Files
radar_system/python_app/orchestration/restart_policy.py

48 lines
2.0 KiB
Python

"""Crash auto-restart back-off policy (pure, GUI-independent)."""
from __future__ import annotations
import math
from dataclasses import dataclass
@dataclass(frozen=True, slots=True)
class RestartPolicy:
"""Decide when to relaunch a crashed pipeline — retry forever with capped back-off.
This is an unattended appliance, so the pipeline never permanently gives up. The
wait between restart attempts grows with the number of consecutive failures (so a
persistently broken pipeline is not hammered) but is capped at ``max_interval_s``,
and the failure streak resets to zero once genuine data flows again. A burst of
crash signals within the current back-off window collapses to a single restart.
"""
min_interval_s: float = 3.0
max_interval_s: float = 60.0
backoff_factor: float = 2.0
def backoff_for(self, consecutive_failures: int) -> float:
"""Return the seconds to wait before the next restart for this failure streak.
``consecutive_failures`` is the number of restarts already attempted without a
recovery: 0 → ``min_interval_s``, then each additional failure multiplies the
wait by ``backoff_factor``, capped at ``max_interval_s``.
"""
if consecutive_failures <= 0:
return self.min_interval_s
# Beyond this many doublings the wait is always capped; clamp the exponent so a
# long streak cannot overflow ``factor ** n``.
max_exponent = max(1, math.ceil(math.log(self.max_interval_s / self.min_interval_s, self.backoff_factor)))
exponent = min(int(consecutive_failures), max_exponent)
return min(self.min_interval_s * (self.backoff_factor ** exponent), self.max_interval_s)
def should_restart_now(
self,
*,
now_s: float,
last_restart_s: float,
consecutive_failures: int,
) -> bool:
"""Return whether enough back-off has elapsed since the last restart to retry."""
return now_s - last_restart_s >= self.backoff_for(consecutive_failures)