home-infra/scripts/rollback-wishlist.sh

#!/usr/bin/env bash
# shellcheck shell=bash

set -Eeuo pipefail
umask 077

LIVE="/srv/nfs/media/wishlist"
LIVE_COMPOSE="${LIVE}/docker-compose.yml"
BACKUP_ROOT="/srv/nfs/media/.deploy-backups/wishlist"
DEPLOY_GROUP="infra-deploy"
CONTAINER="wishlist"
LOCK_FILE="/var/lib/gitea-runner/locks/wishlist.lock"
START_TIMEOUT=300
CHECK_INTERVAL=5

log() {
    echo
    echo "============================================================"
    echo "$*"
    echo "============================================================"
}

fail() {
    echo >&2
    echo "ERROR: $*" >&2
    exit 1
}

wait_for_wishlist() {
    local elapsed=0
    local running

    while (( elapsed < START_TIMEOUT )); do
        running="$(docker inspect -f '{{.State.Running}}' "${CONTAINER}" 2>/dev/null || echo false)"

        if [[ "${running}" != "true" ]]; then
            docker logs --tail 200 "${CONTAINER}" 2>/dev/null || true
            fail "Wishlist container is not running after rollback"
        fi

        if curl --fail --silent --show-error --max-time 5 \
            --output /dev/null http://127.0.0.1:3280/; then
            echo "Wishlist HTTP check: OK"
            return 0
        fi

        sleep "${CHECK_INTERVAL}"
        elapsed=$((elapsed + CHECK_INTERVAL))
        echo "Waiting after rollback... ${elapsed}s"
    done

    docker logs --tail 200 "${CONTAINER}" 2>/dev/null || true
    fail "Timeout waiting for Wishlist after rollback"
}

resolve_backup() {
    local requested="${1:-latest}"
    local latest

    if [[ "${requested}" != "latest" ]]; then
        [[ -d "${requested}" ]] && { printf '%s\n' "${requested}"; return 0; }
        [[ -d "${BACKUP_ROOT}/${requested}" ]] && { printf '%s\n' "${BACKUP_ROOT}/${requested}"; return 0; }
        return 1
    fi

    latest="$(
        find "${BACKUP_ROOT}" -mindepth 1 -maxdepth 1 -type d -printf '%f\n' \
            | grep -E '^[0-9]{8}-[0-9]{6}$' \
            | sort -r \
            | while IFS= read -r name; do
                if [[ -f "${BACKUP_ROOT}/${name}/.complete" ]]; then
                    echo "${BACKUP_ROOT}/${name}"
                    break
                fi
              done
    )"

    [[ -n "${latest}" ]] || return 1
    printf '%s\n' "${latest}"
}

for cmd in docker tar gzip sha256sum curl find grep sort flock chgrp chmod; do
    command -v "${cmd}" >/dev/null 2>&1 || fail "Required command not found: ${cmd}"
done

docker compose version >/dev/null 2>&1 || fail "Docker Compose plugin is not available"
getent group "${DEPLOY_GROUP}" >/dev/null 2>&1 || fail "Group ${DEPLOY_GROUP} does not exist"

mkdir -p "$(dirname "${LOCK_FILE}")"
if [[ "${WISHLIST_LOCK_HELD:-0}" != "1" ]]; then
    exec 9>"${LOCK_FILE}"
    flock -n 9 || fail "Another Wishlist deploy/rollback is already running"
fi

BACKUP_DIR="$(resolve_backup "${1:-latest}")" || fail "Could not find requested complete backup"

log "Rollback Wishlist from ${BACKUP_DIR}"

for file in .complete docker-compose.yml metadata.env state.tar.gz old-image.tar.gz SHA256SUMS; do
    [[ -e "${BACKUP_DIR}/${file}" ]] || fail "Backup file missing: ${BACKUP_DIR}/${file}"
done

(
    cd "${BACKUP_DIR}"
    sha256sum -c SHA256SUMS
)

tar -tzf "${BACKUP_DIR}/state.tar.gz" >/dev/null
gzip -t "${BACKUP_DIR}/old-image.tar.gz"

# shellcheck disable=SC1090
source "${BACKUP_DIR}/metadata.env"

: "${OLD_IMAGE_ID:?OLD_IMAGE_ID is missing from metadata}"
: "${OLD_IMAGE_REF:?OLD_IMAGE_REF is missing from metadata}"

FAILED_IMAGE_ID=""
if docker inspect "${CONTAINER}" >/dev/null 2>&1; then
    FAILED_IMAGE_ID="$(docker inspect -f '{{.Image}}' "${CONTAINER}" 2>/dev/null || true)"
    docker logs "${CONTAINER}" > "${BACKUP_DIR}/failed-container.log" 2>&1 || true
    docker rm -f "${CONTAINER}" >/dev/null || true
fi

log "Restore previous Compose"
cp "${BACKUP_DIR}/docker-compose.yml" "${LIVE_COMPOSE}"

log "Restore persistent state"
mkdir -p "${LIVE}/data" "${LIVE}/uploads"
find "${LIVE}/data" -mindepth 1 -maxdepth 1 -exec rm -rf -- {} +
find "${LIVE}/uploads" -mindepth 1 -maxdepth 1 -exec rm -rf -- {} +

tar --no-same-owner -xzf "${BACKUP_DIR}/state.tar.gz" -C "${LIVE}"

# Files extracted by the runner become runner-owned. Normalize the group and
# group permissions so the container can access them through the shared GID.
find "${LIVE}/data" "${LIVE}/uploads" -mindepth 1 -exec chgrp "${DEPLOY_GROUP}" {} +
find "${LIVE}/data" "${LIVE}/uploads" -type d -exec chmod g+rwx,g+s {} +
find "${LIVE}/data" "${LIVE}/uploads" -type f -exec chmod g+rw {} +

log "Restore exact previous Docker image"
gzip -dc "${BACKUP_DIR}/old-image.tar.gz" | docker image load

docker image inspect "${OLD_IMAGE_ID}" >/dev/null 2>&1 || \
    fail "Previous image ${OLD_IMAGE_ID} was not restored"

docker image tag "${OLD_IMAGE_ID}" "${OLD_IMAGE_REF}"

log "Start previous Wishlist version"
cd "${LIVE}"
docker compose -p wishlist -f "${LIVE_COMPOSE}" up -d --pull never --remove-orphans
wait_for_wishlist

RUNNING_IMAGE_ID="$(docker inspect -f '{{.Image}}' "${CONTAINER}")"
[[ "${RUNNING_IMAGE_ID}" == "${OLD_IMAGE_ID}" ]] || \
    fail "Rollback started unexpected image: ${RUNNING_IMAGE_ID}"

if [[ -n "${FAILED_IMAGE_ID}" && "${FAILED_IMAGE_ID}" != "${OLD_IMAGE_ID}" ]]; then
    docker image rm "${FAILED_IMAGE_ID}" >/dev/null 2>&1 || true
fi

touch "${BACKUP_DIR}/.rolled-back"

log "Wishlist rollback successful"
echo "Backup:        ${BACKUP_DIR}"
echo "Restored ID:   ${OLD_IMAGE_ID}"
echo "Restored ref:  ${OLD_IMAGE_REF}"