feat: add openai-compatible transport with sse parsing and error translation

This commit is contained in:
2026-07-20 06:45:55 -04:00
parent 315142ceb1
commit 454a8b5e0f
2 changed files with 559 additions and 0 deletions
+304
View File
@@ -0,0 +1,304 @@
"""OpenAI 兼容 transport(D2 默认): 手写 httpx + SSE 解析 + 错误翻译。
SSE 纯函数移植 VT `adapters/llm.py:51-124`;错误翻译移植 CHS
`app/providers/invokers.py:127-227`。职责只到"一次原始调用"——重试/限流/
缓存归中间件。看门狗活性口径: content 与 reasoning_content 增量都作为流
元素产出,思考流天然刷新计时(CHS 迁移约束 R1)。
"""
from __future__ import annotations
import json
import re
import time
from typing import TYPE_CHECKING, Any
import httpx
from polygateway.errors import RequestRejectedError, SourceDeadError, TransientError
from polygateway.providers import ProviderProfile, get_provider
from polygateway.streaming import StreamLivenessTimeout, stream_with_liveness_timeouts
from polygateway.types import SourceConfig, TransportResult
if TYPE_CHECKING:
from collections.abc import AsyncIterator, Callable, Mapping
_THINK_PATTERN = re.compile(r"<think>(.*?)</think>", re.DOTALL)
# —— SSE 纯函数(VT llm.py 同款)——
def _sse_data_payload(raw: str) -> str | None:
"""提取 SSE data 行载荷;ping(: 开头)/空行/非 data 行返回 None 跳过。"""
line = raw.strip()
if not line or line.startswith(":") or not line.startswith("data:"):
return None
return line[len("data:") :].strip()
def _sse_delta(chunk: dict[str, Any], usage_sink: dict[str, Any]) -> tuple[bool, str] | None:
"""从 chunk 提取增量: (True, content) 或 (False, reasoning);usage 帧旁路进 sink。"""
if chunk.get("usage"):
usage_sink["usage"] = chunk["usage"]
choices = chunk.get("choices") or []
if not choices:
return None
delta = choices[0].get("delta") or {}
content = delta.get("content")
if content:
return (True, content)
reasoning = delta.get("reasoning_content")
if reasoning:
return (False, reasoning)
return None
async def _iter_sse_deltas(
lines: AsyncIterator[str], usage_sink: dict[str, Any]
) -> AsyncIterator[tuple[bool, str]]:
"""逐行解析 SSE 流;[DONE] 置 sink["done"];畸形 JSON 帧 → 瞬时错误(可重试)。"""
async for raw in lines:
data = _sse_data_payload(raw)
if data is None:
continue
if data == "[DONE]":
usage_sink["done"] = True
return
try:
chunk = json.loads(data)
except json.JSONDecodeError as exc:
raise TransientError(f"SSE 帧畸形(malformed_json): {data[:80]!r}", operation="chat") from exc
delta = _sse_delta(chunk, usage_sink)
if delta is not None:
yield delta
# —— 错误翻译(CHS invokers.py 同款)——
def _parse_retry_after(raw: str | None) -> float | None:
"""解析 Retry-After 头;仅支持秒数形态,HTTP-date 返回 None(CHS 同款)。"""
if raw is None:
return None
try:
seconds = float(raw.strip())
except ValueError:
return None
return seconds if seconds > 0 else None
def _translate_429(source: SourceConfig, body_text: str, headers: Mapping[str, str]) -> Exception:
try:
err_type = json.loads(body_text).get("error", {}).get("type", "")
except (json.JSONDecodeError, AttributeError):
err_type = ""
if err_type == "insufficient_quota":
return SourceDeadError(
f"{source.name} 配额耗尽(insufficient_quota)",
source_name=source.name, status_code=429, operation="chat",
)
return TransientError(
f"{source.name} 限速: 429",
retry_after_s=_parse_retry_after(headers.get("retry-after")),
source_name=source.name, status_code=429, operation="chat",
)
def _status_to_error(
source: SourceConfig, status: int, body_text: str, headers: Mapping[str, str]
