14b6bbecb4
上一版给 push.sh 加的 pid 比对刚跑就露馅了：`sh: pgrep: command not found`， 于是 BEFORE 和 AFTER 都是空字符串，`[ -n "$BEFORE" ]` 直接跳过检查——我刚加 的安全网自己就是个摆设。 顺手查了一圈，stop.sh 和 start.sh 里同样的 pgrep 也一直在空转： - stop.sh 的 `pgrep ... || exit 0` 每次都以 127 失败，永远匹配不到那个提前 返回，所以每次停服都白等满 15 秒，也从没真正确认过残留 worker 已经没了。 - start.sh 的等待循环同理，白等 20 秒。 DSM 有 pkill 没有 pgrep，而且非特权的 ps 看不见 root 起的进程（服务是开机 以 root 启动的），两者任一都会让检查静默返回「没有」——这正是最危险的答案， 因为它长得和「已经停干净了」一模一样。 - push.sh 改用 `sudo ps -eo pid,args | grep`；并且 AFTER 为空时直接报错退出， 不再把「看不见」当成「通过」 - stop.sh / start.sh 改用 `ps -eo args | grep -q` Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
113 lines
4.8 KiB
Bash
Executable File
113 lines
4.8 KiB
Bash
Executable File
#!/bin/sh
|
|
# Push this working tree to the NAS and restart it.
|
|
#
|
|
# The NAS only accepts password auth, so one ssh master connection is opened
|
|
# up front and every later step rides on it: you type the password once, not
|
|
# five times. (macOS's bundled rsync 2.6.9 cannot carry a password through
|
|
# -e at all — hence tar over ssh.)
|
|
#
|
|
# ./deploy/push.sh [user@host] [port]
|
|
#
|
|
# The password is never stored here — ssh asks for it on your terminal. To
|
|
# stop being asked at all, run `ssh-copy-id -p 2222 ericwyuan@192.168.50.64`
|
|
# once; after that this script runs unattended.
|
|
#
|
|
# Never touches .env, .venv or the database on the far side.
|
|
set -e
|
|
|
|
HOST="${1:-ericwyuan@192.168.50.64}"
|
|
PORT="${2:-2222}"
|
|
# Needed for the sudo restart below. Prompted for rather than stored, and only
|
|
# used over the already-authenticated ssh master connection.
|
|
PASS="${NAS_PASSWORD:-}"
|
|
REPO="$(cd "$(dirname "$0")/.." && pwd)"
|
|
CTL="$(mktemp -u /tmp/garmin-deploy-XXXXXX)"
|
|
|
|
sh_() { ssh -S "$CTL" -o BatchMode=yes "$HOST" "$@"; }
|
|
|
|
cleanup() { ssh -S "$CTL" -O exit "$HOST" 2>/dev/null || true; }
|
|
trap cleanup EXIT
|
|
|
|
echo "==> connecting to $HOST:$PORT (password prompt follows, once)"
|
|
ssh -M -S "$CTL" -fN -p "$PORT" -o ControlPersist=300 "$HOST"
|
|
|
|
if [ -z "$PASS" ]; then
|
|
# Same password as the ssh login; asked for separately because ssh consumed
|
|
# the first one itself and sudo on the far side needs it on stdin.
|
|
printf 'sudo password for %s (needed to restart the root-owned service): ' "$HOST" >&2
|
|
stty -echo 2>/dev/null; read PASS; stty echo 2>/dev/null; echo >&2
|
|
fi
|
|
|
|
# The app dir has moved before; find it rather than assume it.
|
|
APP=$(sh_ 'for d in ~/apps/garmin-health-lab /volume1/web/garmin-health-lab; do
|
|
[ -d "$d/backend" ] && { echo "$d"; break; }; done')
|
|
[ -n "$APP" ] || { echo "cannot find the app dir on $HOST" >&2; exit 1; }
|
|
echo "==> app dir: $APP"
