Files
RosterChief/deploy/deploy-dev.sh
Bernard Siebens 10b113f244 Read the health-check port from the server's WEB_PORT
The probe now reads WEB_PORT from the server's .env — the same file compose reads —
and builds the health URL from it, falling back to 8001 (compose's own default) when
it is unset. No more passing HEALTH_URL by hand when the published port changes.

Parsed the way compose parses it: last assignment wins, surrounding quotes and
inline whitespace stripped, commented lines ignored.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-15 13:25:39 +02:00

112 lines
5.1 KiB
Bash
Executable File

#!/usr/bin/env bash
# Deploy the test instance to the dev server, behind its existing Caddy.
#
# deploy/deploy-dev.sh # deploy the current branch
# BRANCH=main deploy/deploy-dev.sh
# deploy/deploy-dev.sh --push # push the branch first, then deploy
#
# Runs FROM your machine, works ON the server over one SSH session: it fetches the pushed
# branch, builds the image, runs migrations explicitly (never from the entrypoint — a
# starting gunicorn worker is a bad place to discover a failed migration), restarts web, and
# waits for /healthz. Any step failing aborts the whole thing with a non-zero exit.
set -Eeuo pipefail
# --- config (override via env) ----------------------------------------------
SSH_HOST="${SSH_HOST:-home.siebens.org}"
SSH_USER="${SSH_USER:-bernard}"
REMOTE_DIR="${REMOTE_DIR:-/home/bernard/RosterChief}"
BRANCH="${BRANCH:-$(git rev-parse --abbrev-ref HEAD)}"
COMPOSE_FILE="${COMPOSE_FILE:-compose.behind-proxy.yaml}"
# The published port lives in the server's .env (WEB_PORT), so it is read there, not here —
# see the remote block. This default only applies if that file omits it, matching compose's
# own `${WEB_PORT:-8001}`.
DEFAULT_WEB_PORT="${DEFAULT_WEB_PORT:-8001}"
SSH_TARGET="${SSH_USER}@${SSH_HOST}"
say() { printf '\033[1;36m==>\033[0m %s\n' "$*"; }
die() { printf '\033[1;31mERROR:\033[0m %s\n' "$*" >&2; exit 1; }
# --- preflight, locally -----------------------------------------------------
# The server deploys what is on the git remote, so unpushed commits would silently ship stale
# code. Catch that here rather than after a confusing "why isn't my change live" round trip.
git rev-parse --verify --quiet "origin/${BRANCH}" >/dev/null \
|| die "origin/${BRANCH} does not exist. Push the branch first, or pass --push."
if [ "${1:-}" = "--push" ]; then
say "Pushing ${BRANCH} to origin"
git push origin "${BRANCH}"
elif [ -n "$(git rev-list "origin/${BRANCH}..HEAD" 2>/dev/null)" ]; then
die "Local ${BRANCH} is ahead of origin — the server would deploy stale code. Push first, or run with --push."
fi
say "Deploying ${BRANCH} to ${SSH_TARGET}:${REMOTE_DIR}"
# --- the work, on the server ------------------------------------------------
# One SSH session runs the whole remote script; args are passed positionally so nothing has to
# be re-quoted inside the heredoc.
ssh -o ConnectTimeout=10 "${SSH_TARGET}" bash -s -- "${REMOTE_DIR}" "${BRANCH}" "${COMPOSE_FILE}" "${DEFAULT_WEB_PORT}" <<'REMOTE'
set -Eeuo pipefail
REMOTE_DIR="$1"; BRANCH="$2"; COMPOSE_FILE="$3"; DEFAULT_WEB_PORT="$4"
step() { printf '\033[1;34m ->\033[0m %s\n' "$*"; }
cd "$REMOTE_DIR" 2>/dev/null || { echo "ERROR: $REMOTE_DIR not found. Clone the repo there first."; exit 1; }
[ -d .git ] || { echo "ERROR: $REMOTE_DIR is not a git checkout."; exit 1; }
# The env files carry secrets and are never committed, so they must already be on the server.
# Fail loudly rather than boot a half-configured stack.
[ -f .env.production ] || { echo "ERROR: .env.production missing (Django config). Copy from .env.production.example."; exit 1; }
[ -f .env ] || { echo "ERROR: .env missing (compose vars: POSTGRES_PASSWORD, ...). Copy from .env.compose.example."; exit 1; }
dc() { docker compose -f "$COMPOSE_FILE" "$@"; }
# The health probe must hit the port the container actually publishes, which is WEB_PORT in
# the same .env compose reads. Parse it the way compose does — last assignment wins, quotes
# and inline whitespace stripped — and fall back to the compose default when it is unset.
WEB_PORT="$(sed -n 's/^[[:space:]]*WEB_PORT[[:space:]]*=[[:space:]]*//p' .env | tail -1 | tr -d '"'"'"' \r')"
WEB_PORT="${WEB_PORT:-$DEFAULT_WEB_PORT}"
HEALTH_URL="http://127.0.0.1:${WEB_PORT}/healthz"
# reset --hard, not pull: a deploy target only receives deploys, so make it exactly match the
# remote branch rather than risk a merge conflict from drift no one meant to leave there.
step "Fetching ${BRANCH}"
git fetch --quiet origin
git checkout --quiet "$BRANCH"
git reset --hard --quiet "origin/${BRANCH}"
echo " at $(git rev-parse --short HEAD)$(git log -1 --pretty=%s)"
step "Building image"
dc build
# Bring the data services up first and wait for Postgres, so the migration below has something
# to connect to on a cold start.
step "Starting db + redis"
dc up -d db redis
step "Running migrations"
# -T and </dev/null are load-bearing: this whole script IS ssh's stdin (a heredoc), and
# `compose run` without them attaches that stdin to the container — swallowing every command
# below it, so web never restarts and the script exits 0 having done half the job.
dc run --rm -T web python manage.py migrate --noinput </dev/null
# Recreate only web, with the freshly built image. db and redis keep running untouched.
step "Restarting web"
dc up -d --no-deps web
step "Waiting for /healthz on :${WEB_PORT}"
for attempt in $(seq 1 20); do
if curl -fsS "$HEALTH_URL" >/dev/null 2>&1; then
echo " healthy after ${attempt} check(s)"
exit 0
fi
sleep 3
done
echo "ERROR: health check never passed. Recent web logs:"
dc logs --tail 40 web
exit 1
REMOTE
say "Done. https://<your-test-domain>/healthz should return ok."