Enhance Selectel balance handling and error messaging in GPU rental system

- Updated Selectel documentation to clarify error responses related to balance issues.
- Introduced `peek_balance_rub` function for best-effort balance checks, handling API failures gracefully.
- Improved error messages for insufficient funds and quota issues, specifying Selectel's 403 policy response.
- Added balance checks in the doctor command to ensure users are informed about their balance status before attempting GPU creation.
- Refactored exception handling in cloud operations to provide clearer feedback on balance-related errors.

Co-authored-by: Cursor <cursoragent@cursor.com>
This commit is contained in:
Leonid Pershin
2026-08-23 17:54:11 +03:00
co-authored by Cursor
parent 322577cf9f
commit 0bad693b12
10 changed files with 389 additions and 28 deletions
+1 -1
View File
@@ -142,7 +142,7 @@ Ingress SG: TCP/22 с адреса оператора. 7801 снаружи не
- квота compute/GPU;
- наличие flavor в сегменте (`OS-FLV-DISABLED:disabled`, extra specs);
- баланс / возможность создать ресурс (ошибка 402/403 — понятное сообщение, не traceback).
- баланс / возможность создать ресурс: Selectel при нулевом балансе часто отвечает **403 policy** (`os_compute_api:servers:create`), не 402. CLI пишет «нет средств», не «квота GPU=0».
Письмо в поддержку — ручной шаг один раз, не часть CLI.
+13
View File
@@ -18,6 +18,8 @@ Log = Callable[[str], None]
BALANCES_URL = "https://api.selectel.ru/v3/balances"
DEFAULT_STEP_RUB = 200.0
# Selectel member role often cannot create compute below this (docs/setup.md).
MEMBER_MIN_RUB = 100.0
@dataclass
@@ -124,6 +126,17 @@ def fetch_balance_rub(api_token: str, *, timeout: float = 20.0) -> float:
return balance_rub_from_payload(payload)
def peek_balance_rub(api_token: str | None, *, timeout: float = 8.0) -> float | None:
"""Best-effort balance; None if token missing or API fails."""
token = (api_token or "").strip()
if not token:
return None
try:
return fetch_balance_rub(token, timeout=timeout)
except Exception:
return None
def spend_steps(baseline_rub: float, current_rub: float, step_rub: float) -> int:
"""How many full step intervals have been spent since baseline."""
if step_rub <= 0:
+18 -3
View File
@@ -52,21 +52,36 @@ def _die(exc: BaseException) -> None:
"(Ctrl+C / Ctrl+D на туннеле гасят GPU, диски остаются)"
)
msg = str(exc)
if "gpu-rent status" not in msg and "Дальше:" not in msg:
skip = any(
s in msg
for s in (
"gpu-rent status",
"Дальше:",
"Пополни",
"нет средств",
"gpu-rent up",
)
)
if not skip:
err(hint)
raise typer.Exit(1)
def _stop_after_failed_up(cfg, cause: BaseException) -> None:
"""Delete compute after a failed up so billing does not continue unnoticed."""
from gpu_rent.state import load_state
state = load_state()
if not getattr(state, "server_id", None) and not getattr(state, "floating_ip", None):
# Create never succeeded — nothing to bill, don't print "гашу GPU".
return
warn(
"up упал — гашу GPU (UP_STOP_ON_FAIL; оставить: --keep-on-fail / UP_STOP_ON_FAIL=false)"
)
try:
from gpu_rent.state import load_state
from gpu_rent.vm_logs import print_log_digest
fip = load_state().floating_ip
fip = getattr(state, "floating_ip", None)
if fip:
print_log_digest(cfg, fip, console=console, log=log)
except Exception as dig_exc:
+101 -12
View File
@@ -2,6 +2,7 @@
from __future__ import annotations
import re
import time
from typing import Any, Callable
@@ -27,16 +28,96 @@ from gpu_rent.os_client import (
)
def _wrap(exc: BaseException, what: str) -> CloudError:
def _http_status(exc: BaseException) -> int | None:
for attr in ("http_status", "status_code"):
val = getattr(exc, attr, None)
if isinstance(val, int) and val >= 400:
return val
match = re.search(r"\b([45]\d{2})\b", str(exc)[:200])
return int(match.group(1)) if match else None
def _exc_brief(exc: BaseException, *, limit: int = 160) -> str:
"""One-line exception for logs — no URLs, no user_data / SSH keys."""
