fix: auto-save state on Ctrl+C/SIGTERM, detect interrupted runs

This commit is contained in:
2026-05-22 20:33:25 +04:00
parent 15bb88527d
commit 3222620d0b
2 changed files with 113 additions and 19 deletions

View File

@@ -6,6 +6,8 @@ main.py — Точка входа. Меню выбора режима.
import sys
import os
import argparse
import signal
import atexit
_PROJECT_ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
if _PROJECT_ROOT not in sys.path:
@@ -15,6 +17,31 @@ from core.color import banner, menu, prompt, info, error as cerror, warn
from core import state
def _save_on_exit():
"""Сохраняет состояние при любом завершении (Ctrl+C, SIGTERM, os._exit)."""
state.flush_state()
def _handle_sigterm(signum, frame):
"""Обработчик SIGTERM (kill, systemd stop)."""
state.set_interrupted()
state.flush_state()
sys.exit(128 + signum)
def _handle_keyboard_interrupt():
"""Обработчик Ctrl+C — сохраняет состояние до выхода."""
state.set_interrupted()
state.flush_state()
print()
warn("Прервано (Ctrl+C). Состояние сохранено. Запустите docker-migrate --resume для продолжения.")
sys.exit(130)
# Регистрируем сохранение state при любом завершении
atexit.register(_save_on_exit)
signal.signal(signal.SIGTERM, _handle_sigterm)
def main():
parser = argparse.ArgumentParser(description="Docker Service Migration Tool")
parser.add_argument("--mode", choices=["source", "target"], help="Режим работы")
@@ -42,44 +69,66 @@ def main():
if args.mode == "source":
from source.source import run_source_mode
run_source_mode()
try:
run_source_mode()
except KeyboardInterrupt:
_handle_keyboard_interrupt()
sys.exit(0)
elif args.mode == "target":
if args.remote_dir:
state.set_stage("INIT", mode="target", target_remote_dir=args.remote_dir)
from target.target import run_target_mode
run_target_mode()
try:
run_target_mode()
except KeyboardInterrupt:
_handle_keyboard_interrupt()
sys.exit(0)
# Интерактивный режим
# Проверим, не было ли прерывания в прошлом запуске
recovered = state.recover_after_interrupt()
if recovered:
banner()
warn(f"⚠ Обнаружено прерывание прошлого запуска!")
info(f" Шаг: {recovered['stage']}")
info(f" Режим: {recovered['mode']}")
print(f" {recovered['hint']}")
print()
if confirm("Продолжить с места прерывания", default="y"):
try:
_do_resume()
except KeyboardInterrupt:
_handle_keyboard_interrupt()
sys.exit(0)
else:
info("Сбрасываем прерванное состояние. Можете начать заново.")
state.set_stage("INIT", mode=None, interrupted_at=None, resumable_hint=None)
banner()
while True:
st = state.load_state()
has_resume = bool(st.get("mode") and st.get("stage") and st.get("stage") != "INIT")
# Это resume — если есть mode, stage != INIT, или было прерывание
has_resume = bool(st.get("mode") and st.get("stage") != "INIT")
menu(has_resume=has_resume)
try:
choice = prompt("Ваш выбор", default="0").strip()
except SystemExit:
except EOFError:
break
except KeyboardInterrupt:
_handle_keyboard_interrupt()
if choice == "1":
from source.source import run_source_mode
try:
run_source_mode()
except KeyboardInterrupt:
warn("Прервано")
except Exception as e:
cerror(f"Ошибка: {e}")
sys.exit(1)
_handle_keyboard_interrupt()
break
elif choice == "2":
from target.target import run_target_mode
try:
run_target_mode()
except KeyboardInterrupt:
warn("Прервано")
except Exception as e:
cerror(f"Ошибка: {e}")
sys.exit(1)
_handle_keyboard_interrupt()
break
elif choice == "3":
if not has_resume:
@@ -88,7 +137,7 @@ def main():
try:
_do_resume()
except KeyboardInterrupt:
warn("Прервано")
_handle_keyboard_interrupt()
break
elif choice == "4":
_do_status_logs()
@@ -107,9 +156,13 @@ def _do_resume():
if not mode or stage == "INIT":
info("Нет сохранённого состояния для продолжения")
return
# Сбрасываем флаг прерывания, если был
state.set_stage(stage, interrupted_at=None, resumable_hint=None)
fsm = FSM(mode=mode)
try:
fsm.resume_from(stage)
except KeyboardInterrupt:
_handle_keyboard_interrupt()
except Exception as e:
cerror(f"Ошибка при resume: {e}")
@@ -123,8 +176,10 @@ def _do_status_logs():
print(" 0 Назад")
try:
c = prompt("Выбор", default="0").strip()
except SystemExit:
except EOFError:
break
except KeyboardInterrupt:
_handle_keyboard_interrupt()
if c == "1":
stmod.show_status()
elif c == "2":

