#!/usr/bin/env bash
# Roll out a published CodeCraft image to a remote installation that was set up with
# scripts/move-server.sh (docker-compose.yml + .env in one folder on the target).
#
#   scripts/release.sh        # build and publish a new version (amd64 + arm64)
#   scripts/deploy.sh         # roll the newest tag out to the target
#
# Steps:
#   1. check that the image exists in the registry for the target's architecture,
#   2. wait until no agent or terminal runs on the target (unless --force),
#   3. set CODECRAFT_VERSION in the target's .env, docker compose pull && up -d
#      (or start.sh on hosts without Compose),
#   4. run scripts/smoke-test.sh on the target; on failure the previous version is
#      started again.
#
# Usage: scripts/deploy.sh [vYYYYMM-NNNNN] [--target user@host:/dir] [--force]
#                          [--wait MINUTES] [--dry-run]
#   vYYYYMM-NNNNN   version to deploy (default: the newest vYYYYMM-NNNNN git tag)
#   --target        installation folder on the target (default: DEPLOY_TARGET from .env)
#   --force         restart even if agents or terminals are running (they end)
#   --wait N        how long to wait for an idle server before giving up (default 15)
#   --dry-run       show the steps only
#
# Environment:
#   REGISTRY        image name without tag (default git.familie-zink.org/zink/codecraft)
#   CONTAINER       container name on the target (default codecraft)
#   PORT            port of CodeCraft on the target, for the smoke test (default: PORT
#                   from the target's .env, else 8787)
set -euo pipefail

REGISTRY="${REGISTRY:-git.familie-zink.org/zink/codecraft}"
CONTAINER="${CONTAINER:-codecraft}"

tag="" target="" force=0 wait_min=15 dry=0
while (($#)); do
  case "$1" in
    --target) target="$2"; shift ;;
    --force) force=1 ;;
    --wait) wait_min="$2"; shift ;;
    --dry-run) dry=1 ;;
    -h|--help) sed -n '2,29p' "$0"; exit 0 ;;
    -*) echo "unknown option $1" >&2; exit 2 ;;
    *) tag="$1" ;;
  esac
  shift
done

cd "$(git rev-parse --show-toplevel)"
step() { printf '\n==> %s\n' "$*"; }
fail() { echo "error: $*" >&2; exit 1; }
q() { printf '%q' "$1"; }
env_get() { [[ -f .env ]] && sed -n "s/^$1=//p" .env | tail -1 | sed -E "s/^'(.*)'\$/\1/; s/^\"(.*)\"\$/\1/"; }

target="${target:-$(env_get DEPLOY_TARGET || true)}"
[[ "$target" == *:/* ]] || fail "no target – pass --target user@host:/dir or set DEPLOY_TARGET in .env"
host="${target%%:*}"
dir="${target#*:}"; dir="${dir%/}"
remote() { ssh -o BatchMode=yes "$host" "$1"; }
run_remote() { if (( dry )); then echo "[dry-run] on $host: $1"; else remote "$1"; fi; }

if [[ -z "$tag" ]]; then
  git fetch --tags --quiet origin || true
  tag=$(git tag -l 'v[0-9][0-9][0-9][0-9][0-9][0-9]-[0-9][0-9][0-9][0-9][0-9]' | sort | tail -1)
  [[ -n "$tag" ]] || fail "no release tag found"
fi
[[ "$tag" =~ ^v[0-9]{6}-[0-9]{5}$ ]] || fail "version must look like vYYYYMM-NNNNN, got $tag"

step "Checking $host:$dir and the image $REGISTRY:$tag"
remote "test -f $(q "$dir/.env") && test -f $(q "$dir/docker-compose.yml")" \
  || fail "$dir on $host has no .env/docker-compose.yml – set it up with scripts/move-server.sh"
current=$(remote "sed -n 's/^CODECRAFT_VERSION=//p' $(q "$dir/.env") | tail -1")
case "$(remote 'uname -m')" in
  x86_64) arch=amd64 ;; aarch64|arm64) arch=arm64 ;; *) arch="" ;;
esac
if [[ -n "$arch" ]]; then
  # Captured first: `inspect | grep -q` fails under pipefail when grep exits early (SIGPIPE).
  manifests=$(docker buildx imagetools inspect "$REGISTRY:$tag" 2>/dev/null || true)
  grep -q "Platform:.*linux/$arch" <<<"$manifests" \
    || fail "$REGISTRY:$tag has no linux/$arch image – run scripts/release.sh first"
fi
echo "deploying $tag (running: ${current:-unknown}, target linux/${arch:-?})"
[[ "$current" == "$tag" ]] && echo "note: $tag is already configured – the container is recreated anyway"

# Busy = the container runs more than the server process (agents' shells, terminals, ...).
busy() { (( $(remote "docker top $(q "$CONTAINER") -o pid 2>/dev/null | tail -n +2 | wc -l") > 1 )); }
if (( ! force )) && remote "docker inspect -f '{{.State.Running}}' $(q "$CONTAINER") 2>/dev/null" | grep -q true; then
  step "Waiting until no agent or terminal runs (up to $wait_min min, --force skips this)"
  deadline=$(( $(date +%s) + wait_min * 60 ))
  while busy; do
    (( $(date +%s) < deadline )) || fail "still busy after $wait_min min – try later or use --force"
    sleep 15
  done
  echo "idle"
fi

step "Updating to $tag"
run_remote "cd $(q "$dir") && sed -i 's/^CODECRAFT_VERSION=.*/CODECRAFT_VERSION=$tag/' .env \
  && (grep -q '^CODECRAFT_VERSION=' .env || echo 'CODECRAFT_VERSION=$tag' >> .env) \
  && docker pull -q $(q "$REGISTRY:$tag") >/dev/null \
  && if docker compose version >/dev/null 2>&1; then docker compose up -d; else ./start.sh; fi"

step "Smoke test"
port="${PORT:-$(remote "sed -n 's/^PORT=//p' $(q "$dir/.env") | tail -1")}"; port="${port:-8787}"
if (( dry )); then
  echo "[dry-run] smoke test on $host: http://127.0.0.1:$port ${tag#v}"
else
  ssh -o BatchMode=yes "$host" "cat > $(q "$dir/smoke-test.sh")" <scripts/smoke-test.sh
  if ! remote "bash $(q "$dir/smoke-test.sh") http://127.0.0.1:$port ${tag#v}"; then
    if [[ -n "$current" && "$current" != "$tag" ]]; then
      echo "smoke test failed – going back to $current" >&2
      remote "cd $(q "$dir") && sed -i 's/^CODECRAFT_VERSION=.*/CODECRAFT_VERSION=$current/' .env \
        && if docker compose version >/dev/null 2>&1; then docker compose up -d; else ./start.sh; fi"
    fi
    fail "deploy of $tag failed"
  fi
fi

step "Done"
echo "$host runs CodeCraft $tag."