0644 "${stage}${site_conf_path}"
# Both targets are leaf paths (the docroot itself, and a single named
# file) so rsync does not traverse /var or /etc parents — `--chown` is
# enough; -A/-X are intentionally absent.
rsync \
--archive \
--hard-links \
--numeric-ids \
--chown root:root \
--rsh='ssh -o BatchMode=yes' \
--rsync-path 'sudo rsync' \
--delete \
"${stage}${web_root}/" \
"${host}:${web_root}/"
rsync \
--archive \
--hard-links \
--numeric-ids \
--chown root:root \
--rsh='ssh -o BatchMode=yes' \
--rsync-path 'sudo rsync' \
"${stage}${site_conf_path}" \
"${host}:${site_conf_path}"
ssh_run "$host" "sudo bash -s -- ${web_root@Q} ${site_conf_path@Q} ${api_upstream_port@Q}" <<'REMOTE_EOF'
set -euo pipefail
web_root="$1"
site_conf_path="$2"
api_upstream_port="$3"
# Allow nginx to make outbound connections to the moments-api upstream
# across the WG mesh.
setsebool -P httpd_can_network_connect on
# Idempotent label: --add fails if the port is already labelled (we suppress
# that one stderr line); --modify is then a no-op or fixes a stale type.
semanage port --add --type=http_port_t --proto=tcp "$api_upstream_port" 2>/dev/null \
|| semanage port --modify --type=http_port_t --proto=tcp "$api_upstream_port"
restorecon -Rv "$web_root" "$site_conf_path"
if ! nginx -t; then
echo "nginx config check failed" >&2
exit 1
fi
systemctl reload nginx
echo "nginx reloaded"
REMOTE_EOF
}
# Dispatch ------------------------------------------------------------------
failed=()
for component in "${components[@]}"; do
mapfile -t hosts < <(yq --raw-output "${env_path}.components.${component}.hosts[]" "$manifest")
for host in "${hosts[@]}"; do
case "$component" in
api) deploy_api "$host" || failed+=("api@$host") ;;
worker) deploy_worker "$host" || failed+=("worker@$host") ;;
web) deploy_web "$host" || failed+=("web@$host") ;;
*) warn "unknown component: $component" ;;
esac
done
done
if [[ ${#failed[@]} -gt 0 ]]; then
die "failed: ${failed[*]}"
fi
log "deploy complete"

View File

@@ -0,0 +1,62 @@
#!/usr/bin/env python3
"""Render the nginx vhost (asset/nginx/site.conf.tmpl) from the environment.
Both .gitea/workflows/deploy.yml (deploy-web) and refresh.yml call this, so the
two pipelines can never disagree about what they substitute. They did once: the
WEB_LISTEN placeholder was added to the template and to deploy.yml but not to
refresh.yml, so the nightly refresh shipped a literal `listen {{WEB_LISTEN}};`
to oolon. `nginx -t` then failed for the whole edge, and because the file is
rsynced straight into the live conf.d before it is tested, every vhost's reload
-- including the step@ cert renewals -- stayed frozen for days while renewed
certs piled up unserved on disk.
Guard rails, so that can't recur:
* every {{PLACEHOLDER}} in the template must have a matching environment
variable, or the render fails before anything leaves the runner;
* no {{...}} may survive substitution.
A forgotten or misnamed variable is now a red build, not a broken edge.
usage: render-site-conf.py [OUTPUT] (default: rendered/site.conf)
"""
import os
import re
import sys
TEMPLATE = "asset/nginx/site.conf.tmpl"
PLACEHOLDER = re.compile(r"\{\{(\w+)\}\}")
def main() -> int:
out = sys.argv[1] if len(sys.argv) > 1 else "rendered/site.conf"
with open(TEMPLATE, encoding="utf-8") as fh:
text = fh.read()
names = sorted(set(PLACEHOLDER.findall(text)))
missing = [n for n in names if n not in os.environ]
if missing:
sys.stderr.write(
"render-site-conf: no environment value for placeholder(s): "
+ ", ".join(missing) + "\n")
return 1
for name in names:
text = text.replace("{{%s}}" % name, os.environ[name])
leftover = sorted(set(PLACEHOLDER.findall(text)))
if leftover:
sys.stderr.write(
"render-site-conf: unrendered placeholder(s) after substitution: "
+ ", ".join(leftover) + "\n")
return 1
os.makedirs(os.path.dirname(out) or ".", exist_ok=True)
with open(out, "w", encoding="utf-8") as fh:
fh.write(text)
sys.stderr.write(
"render-site-conf: wrote %s (%d substitutions: %s)\n"
% (out, len(names), ", ".join(names)))
return 0
if __name__ == "__main__":
raise SystemExit(main())