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)
75 lines
3.6 KiB
Python
75 lines
3.6 KiB
Python
"""Shared prompt-caching translation for OpenAI model providers.
|
|
|
|
OpenAI caches prompt prefixes automatically server-side and routes reads on a caller-supplied
|
|
``prompt_cache_key``. It exposes no cache-point placement knobs, so of ``CacheConfig`` only
|
|
``cache_key`` (and, when it already names a valid retention literal, ``ttl``) maps onto the request.
|
|
"""
|
|
|
|
import warnings
|
|
from typing import Any
|
|
|
|
from ..agent.agent_metadata import AgentMetadata
|
|
from ._validation import warn_on_cache_config_not_supported
|
|
from .model import CacheConfig
|
|
|
|
# OpenAI's prompt_cache_retention accepts only these literals. ttl maps through only on an exact
|
|
# match - the SDK never guesses a conversion from an arbitrary duration string.
|
|
#
|
|
# prompt_cache_retention is deprecated in openai 2.54.0 in favor of prompt_cache_options.ttl, whose only
|
|
# accepted value today is "30m" - so "24h"/"in_memory" remain expressible only through this field.
|
|
_RETENTION_LITERALS = frozenset({"in_memory", "24h"})
|
|
|
|
|
|
def _resolve_cache_key(cache_config: CacheConfig, agent_metadata: AgentMetadata | None) -> str | None:
|
|
"""Resolve the prompt-cache routing key: configured value wins, else derive from the session.
|
|
|
|
Returns the configured ``cache_key`` when it names one; ``cache_key=False`` is an explicit opt-out
|
|
(no key). Left unset, it derives ``strands-<session_id>`` when the agent carries a session id, else None.
|
|
"""
|
|
if cache_config.cache_key is False:
|
|
return None
|
|
if cache_config.cache_key is not None:
|
|
return cache_config.cache_key
|
|
if agent_metadata is not None and agent_metadata.session_id is not None:
|
|
return f"strands-{agent_metadata.session_id}"
|
|
return None
|
|
|
|
|
|
def apply_cache_config(
|
|
request: dict[str, Any], cache_config: CacheConfig | None, agent_metadata: AgentMetadata | None = None
|
|
) -> None:
|
|
"""Map a ``CacheConfig`` onto an OpenAI request in place.
|
|
|
|
An explicit value already present in ``request`` (carried in from the user's ``params``) always
|
|
wins; this fills in only what ``params`` did not set. ``strategy`` and ``system_prompt_ttl`` are
|
|
accepted but have no effect, and a ``ttl`` that is not an OpenAI retention literal is ignored;
|
|
each such no-op is surfaced through ``warnings.warn`` (deduped per call site by the standard
|
|
library's default filter), matching the config-validation warnings in ``_validation.py``.
|
|
|
|
The prompt-cache routing key resolves as: the configured ``cache_key`` wins when set to a string,
|
|
``cache_key=False`` opts out; otherwise it falls back to ``strands-<session_id>`` when the agent
|
|
carries a session id. A falsy result (empty or None) emits no key.
|
|
|
|
Args:
|
|
request: The request dict being assembled; mutated in place.
|
|
cache_config: The provider's configured cache settings, if any.
|
|
agent_metadata: The invoking agent's metadata, used to derive a routing key when one is unset.
|
|
"""
|
|
if cache_config is None:
|
|
return
|
|
|
|
cache_key = _resolve_cache_key(cache_config, agent_metadata)
|
|
if cache_key and "prompt_cache_key" not in request:
|
|
request["prompt_cache_key"] = cache_key
|
|
|
|
if cache_config.ttl is not None and "prompt_cache_retention" not in request:
|
|
if cache_config.ttl in _RETENTION_LITERALS:
|
|
request["prompt_cache_retention"] = cache_config.ttl
|
|
else:
|
|
warnings.warn(
|
|
f"cache_config.ttl={cache_config.ttl!r} is not an openai retention value "
|
|
"('in_memory' or '24h') and will be ignored",
|
|
stacklevel=4,
|
|
)
|
|
|
|
warn_on_cache_config_not_supported(cache_config, "OpenAI", supported={"cache_key", "ttl"})
|