text = str(exc)
low = text.lower()
if "402" in text or "payment" in low:
return CloudError(f"{what}: нет средств / 402. {text}")
if "403" in text or "forbidden" in low or "quota" in low:
return CloudError(
f"{what}: квота или запрет (403). Часто GPU=0 — тикет в поддержку, docs/setup.md. {text}"
text = re.split(r"\nValue:|\bValue:|\n#cloud-config|\nusers:", text, maxsplit=1)[0]
text = re.sub(r"https?://\S+", "", text)
text = re.sub(r"ssh-(?:ed25519|rsa|dss) \S+", "ssh-…", text, flags=re.I)
text = re.sub(r"\s+", " ", text).strip(" :.,")
if len(text) > limit:
text = text[: limit - 1] + ""
return text or exc.__class__.__name__
def _is_create_policy_deny(exc: BaseException) -> bool:
low = str(exc).lower()
return "os_compute_api:servers:create" in low or (
"policy" in low and "servers:create" in low
)
def _funds_message(what: str, balance_rub: float | None) -> str:
from gpu_rent.balance import MEMBER_MIN_RUB
if balance_rub is not None and balance_rub <= 0:
return (
f"{what}: нет средств на балансе ({balance_rub:.0f} ₽). "
"Selectel отвечает 403 policy, не 402 и не квота GPU. "
"Пополни в панели Selectel, затем gpu-rent up"
)
return CloudError(f"{what}: {text}")
if balance_rub is not None and balance_rub < MEMBER_MIN_RUB:
return (
f"{what}: Selectel запретил create (403 policy) при балансе "
f"{balance_rub:.0f} ₽ — для member обычно нужно ≥{MEMBER_MIN_RUB:.0f} ₽. "
"Пополни в панели, затем gpu-rent up"
)
if balance_rub is not None:
return (
f"{what}: запрет create (403 policy) при балансе {balance_rub:.0f} ₽ — "
"это IAM/роль сервисного пользователя, не квота GPU. docs/setup.md"
)
return (
f"{what}: Selectel запретил create (403 policy). "
"Обычно кончились деньги, не квота GPU. "
"Пополни баланс в панели; SELECTEL_API_TOKEN покажет сумму. "
"Затем gpu-rent up"
)
def _wrap(
exc: BaseException,
what: str,
*,
api_token: str | None = None,
) -> CloudError:
status = _http_status(exc)
low = str(exc).lower()
brief = _exc_brief(exc)
if status == 402 or "payment required" in low:
return CloudError(
f"{what}: нет средств. Пополни баланс в панели Selectel, затем gpu-rent up"
)
if "quota" in low or "overlimit" in low or "over limit" in low:
return CloudError(
f"{what}: квота исчерпана. Панель IAM → проект → квоты; "
f"GPU=0 — тикет, docs/setup.md. {brief}"
)
create_deny = _is_create_policy_deny(exc) or (
status == 403 and ("create" in what.lower() or what == "сервер")
)
if create_deny:
from gpu_rent.balance import peek_balance_rub
return CloudError(_funds_message(what, peek_balance_rub(api_token)))
if status == 403 or "forbidden" in low:
return CloudError(
f"{what}: запрет (403). Часто нет денег на балансе, не квота GPU. {brief}"
)
return CloudError(f"{what}: {brief}")
def _oid(obj: Any) -> str:
@@ -385,6 +466,7 @@ def create_gpu_server(
spot: bool,
public_key: str,
log: Callable[[str], None],
api_token: str | None = None,
) -> Any:
bdm = [
{
@@ -415,13 +497,12 @@ def create_gpu_server(
"security_groups": [{"name": sg_name}],
}
ud_b64 = _user_data_b64(public_key)
ud_plain = _ssh_user_data(public_key)