) -> Exception:
ctx: dict[str, Any] = {"source_name": source.name, "status_code": status, "operation": "chat"}
if status in (401, 403):
return SourceDeadError(f"{source.name} 凭据失效/欠费: {status}", **ctx)
if status == 400:
return RequestRejectedError(f"{source.name} 请求被拒: 400", **ctx)
if status == 429:
return _translate_429(source, body_text, headers)
if status >= 500:
return TransientError(f"{source.name} 瞬时错误: {status}", **ctx)
return RequestRejectedError(f"{source.name} 客户端错误: {status}", **ctx)
def _strip_think(content: str) -> tuple[str, str]:
"""剥离 <think> 标签(qwen 系),返回 (正文, 思考流)。VT llm.py:147-164 同款。"""
match = _THINK_PATTERN.search(content)
if match is None:
return content, ""
return _THINK_PATTERN.sub("", content).strip(), match.group(1).strip()
def _resolve_usage(usage: dict[str, Any], source: SourceConfig) -> tuple[int, int, str]:
"""usage 帧读取;缺失/非法按 est_tokens 保守兜底并标 estimated(CHS invokers.py:241)。"""
prompt, completion = usage.get("prompt_tokens"), usage.get("completion_tokens")
if isinstance(prompt, int) and isinstance(completion, int) and prompt + completion > 0:
return prompt, completion, "measured"
return 0, source.est_tokens, "estimated"
def _default_client_factory(source: SourceConfig) -> httpx.AsyncClient:
return httpx.AsyncClient(
headers={"Authorization": f"Bearer {source.api_key}"},
timeout=httpx.Timeout(source.timeout_s),
trust_env=source.trust_env,
)
class OpenAICompatTransport:
"""默认 transport: 每源一个预配 httpx client,懒创建,aclose 统一释放。"""
def __init__(
self,
*,
registry: Mapping[str, ProviderProfile] | None = None,
client_factory: Callable[[SourceConfig], httpx.AsyncClient] | None = None,
) -> None:
self._registry = registry
self._client_factory = client_factory or _default_client_factory
self._clients: dict[str, httpx.AsyncClient] = {}
def _client_for(self, source: SourceConfig) -> httpx.AsyncClient:
client = self._clients.get(source.name)
if client is None:
client = self._client_factory(source)
self._clients[source.name] = client
return client
def _build_payload(
self, *, messages: list[dict[str, Any]], source: SourceConfig,
profile: ProviderProfile, stream: bool, overlay: dict[str, Any],
) -> dict[str, Any]:
payload: dict[str, Any] = {"model": source.model, "messages": messages, "stream": stream}
if stream:
payload["stream_options"] = {"include_usage": True} # 强制 usage 帧(三项目同款)
if source.enable_thinking is True:
payload.update(profile.thinking_on)
elif source.enable_thinking is False:
payload.update(profile.thinking_off)
payload.update(overlay)
return payload
async def complete(
self, *, messages: list[dict[str, Any]], source: SourceConfig,
stream: bool, overlay: dict[str, Any], call_id: str,
) -> TransportResult:
"""一次原始调用;HTTP/线路/流式异常按 ARCH §6.2 翻译为领域错误。"""
profile = get_provider(source.provider, registry=self._registry)
payload = self._build_payload(
messages=messages, source=source, profile=profile, stream=stream, overlay=overlay
)
url = source.base_url.rstrip("/") + "/chat/completions"
client = self._client_for(source)
ctx: dict[str, Any] = {"source_name": source.name, "operation": "chat"}
try:
if stream:
return await self._complete_stream(client, url, payload, source, profile)
return await self._complete_once(client, url, payload, source, profile)
except StreamLivenessTimeout as exc:
raise TransientError(f"{source.name} 流活性超时({exc.kind})", **ctx) from exc
