b60439152b
- harness/es.py: append-only event store + проекции (образец — кооп-портал) - harness/levels.py: уровни 1с/1м/1ч/1д, контекст ∝ масштабу - harness/scheduler.py: уровневые потоки, идемпотентные тики, absorbed-стабы - harness/isolated_runner.py + _llm_worker.py: LLM-тик в сабпроцессе (глобальный asyncio-рантайм Strands не переживает 2+ create_harness) - cli.py: init/task/tasks/status/events + requeue (написан самим минутным агентом) - 9router: отдельный ключ sk-simbiot, модель auto (glm-5.3 первым)
40 lines
1.9 KiB
Python
40 lines
1.9 KiB
Python
"""The default routing strategy: ordered failover that prefers the candidates failing least."""
|
|
|
|
from __future__ import annotations
|
|
|
|
from typing import TYPE_CHECKING, Any
|
|
|
|
from .strategy import RoutingContext, _attempts_since_success
|
|
|
|
if TYPE_CHECKING:
|
|
from .router import RoutingCandidate
|
|
|
|
|
|
class FallbackStrategy:
|
|
"""Picks the healthiest candidate not yet tried since the last success.
|
|
|
|
Candidates already tried since the last success are excluded, then the fewest recorded failures
|
|
wins, ties going to the earlier declaration. So an invocation with no failures behind it is plain
|
|
declaration order, and a model that keeps failing sinks below healthier ones rather than being
|
|
re-tried in its declared slot.
|
|
|
|
A success re-arms every candidate, since exclusion looks only at attempts since the last success, and
|
|
clears the succeeding candidate's own failure count. Every other candidate keeps its failure history,
|
|
so a model that fails repeatedly stays below healthier ones across successes. Returns ``None`` once
|
|
every candidate has been tried since the last success.
|
|
"""
|
|
|
|
async def select(self, context: RoutingContext, **kwargs: Any) -> RoutingCandidate | None:
|
|
"""Return the least-failed candidate not yet tried since the last success, else ``None``."""
|
|
failures: dict[int, int] = {}
|
|
for attempt in context.attempts:
|
|
if attempt.exception is None:
|
|
failures.pop(id(attempt.candidate), None)
|
|
else:
|
|
failures[id(attempt.candidate)] = failures.get(id(attempt.candidate), 0) + 1
|
|
|
|
tried_now = {id(attempt.candidate) for attempt in _attempts_since_success(context.attempts)}
|
|
available = [candidate for candidate in context.candidates if id(candidate) not in tried_now]
|
|
if not available:
|
|
return None
|
|
return min(available, key=lambda candidate: failures.get(id(candidate), 0))
|