bugfix: comprehensive stability audit and fixes
- discover/docker.py: handle comma-separated compose_file in labels
- discover/network.py: replace os.getlogin() with robust user detection
- target.py: add lsb_release fallback via hostnamectl, guard None compose_file,
use container name (not cached CID) for docker logs
- main.py: call reset_state(mode='target') before target mode,
improve EOF handling info message
- source.py: remove redundant set_stage('DONE') inside transfer_offer
- transfer.py: fix stage naming for resume after transfer
- add dry_run.py for local logic validation
This commit is contained in:
+44
-25
@@ -28,9 +28,17 @@ def do_preflight():
|
||||
state.set_stage("TARGET_PREFLIGHT")
|
||||
|
||||
# Ubuntu version
|
||||
out = run("lsb_release -ds", check=False).stdout.strip()
|
||||
info(f"OS: {out}")
|
||||
version_match = re.search(r'(\d+\.\d+)', out)
|
||||
try:
|
||||
out = run("lsb_release -ds", check=False).stdout.strip()
|
||||
except RuntimeError:
|
||||
out = ""
|
||||
if not out and exists("hostnamectl"):
|
||||
try:
|
||||
out = run("hostnamectl | grep 'Operating System'", check=False).stdout.strip()
|
||||
except RuntimeError:
|
||||
out = ""
|
||||
info(f"OS: {out or 'не определено'}")
|
||||
version_match = re.search(r'(\d+\.\d+)', out) or re.search(r'Ubuntu\s+(\d+\.\d+)', out, re.I)
|
||||
ubuntu_version = version_match.group(1) if version_match else None
|
||||
if not ubuntu_version:
|
||||
warn("Не удалось определить версию Ubuntu. Продолжаем на свой риск.")
|
||||
@@ -346,23 +354,27 @@ def do_verify():
|
||||
|
||||
# Docker compose up
|
||||
cf = manifest["docker"].get("compose_file")
|
||||
if cf and os.path.isfile(cf):
|
||||
info("Запускаем docker compose up ...")
|
||||
compose_dir = os.path.dirname(cf)
|
||||
r = run(f"cd {compose_dir} \u0026\u0026 {sudo} docker compose up -d", check=False)
|
||||
if r.returncode != 0:
|
||||
print(f"\n{r.stdout}")
|
||||
print(f"\n{r.stderr}")
|
||||
state.set_error(
|
||||
step="docker_compose_up",
|
||||
stdout=r.stdout,
|
||||
stderr=r.stderr,
|
||||
suggestion="Проверьте compose-файл, доступность image, volumes. После исправления запустите: ./migrate --resume"
|
||||
)
|
||||
raise RuntimeError("docker compose up failed")
|
||||
success("Docker compose up выполнен")
|
||||
if cf:
|
||||
cf_path = cf
|
||||
if os.path.isfile(cf_path):
|
||||
info("Запускаем docker compose up ...")
|
||||
compose_dir = os.path.dirname(cf_path)
|
||||
r = run(f"cd {compose_dir} && {sudo} docker compose up -d", check=False)
|
||||
if r.returncode != 0:
|
||||
print(f"\n{r.stdout}")
|
||||
print(f"\n{r.stderr}")
|
||||
state.set_error(
|
||||
step="docker_compose_up",
|
||||
stdout=r.stdout,
|
||||
stderr=r.stderr,
|
||||
suggestion="Проверьте compose-файл, доступность image, volumes. После исправления запустите: docker-migrate --resume"
|
||||
)
|
||||
raise RuntimeError("docker compose up failed")
|
||||
success("Docker compose up выполнен")
|
||||
else:
|
||||
warn(f"Compose-файл из манифеста не найден на target: {cf_path}")
|
||||
elif manifest["docker"].get("container_name"):
|
||||
info("Compose не найден, пробуем docker run по параметрам контейнера ...")
|
||||
info("Compose не найден, запуск по docker run не реализован. Используйте docker run вручную.")
|
||||
|
||||
# Sidecar units start
|
||||
for u in manifest.get("systemd_units", []):
|
||||
@@ -371,12 +383,19 @@ def do_verify():
|
||||
run(f"{sudo} systemctl start {uname}", check=False)
|
||||
info(f"Запущен unit: {uname}")
|
||||
|
||||
# Логи
|
||||
cid = manifest["docker"].get("container_id")
|
||||
if cid:
|
||||
info("Последние логи контейнера:")
|
||||
log_out = run(f"docker logs --tail 30 {cid[:12]}", check=False)
|
||||
print(log_out.stdout)
|
||||
# Логи — ищем контейнер по имени, а не по ID (ID с другого сервера)
|
||||
cname = manifest["docker"].get("container_name")
|
||||
if cname:
|
||||
try:
|
||||
out = run(f"docker ps -q -f name={cname}", check=False)
|
||||
if out.stdout.strip():
|
||||
info("Последние логи контейнера:")
|
||||
log_out = run(f"docker logs --tail 30 {cname}", check=False)
|
||||
print(log_out.stdout)
|
||||
else:
|
||||
warn(f"Контейнер '{cname}' пока не запущен, логи недоступны")
|
||||
except Exception as e:
|
||||
warn(f"Не удалось получить логи: {e}")
|
||||
|
||||
success("Верификация завершена. Сервис должен работать.")
|
||||
state.set_stage("DONE")
|
||||
|
||||
Reference in New Issue
Block a user