gatehouse-api/deploy/deploy.sh

#!/usr/bin/env bash
# Rolling deploy: drains each api container from nginx before replacing it,
# so users see zero downtime. Job workers are restarted last without draining.
#
# Usage:
#   ./deploy.sh <image-tag>    e.g. ./deploy.sh a1b2c3d
#
# Overridable via env vars (defaults below are gatehouse-api):
#   IMAGE_NAME   docker image name
#   SVC1/SVC2    compose service names
#   SVC1_PORT    host port for SVC1
#   SVC2_PORT    host port for SVC2
#   HEALTH_PATH  HTTP path for health check
#   NGINX_CONF   path to nginx site config
#   COMPOSE_DIR  directory with docker-compose.yml and .env
#
# The runner user needs passwordless sudo for nginx:
#   echo "runner ALL=(ALL) NOPASSWD: /usr/sbin/nginx, /bin/systemctl reload nginx" \
#     | sudo tee /etc/sudoers.d/runner-nginx
set -euo pipefail

TAG="${1:?Usage: deploy.sh <image-tag>  (e.g. deploy.sh a1b2c3d)}"

IMAGE_NAME="${IMAGE_NAME:-gatehouse-api}"
NGINX_CONF="${NGINX_CONF:-/etc/nginx/conf.d/gatehouse-api.conf}"
COMPOSE_DIR="${COMPOSE_DIR:-/opt/gatehouse-api}"
SVC1="${SVC1:-api1}"
SVC2="${SVC2:-api2}"
SVC1_PORT="${SVC1_PORT:-5000}"
SVC2_PORT="${SVC2_PORT:-5001}"
HEALTH_PATH="${HEALTH_PATH:-/api/health}"
HEALTH_RETRIES=18     # 18 × 5 s = 90 s max
HEALTH_INTERVAL=5

# ── helpers ───────────────────────────────────────────────────────────────────

log()  { echo "[$(date '+%H:%M:%S')] $*"; }
die()  { log "ERROR: $*" >&2; exit 1; }
step() { log ""; log "── $* ──"; }

health_check() {
    local port=$1 label=$2 attempt=0
    log "Waiting for ${label} on :${port}${HEALTH_PATH} ..."
    while (( attempt < HEALTH_RETRIES )); do
        if curl -sf "http://127.0.0.1:${port}${HEALTH_PATH}" -o /dev/null; then
            log "✓ ${label} healthy"
            return 0
        fi
        attempt=$(( attempt + 1 ))
        log "  [${attempt}/${HEALTH_RETRIES}] not ready, retrying in ${HEALTH_INTERVAL}s"
        sleep "${HEALTH_INTERVAL}"
    done
    log "ERROR: ${label} failed health check after $((HEALTH_RETRIES * HEALTH_INTERVAL))s"
    return 1
}

get_service_tag() {
    docker compose ps -q "$1" 2>/dev/null \
        | xargs -r docker inspect --format '{{.Config.Image}}' 2>/dev/null \
        | cut -d: -f2
}

rollback() {
    local service=$1 port=$2 old_tag=$3
    if [[ -z "${old_tag}" ]]; then
        nginx_restore "${port}"
        die "Deploy aborted — ${service} failed health check, no previous tag to roll back to"
    fi
    log "Rolling back ${service} to ${old_tag}..."
    IMAGE_TAG="${old_tag}" docker compose up -d --no-deps --force-recreate "${service}"
    nginx_restore "${port}"
    die "Deploy aborted — ${service} rolled back to ${old_tag}"
}

nginx_drain() {
    local port=$1
    sudo sed -i "s|server 127.0.0.1:${port};|server 127.0.0.1:${port} down;|" "$NGINX_CONF"
    sudo nginx -t 2>&1 | tail -2
    sudo nginx -s reload
    log "nginx: drained :${port}"
}

nginx_restore() {
    local port=$1
    sudo sed -i "s|server 127.0.0.1:${port} down;|server 127.0.0.1:${port};|" "$NGINX_CONF"
    sudo nginx -t 2>&1 | tail -2
    sudo nginx -s reload
    log "nginx: restored :${port}"
}

# ── pre-flight ────────────────────────────────────────────────────────────────

cd "${COMPOSE_DIR}"
pwd; ls -la
log "Deploying ${IMAGE_NAME}:${TAG}"

docker image inspect "${IMAGE_NAME}:${TAG}" > /dev/null 2>&1 \
    || die "Image ${IMAGE_NAME}:${TAG} not found locally — build it first."

# ── roll SVC1 ─────────────────────────────────────────────────────────────────

step "${SVC1} → ${TAG}  (traffic: ${SVC2} only)"
old_svc1=$(get_service_tag "${SVC1}")
nginx_drain "${SVC1_PORT}"
log "Waiting 15s for in-flight requests to drain..."
sleep 15

IMAGE_TAG="${TAG}" docker compose up -d --no-deps --force-recreate "${SVC1}"

health_check "${SVC1_PORT}" "${SVC1}" || rollback "${SVC1}" "${SVC1_PORT}" "${old_svc1}"
nginx_restore "${SVC1_PORT}"

# ── roll SVC2 ─────────────────────────────────────────────────────────────────

step "${SVC2} → ${TAG}  (traffic: ${SVC1} only)"
old_svc2=$(get_service_tag "${SVC2}")
nginx_drain "${SVC2_PORT}"
log "Waiting 15s for in-flight requests to drain..."
sleep 15

IMAGE_TAG="${TAG}" docker compose up -d --no-deps --force-recreate "${SVC2}"

health_check "${SVC2_PORT}" "${SVC2}" || rollback "${SVC2}" "${SVC2_PORT}" "${old_svc2}"
nginx_restore "${SVC2_PORT}"

# ── job workers ───────────────────────────────────────────────────────────────

step "job workers → ${TAG}"
IMAGE_TAG="${TAG}" docker compose up -d --no-deps --force-recreate zerotier-reconciler mfa-compliance

# ── done ──────────────────────────────────────────────────────────────────────

log ""
log "Deploy complete ✓  ${IMAGE_NAME}:${TAG}"