Class ApiKey

contrib/multilingual/api_pool.py:81–123

A single API key with concurrency and rate-limit metadata. Attributes ---------- key : API key string (e.g. ``"sk-or-xxx"``). base_url : Optional base URL override for the provider endpoint. model : Model label to use with this key. rate_limited :

@dataclass
class ApiKey:
    """A single API key with concurrency and rate-limit metadata.

    Attributes
    ----------
    key :
        API key string (e.g. ``"sk-or-xxx"``).
    base_url :
        Optional base URL override for the provider endpoint.
    model :
        Model label to use with this key.
    rate_limited :
        ``True`` when this key is cooling down after a 429 response.
    rate_limited_until :
        Monotonic timestamp when this key becomes eligible again after a
        429.  Only meaningful when *rate_limited* is ``True``.
    consecutive_429 :
        Count of consecutive rate-limit hits.  Used to compute the next
        backoff duration via :math:`30 \\times 2^n` seconds, capped at 300.
    total_requests :
        Cumulative request count served by this key.  Used for
        least-loaded scheduling.
    active_requests :
        Number of callers currently using this key.
    max_concurrent :
        Maximum number of simultaneous callers allowed on this key
        (default 5).  One key serves up to this many concurrent LLM calls.
    """

    key: str
    base_url: str | None
    model: str
    rate_limited: bool = False
    rate_limited_until: float = 0.0
    consecutive_429: int = 0
    total_requests: int = 0
    active_requests: int = 0
    max_concurrent: int = _DEFAULT_MAX_CONCURRENT_PER_KEY

    @property
    def available(self) -> bool:
        """``True`` when this key can accept at least one more caller."""
        return not self.rate_limited and self.active_requests < self.max_concurrent


# ---------------------------------------------------------------------------