# Selectel: OpenStack API expects Base64 user_data; config_drive often unsupported.
# Do not send plaintext user_data — Nova 400 dumps the whole cloud-config (SSH keys).
attempts: list[tuple[str, dict[str, Any]]] = [
("tags+user_data_b64", {**base, "tags": tags, "user_data": ud_b64}),
("user_data_b64", {**base, "user_data": ud_b64}),
("user_data_plain", {**base, "user_data": ud_plain}),
("key_name_only", {**base, "tags": tags}),
("minimal", dict(base)),
]
@@ -435,9 +516,17 @@ def create_gpu_server(
break
except Exception as exc:
last_exc = exc
log(f"create {label}: {exc}")
status = _http_status(exc)
# 401/402/403 / policy deny — payload variants cannot help; stop.
if status in {401, 402, 403} or _is_create_policy_deny(exc):
break
log(f"create {label}: {_exc_brief(exc)}")
if server is None:
raise _wrap(last_exc or RuntimeError("create failed"), "create server")
raise _wrap(
last_exc or RuntimeError("create failed"),
"сервер",
api_token=api_token,
)
if "user_data" not in used:
log(
+43
View File
@@ -40,6 +40,47 @@ class Check:
detail: str
def _check_balance(cfg: Config, checks: list[Check]) -> None:
from gpu_rent.balance import MEMBER_MIN_RUB, fetch_balance_rub
token = (cfg.selectel_api_token or "").strip()
if not token:
checks.append(
Check(
"баланс ₽",
True,
False,
"нет SELECTEL_API_TOKEN — ₽ не проверяем (403 create часто = нет денег)",
)
)
return
try:
rub = fetch_balance_rub(token)
except Exception as exc:
checks.append(Check("баланс ₽", True, False, f"не прочитать: {exc}"))
return
if rub <= 0:
checks.append(
Check(
"баланс ₽",
False,
True,
f"{rub:.0f} ₽ — GPU не создать. Пополни в панели Selectel.",
)
)
elif rub < MEMBER_MIN_RUB:
checks.append(
Check(
"баланс ₽",
True,
False,
f"{rub:.0f} ₽ — мало (member обычно ≥{MEMBER_MIN_RUB:.0f} ₽)",
)
)
else:
checks.append(Check("баланс ₽", True, False, f"{rub:.0f}"))
def run_doctor() -> list[Check]:
checks: list[Check] = []
runtime_dir().mkdir(parents=True, exist_ok=True)
@@ -123,6 +164,8 @@ def run_doctor() -> list[Check]:
except CloudError as exc:
checks.append(Check("квота GPU", False, True, str(exc)))
_check_balance(cfg, checks)
flavors = list(iter_flavors(conn))
gpu_flavors = [f for f in flavors if looks_like_gpu(f)]
ranked = rank_flavors(
+1
View File
@@ -613,6 +613,7 @@ def cmd_up(
spot=spot,
public_key=public_key,
log=log,
api_token=cfg.selectel_api_token,
)
state.server_id = server.id
state.server_name = getattr(server, "name", None)
+16
View File
@@ -1,9 +1,11 @@
from gpu_rent.balance import (
BalanceWatchState,
balance_rub_from_payload,
peek_balance_rub,
plan_balance_notices,
spend_steps,
)
from gpu_rent.errors import CloudError
def test_balance_kopecks():
@@ -54,3 +56,17 @@ def test_plan_low_once():
assert any("низкий" in m for m in msgs)
state, msgs2 = plan_balance_notices(state, 100.0, step_rub=200.0, low_rub=200.0)
assert not any("низкий" in m for m in msgs2)
def test_peek_balance_rub_empty_token():
assert peek_balance_rub("") is None
assert peek_balance_rub(None) is None
def test_peek_balance_rub_swallows_errors(monkeypatch):
def boom(*_a, **_k):
raise CloudError("сеть")
monkeypatch.setattr("gpu_rent.balance.fetch_balance_rub", boom)
assert peek_balance_rub("tok") is None
+111
View File
@@ -0,0 +1,111 @@
from types import SimpleNamespace
import pytest
from gpu_rent.cloud import (
_exc_brief,
_http_status,
_is_create_policy_deny,
_wrap,
create_gpu_server,
)
from gpu_rent.errors import CloudError
POLICY_403 = (
"ForbiddenException: 403: Client Error for url: "
"https://ru-6.cloud.api.selcloud.ru/compute/v2.1/servers, "
"Policy doesn't allow os_compute_api:servers:create to be performed."