View File

@@ -7,17 +7,17 @@ state.py — Управление состоянием (FSM) и resume
import os
import json
import sys
import time
import signal
import atexit
from datetime import datetime
STATE_FILE = None
_LAST_STATE = None
def _state_path():
global STATE_FILE
if STATE_FILE:
return STATE_FILE
# Ищем рядом со скриптом или в /var/lib/migrate
candidates = [
os.path.join(os.path.dirname(os.path.dirname(os.path.abspath(__file__))), ".migrate-state.json"),
os.path.join(os.path.expanduser("~"), ".migrate", "state.json"),
@@ -40,14 +40,18 @@ def _state_path():
def load_state():
global _LAST_STATE
sp = _state_path()
if os.path.isfile(sp):
try:
with open(sp, "r", encoding="utf-8") as f:
return json.load(f)
data = json.load(f)
_LAST_STATE = data
return data
except json.JSONDecodeError:
pass
return _default_state()
_LAST_STATE = _default_state()
return _LAST_STATE
def _default_state():
@@ -61,14 +65,25 @@ def _default_state():
"target_port": 22,
"last_error": None,
"completed_steps": [],
"interrupted_at": None,
"paused": False,
"created_at": datetime.now().isoformat(),
}
def flush_state():
"""Сохраняет текущее _LAST_STATE на диск."""
global _LAST_STATE
sp = _state_path()
with open(sp, "w", encoding="utf-8") as f:
json.dump(_LAST_STATE, f, indent=2, ensure_ascii=False)
def save_state(state):
global _LAST_STATE
sp = _state_path()
state["updated_at"] = datetime.now().isoformat()
_LAST_STATE = state
with open(sp, "w", encoding="utf-8") as f:
json.dump(state, f, indent=2, ensure_ascii=False)
@@ -112,6 +127,26 @@ def clear_error():
save_state(state)
def set_interrupted():
"""Помечает, что скрипт был прерван (Ctrl+C, SIGTERM)."""
state = load_state()
state["interrupted_at"] = datetime.now().isoformat()
state["resumable_hint"] = f"Скрипт был прерван на шаге '{state.get('stage', 'INIT')}'. Запустите: docker-migrate --resume"
save_state(state)
def recover_after_interrupt():
"""Проверяет, был ли предыдущий запуск прерван. Если да — сбрасывает флаг."""
state = load_state()
if state.get("interrupted_at"):
return {
"stage": state.get("stage"),
"mode": state.get("mode"),
"hint": state.pop("resumable_hint", ""),
}
return None
def show_status():
from core.color import info, warn, error as cerror, bold, divider
state = load_state()
@@ -123,6 +158,9 @@ def show_status():
info(f"Completed steps: {', '.join(state.get('completed_steps', [])) or '(none)'}")
if state.get("paused"):
warn("Скрипт на паузе из-за ошибки")
if state.get("interrupted_at"):
warn(f"Последний запуск был прерван в {state['interrupted_at']} на шаге '{state.get('stage', 'INIT')}'")
print(f" Запустите: docker-migrate --resume")
if state.get("manifest_path"):
info(f"Manifest: {state['manifest_path']}")
if state.get("archive_path"):
@@ -154,3 +192,4 @@ def show_logs():
if err.get("suggestion"):
print(f"\nРекомендация:")
print(err["suggestion"])