"""T5.4 能力协商 / 降级(ARCH §4.4)。 结构化输出:原生不支持时降级(instructor JSON-提示路径仍由适配器自处理;网关 层负责在链内**优先选支持结构化输出的 provider**,无则降级到首个可用、对上层透明)。 降级时 `served_by.degraded=True` 标注,记账/日志可见,正确性不受影响。 """ from __future__ import annotations import uuid from fakes_resilience import FakeLedger, ScriptedAdapter, chain, chain_resolver from pydantic import BaseModel from ww_llm_gateway.adapters.base import Capabilities from ww_llm_gateway.gateway import Gateway from ww_llm_gateway.types import Block, LlmRequest, Scope class Tiny(BaseModel): x: str def _structured_req() -> LlmRequest: return LlmRequest( tier="analyst", input="给我结构化", output_schema=Tiny, scope=Scope(user_id=uuid.UUID(int=1)), ) async def test_prefers_structured_capable_provider_in_chain() -> None: # 主 provider 不支持结构化输出,回退 provider 支持 → 网关优先选支持者服务结构化请求。 no_struct = ScriptedAdapter( "weakprov", capabilities_=Capabilities(structured_output=False, prefix_cache=False), ) struct = ScriptedAdapter( "strongprov", capabilities_=Capabilities(structured_output=True, prefix_cache=True), ) ledger = FakeLedger() gw = Gateway( {"weakprov": no_struct, "strongprov": struct}, ledger, chain_resolver=chain_resolver(chain(("weakprov", "w"), ("strongprov", "s"))), ) resp = await gw.run(_structured_req()) assert resp.served_by.provider == "strongprov" assert resp.parsed is not None assert no_struct.complete_calls == 0 async def test_degrades_when_no_structured_capable_provider() -> None: # 链上无 provider 支持结构化输出 → 降级用首个可用(适配器自走 instructor JSON 提示) # 并标 served_by.degraded=True,不硬失败。 weak = ScriptedAdapter( "weakprov", capabilities_=Capabilities(structured_output=False), ) gw = Gateway( {"weakprov": weak}, FakeLedger(), chain_resolver=chain_resolver(chain(("weakprov", "w"))), ) resp = await gw.run(_structured_req()) assert resp.served_by.provider == "weakprov" assert resp.served_by.degraded is True assert weak.complete_calls == 1 async def test_no_degradation_flag_for_plain_text() -> None: # 纯文本请求对任何 provider 都不算降级。 weak = ScriptedAdapter("weakprov", capabilities_=Capabilities(structured_output=False)) gw = Gateway( {"weakprov": weak}, FakeLedger(), chain_resolver=chain_resolver(chain(("weakprov", "w"))), ) req = LlmRequest(tier="writer", input="正文", scope=Scope(user_id=uuid.UUID(int=1))) resp = await gw.run(req) assert resp.served_by.degraded is False async def test_cache_blocks_passed_through_regardless_of_capability() -> None: # 前缀缓存不支持时只是跳过,不改正确性、不算降级。 weak = ScriptedAdapter("weakprov", capabilities_=Capabilities(prefix_cache=False)) gw = Gateway( {"weakprov": weak}, FakeLedger(), chain_resolver=chain_resolver(chain(("weakprov", "w"))), ) req = LlmRequest( tier="writer", input="正文", system=[Block(text="世界观硬规则", cache=True)], scope=Scope(user_id=uuid.UUID(int=1)), ) resp = await gw.run(req) assert resp.text assert resp.served_by.degraded is False