BerriAI/litellm · error · KeyNotInCacheError

Key doesn't exist in cache + check_cache_only=True. key={has

Error message

Key doesn't exist in cache + check_cache_only=True. key={hashed_token}.

What it means

Raised by _resolve_key when the caller demanded cache-only resolution (check_cache_only=True — typically under high load or DB outage) and the hashed token had no cache entry. It means the key was never seen (unknown/invalid token) or its cache entry expired and DB fallback is forbidden, so authentication fails without hitting the database; the hashed token is embedded for correlation.

Source

Thrown at litellm/proxy/auth/resolvers/store.py:111

        Stopgap for the request flow that still consumes ``UserAPIKeyAuth`` for
        budget, rate-limit, and policy state. Only Principals produced by
        ``resolve`` carry a source key.
        """
        if principal.source_key is None:
            raise PrincipalMissingSourceKeyError()
        return principal.source_key

    async def _resolve_key(self, hashed_token: str) -> UserAPIKeyAuth:
        if self._prisma is None:
            raise NoDatabaseConnectionError()

        cached: Final = await self._cache.async_get_cache(key=hashed_token, model_type=UserAPIKeyAuth)
        if cached is not None:
            return _copy_user_api_key_auth_for_cache(user_api_key_obj=cached)

        if self._check_cache_only:
            raise KeyNotInCacheError(hashed_token)

        from_db: Final[BaseModel | None] = await _fetch_key_object_from_db_with_reconnect(
            hashed_token=hashed_token,
            prisma_client=self._prisma,
            parent_otel_span=self._parent_otel_span,
            proxy_logging_obj=self._proxy_logging_obj,
        )
        if from_db is None:
            raise KeyNotFoundError(hashed_token)

        key: Final = UserAPIKeyAuth.model_validate(from_db.model_dump(exclude_none=True))

        if key.object_permission_id and not key.object_permission:
            try:
                key.object_permission = await get_object_permission(
                    object_permission_id=key.object_permission_id,
                    prisma_client=self._prisma,
                    user_api_key_cache=self._cache,

View on GitHub (pinned to 77b7c6c40c)

Solutions

1. Disable check_cache_only so the resolver can fall back to the database.
2. Warm the cache or reissue the key so it exists in cache.

Example fix

Retry the request without check_cache_only=True, or restart the proxy to refresh its cache.

When it happens

Trigger: Thrown at litellm/proxy/auth/resolvers/store.py:111 when the library encounters an invalid state.

Common situations: The key was looked up with check_cache_only=True and was not present in the cache.

AI-assisted analysis of BerriAI/litellm@77b7c6c40c (2026-08-18). Data as JSON: /api/errors/e64fa54d75e4b047. Report an issue: GitHub.