)
class Forbidden(Exception):
def __init__(self, msg: str, http_status: int = 403):
super().__init__(msg)
self.http_status = http_status
def test_http_status_from_attr_and_text():
assert _http_status(Forbidden(POLICY_403)) == 403
assert _http_status(RuntimeError("no status")) is None
def test_is_create_policy_deny():
assert _is_create_policy_deny(Forbidden(POLICY_403))
assert not _is_create_policy_deny(RuntimeError("quota exceeded for cores"))
def test_exc_brief_strips_url_and_user_data():
dump = (
"BadRequestException: 400: Invalid input for field/attribute user_data. "
"Value: #cloud-config\nssh_authorized_keys:\n"
" - ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIFake gpu-rent\n"
)
brief = _exc_brief(RuntimeError(dump))
assert "ssh-ed25519" not in brief
assert "AAAAC3" not in brief
assert "http" not in brief.lower() or "https://" not in brief
def test_wrap_403_create_is_funds_not_gpu_quota():
err = _wrap(Forbidden(POLICY_403), "create server")
msg = str(err)
assert isinstance(err, CloudError)
assert "GPU=0" not in msg
assert "тикет" not in msg
assert "ssh-ed25519" not in msg
assert "https://" not in msg
assert "Пополни" in msg or "нет средств" in msg or "кончились деньги" in msg
def test_wrap_403_create_with_zero_balance(monkeypatch):
monkeypatch.setattr("gpu_rent.balance.peek_balance_rub", lambda *_a, **_k: 0.0)
msg = str(_wrap(Forbidden(POLICY_403), "create server", api_token="tok"))
assert "0 ₽" in msg
assert "нет средств" in msg
assert "Пополни" in msg
assert "GPU=0" not in msg
def test_wrap_403_create_with_healthy_balance_is_iam(monkeypatch):
monkeypatch.setattr("gpu_rent.balance.peek_balance_rub", lambda *_a, **_k: 1500.0)
msg = str(_wrap(Forbidden(POLICY_403), "create server", api_token="tok"))
assert "1500 ₽" in msg
assert "IAM" in msg
assert "GPU=0" not in msg
def test_wrap_quota_message_only_when_quota_in_error():
exc = RuntimeError("403 OverLimit: quota exceeded for instances")
msg = str(_wrap(exc, "create server"))
assert "квота" in msg.lower()
def test_create_gpu_server_stops_on_first_403():
calls: list[int] = []
class Compute:
def create_server(self, **kw):
calls.append(1)
raise Forbidden(POLICY_403)
logs: list[str] = []
with pytest.raises(CloudError) as caught:
create_gpu_server(
SimpleNamespace(compute=Compute()),
flavor_id="1",
net_id="n",
sg_name="g",
boot_volume_id="b",
data_volume_id="d",
az="ru-6a",
spot=True,
public_key="ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIFake gpu-rent",
log=logs.append,
)
assert len(calls) == 1
assert logs == []
msg = str(caught.value)
assert "GPU=0" not in msg
assert "тикет" not in msg
assert "ssh-ed25519" not in msg
assert "Пополни" in msg or "нет средств" in msg or "кончились деньги" in msg
+51
View File
@@ -0,0 +1,51 @@
from types import SimpleNamespace
from gpu_rent.doctor import Check, _check_balance
from gpu_rent.errors import CloudError
def test_check_balance_skips_without_token():
checks: list[Check] = []
_check_balance(SimpleNamespace(selectel_api_token=""), checks)
assert checks[0].name == "баланс ₽"
assert checks[0].ok is True
assert checks[0].blocking is False
def test_check_balance_blocks_when_empty(monkeypatch):
monkeypatch.setattr("gpu_rent.balance.fetch_balance_rub", lambda *_a, **_k: 0.0)
checks: list[Check] = []
_check_balance(SimpleNamespace(selectel_api_token="tok"), checks)
assert checks[0].ok is False
assert checks[0].blocking is True
assert "0 ₽" in checks[0].detail
assert "Пополни" in checks[0].detail
def test_check_balance_warns_when_low(monkeypatch):
monkeypatch.setattr("gpu_rent.balance.fetch_balance_rub", lambda *_a, **_k: 40.0)
checks: list[Check] = []
