fix(alerts): краш при restart-политике, гистерезис закрытия, редкие правки
CI / ruff · mypy · pytest (pull_request) Successful in 24m47s
CI / ruff · mypy · pytest (pull_request) Successful in 24m47s
Найдено смоуком на живых контейнерах (examples/, #19): - контейнер с restart: on-failure поднимается за секунду — между раундами его не видно в exited; краш теперь распознаётся и по росту RestartCount (сравнение со снапшотами до текущего раунда), повторы — счётчик - алерт закрывается, только когда условия нет resolve_after (60 с): иначе флапающий сервис получал бы «✅ восстановлено» после каждого раунда - правки сообщений со счётчиком — не чаще раза в минуту на алерт Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
+10
-2
@@ -28,6 +28,7 @@ class AlertsConfig(BaseModel):
|
||||
disk_path: Path = Path("/data")
|
||||
dependency_down_for: int = Field(default=300, gt=0)
|
||||
repeat_window: int = Field(default=600, ge=0)
|
||||
resolve_after: int = Field(default=60, ge=0)
|
||||
log_tail: int = Field(default=30, ge=0, le=200)
|
||||
|
||||
|
||||
@@ -56,6 +57,7 @@ class Condition:
|
||||
service: str | None
|
||||
payload: dict[str, Any] = field(default_factory=dict)
|
||||
container_id: str | None = None # for the log tail and the «📜 Логи» / «🩺 Статус» buttons
|
||||
bump: bool = False # a fresh incident this round: if the alert is still open, its counter grows
|
||||
|
||||
@property
|
||||
def key(self) -> tuple[AlertKind, str | None]:
|
||||
@@ -69,13 +71,19 @@ def container_conditions(
|
||||
d = o.details
|
||||
found: list[Condition] = []
|
||||
|
||||
def add(kind: AlertKind, **payload: Any) -> None:
|
||||
found.append(Condition(kind, d.name, payload, d.id))
|
||||
def add(kind: AlertKind, bump: bool = False, **payload: Any) -> None:
|
||||
found.append(Condition(kind, d.name, payload, d.id, bump))
|
||||
|
||||
# A restart policy brings a crashed container back within a second — between two rounds it was already
|
||||
# restarted, so a grown restart count is a crash too (the exit code of that run is gone by then).
|
||||
before = [s for s in history if o.at is None or as_utc(s.at) < o.at] # not this round's own row
|
||||
restarted = bool(before) and d.restart_count > before[-1].restart_count
|
||||
if d.state in ("exited", "dead") and d.oom_killed:
|
||||
add(AlertKind.OOM, exit_code=d.exit_code)
|
||||
elif d.state in ("exited", "dead") and d.exit_code != 0:
|
||||
add(AlertKind.CRASH, exit_code=d.exit_code)
|
||||
elif restarted or d.state == "restarting":
|
||||
add(AlertKind.CRASH, bump=restarted, exit_code=d.exit_code or "≠ 0")
|
||||
|
||||
window = [s for s in history if as_utc(s.at) >= now - timedelta(seconds=cfg.restart_loop_window)]
|
||||
if window:
|
||||
|
||||
Reference in New Issue
Block a user