|
|
|
|
# STATIC_DIR is ./static relative to backend/, which is where start.sh cds to.
|
|
STATIC="$APP/backend/static"
|
|
# `cmd && VAR=x` would trip `set -e` when cmd fails, so spell it out.
|
|
if sh_ "[ -f '$APP/static/index.html' ]" 2>/dev/null; then
|
|
STATIC="$APP/static"
|
|
fi
|
|
echo "==> static dir: $STATIC"
|
|
|
|
# macOS bsdtar writes com.apple.provenance xattrs and ._ resource forks that
|
|
# the NAS's tar cannot read; it warns once per file and copies nothing useful.
|
|
TAR="tar czf - --no-xattrs"
|
|
export COPYFILE_DISABLE=1
|
|
|
|
echo "==> backend"
|
|
$TAR --exclude .venv --exclude .env --exclude __pycache__ \
|
|
--exclude '*.db' --exclude tests --exclude .pytest_cache \
|
|
-C "$REPO/backend" . | sh_ "tar xzf - -C '$APP/backend'"
|
|
|
|
echo "==> static (cleared first, so stale JS chunks do not pile up)"
|
|
if [ ! -f "$REPO/client/build/index.html" ]; then
|
|
echo "client/build is missing — run 'npm run build' first" >&2
|
|
exit 1
|
|
fi
|
|
sh_ "rm -rf '$STATIC' && mkdir -p '$STATIC'"
|
|
$TAR -C "$REPO/client/build" . | sh_ "tar xzf - -C '$STATIC'"
|
|
|
|
# The service is started at boot as root (DSM Task Scheduler -> S99garmin.sh),
|
|
# so logs/error.log and logs/access.log are root-owned. Restarting as
|
|
# ericwyuan therefore fails instantly — gunicorn cannot open its own error log
|
|
# — and this went unnoticed for a whole deploy: stop.sh could not kill a root
|
|
# process either, so the OLD master kept serving :8124, start.sh's health
|
|
# check saw a 200 and reported "started", and the new code was never loaded.
|
|
# Hence sudo, matching how the service actually runs.
|
|
SUDO="echo '$PASS' | sudo -S"
|
|
echo "==> restart (sudo: the service runs as root, started at boot)"
|
|
# DSM has no pgrep, and an unprivileged `ps` cannot see a root-owned process
|
|
# — either one silently returns nothing, which would make the pid comparison
|
|
# below always "pass" and put the check right back where it started.
|
|
pids_() {
|
|
sh_ "echo '$PASS' | sudo -S ps -eo pid,args 2>/dev/null \
|
|
| grep '$APP/backend/.venv/bin/gunicorn' | grep -v grep \
|
|
| awk '{print \$1}' | sort -n | tr '\n' ','"
|
|
}
|
|
BEFORE=$(pids_ || true)
|
|
sh_ "cd '$APP' && $SUDO sh deploy/stop.sh >/dev/null 2>&1; sleep 3; $SUDO sh deploy/start.sh" \
|
|
|| { echo "restart failed" >&2; exit 1; }
|
|
|
|
echo "==> health"
|
|
sh_ "curl -sf -m 5 -o /dev/null -w 'local api: %{http_code}\n' \
|
|
http://127.0.0.1:8124/api/health/status" || echo "local api: unreachable"
|
|
|
|
# A 200 alone proves nothing: it is exactly what a surviving old master
|
|
# returns. The master pid must have changed for the new code to be loaded.
|
|
AFTER=$(pids_ || true)
|
|
if [ -z "$AFTER" ]; then
|
|
echo "cannot see any gunicorn process - the pid check could not run." >&2
|
|
echo "Verify by hand before trusting this deploy." >&2
|
|
exit 1
|
|
fi
|
|
if [ "$BEFORE" = "$AFTER" ]; then
|
|
echo "the gunicorn pids did not change ($AFTER) - the old process is still" >&2
|
|
echo "serving and your changes are NOT live. Check $APP/logs/error.log." >&2
|
|
exit 1
|
|
fi
|
|
echo "==> gunicorn restarted: $BEFORE -> $AFTER"
|
|
echo "done. The public URL takes a few seconds longer (frp reconnecting)."
|