_check_balance(SimpleNamespace(selectel_api_token="tok"), checks)
assert checks[0].ok is True
assert checks[0].blocking is False
assert "40 ₽" in checks[0].detail
def test_check_balance_ok_when_healthy(monkeypatch):
monkeypatch.setattr("gpu_rent.balance.fetch_balance_rub", lambda *_a, **_k: 1200.0)
checks: list[Check] = []
_check_balance(SimpleNamespace(selectel_api_token="tok"), checks)
assert checks[0].ok is True
assert "1200 ₽" in checks[0].detail
def test_check_balance_fetch_error_is_warning(monkeypatch):
def boom(*_a, **_k):
raise CloudError("сеть")
monkeypatch.setattr("gpu_rent.balance.fetch_balance_rub", boom)
checks: list[Check] = []
_check_balance(SimpleNamespace(selectel_api_token="tok"), checks)
assert checks[0].ok is True
assert checks[0].blocking is False
assert "не прочитать" in checks[0].detail
+34 -12
View File
@@ -3,7 +3,14 @@ from types import SimpleNamespace
from gpu_rent.cli import _stop_after_failed_up
def test_stop_after_failed_up_calls_cmd_stop(monkeypatch):
def _quiet(monkeypatch):
monkeypatch.setattr("gpu_rent.cli.warn", lambda *_a, **_k: None)
monkeypatch.setattr("gpu_rent.cli.ok", lambda *_a, **_k: None)
monkeypatch.setattr("gpu_rent.cli.err", lambda *_a, **_k: None)
monkeypatch.setattr("gpu_rent.cli.log", lambda *_a, **_k: None)
def test_stop_after_failed_up_skips_when_no_server(monkeypatch):
calls: list[dict] = []
def fake_stop(cfg, *, no_pull=False, log=None, destroy_disks=False):
@@ -11,13 +18,28 @@ def test_stop_after_failed_up_calls_cmd_stop(monkeypatch):
return SimpleNamespace()
monkeypatch.setattr("gpu_rent.cli.cmd_stop", fake_stop)
monkeypatch.setattr("gpu_rent.cli.warn", lambda *_a, **_k: None)
monkeypatch.setattr("gpu_rent.cli.ok", lambda *_a, **_k: None)
monkeypatch.setattr("gpu_rent.cli.err", lambda *_a, **_k: None)
monkeypatch.setattr("gpu_rent.cli.log", lambda *_a, **_k: None)
_quiet(monkeypatch)
monkeypatch.setattr(
"gpu_rent.state.load_state",
lambda: SimpleNamespace(floating_ip=None),
lambda: SimpleNamespace(server_id=None, floating_ip=None),
)
_stop_after_failed_up(SimpleNamespace(), RuntimeError("boom"))
assert calls == []
def test_stop_after_failed_up_runs_when_server_exists(monkeypatch):
calls: list[dict] = []
def fake_stop(cfg, *, no_pull=False, log=None, destroy_disks=False):
calls.append({"no_pull": no_pull, "destroy_disks": destroy_disks})
return SimpleNamespace()
monkeypatch.setattr("gpu_rent.cli.cmd_stop", fake_stop)
_quiet(monkeypatch)
monkeypatch.setattr(
"gpu_rent.state.load_state",
lambda: SimpleNamespace(server_id="s1", floating_ip=None),
)
_stop_after_failed_up(SimpleNamespace(), RuntimeError("boom"))
@@ -33,20 +55,20 @@ def test_stop_after_failed_up_prints_digest_when_fip(monkeypatch):
return SimpleNamespace()
monkeypatch.setattr("gpu_rent.cli.cmd_stop", fake_stop)
monkeypatch.setattr("gpu_rent.cli.warn", lambda *_a, **_k: None)
monkeypatch.setattr("gpu_rent.cli.ok", lambda *_a, **_k: None)
monkeypatch.setattr("gpu_rent.cli.err", lambda *_a, **_k: None)
monkeypatch.setattr("gpu_rent.cli.log", lambda *_a, **_k: None)
_quiet(monkeypatch)
monkeypatch.setattr(
"gpu_rent.state.load_state",
lambda: SimpleNamespace(floating_ip="10.0.0.1"),
lambda: SimpleNamespace(server_id="s1", floating_ip="10.0.0.1"),
)
monkeypatch.setattr(
"gpu_rent.vm_logs.print_log_digest",
lambda cfg, host, **kw: digests.append(host),
)
_stop_after_failed_up(SimpleNamespace(enable_swarmui=True, llm_runtime="none"), RuntimeError("boom"))
_stop_after_failed_up(
SimpleNamespace(enable_swarmui=True, llm_runtime="none"),
RuntimeError("boom"),
)
assert digests == ["10.0.0.1"]
assert calls == [{"no_pull": True}]