except httpx.TimeoutException as exc:
raise TransientError(f"{source.name} 超时: {exc}", **ctx) from exc
except httpx.TransportError as exc:
# VT 宽集: 覆盖断连/协议错误/读写失败(设计 §9 行 8)
raise TransientError(f"{source.name} 网络错误: {exc}", **ctx) from exc
async def _complete_stream(
self, client: httpx.AsyncClient, url: str, payload: dict[str, Any],
source: SourceConfig, profile: ProviderProfile,
) -> TransportResult:
started = time.monotonic()
async with client.stream("POST", url, json=payload) as resp:
if resp.status_code != 200:
body = (await resp.aread()).decode("utf-8", errors="replace")
raise _status_to_error(source, resp.status_code, body, resp.headers)
sink: dict[str, Any] = {}
guarded = stream_with_liveness_timeouts(
_iter_sse_deltas(resp.aiter_lines(), sink),
ttft_s=source.ttft_timeout_s or source.timeout_s,
inter_token_s=source.inter_token_timeout_s or source.timeout_s,
total_s=source.timeout_s,
)
content_parts: list[str] = []
thinking_parts: list[str] = []
ttft_ms: float | None = None
last = started
max_gap = 0.0
async for is_content, text in guarded:
now = time.monotonic()
if ttft_ms is None:
ttft_ms = (now - started) * 1000
else:
max_gap = max(max_gap, (now - last) * 1000)
last = now
(content_parts if is_content else thinking_parts).append(text)
salvaged = self._check_done(sink, content_parts, thinking_parts, source)
content, thinking = self._finalize_text(content_parts, thinking_parts, profile)
prompt, completion, usage_source = _resolve_usage(sink.get("usage") or {}, source)
if salvaged:
usage_source = "estimated" # 打捞路径强制 estimated(设计 §6)
return TransportResult(
content=content, thinking=thinking, prompt_tokens=prompt,
completion_tokens=completion, usage_source=usage_source,
ttft_ms=ttft_ms, max_inter_token_ms=(max_gap if ttft_ms is not None else None),
raw={"usage": sink.get("usage")},
)
def _check_done(
self, sink: dict[str, Any], content_parts: list[str],
thinking_parts: list[str], source: SourceConfig,
) -> bool:
"""缺 [DONE] 语义(设计 §6): 零内容恒 retry;有内容按 missing_done 策略。"""
if sink.get("done"):
return False
ctx: dict[str, Any] = {"source_name": source.name, "operation": "chat"}
if not content_parts and not thinking_parts:
raise TransientError(f"{source.name} SSE early_eof: 零内容断流", **ctx)
if source.missing_done == "retry":
raise TransientError(f"{source.name} SSE missing_done: 截断且无 [DONE]", **ctx)
return True
def _finalize_text(
self, content_parts: list[str], thinking_parts: list[str], profile: ProviderProfile
) -> tuple[str, str]:
content = "".join(content_parts)
thinking = "".join(thinking_parts)
if profile.strip_think_tags and "<think>" in content:
content, tag_thinking = _strip_think(content)
if tag_thinking:
thinking = tag_thinking
return content, thinking
async def _complete_once(
self, client: httpx.AsyncClient, url: str, payload: dict[str, Any],
source: SourceConfig, profile: ProviderProfile,
) -> TransportResult:
"""非流式快路径(三项目均无,库新增): 单 JSON 响应,仅 total 超时。"""
resp = await client.post(url, json=payload)
if resp.status_code != 200:
raise _status_to_error(source, resp.status_code, resp.text, resp.headers)
try:
body = resp.json()
except json.JSONDecodeError as exc:
raise TransientError(
f"{source.name} 非流式响应非法 JSON", source_name=source.name, operation="chat"
) from exc
choices = body.get("choices") or []
if not choices:
raise TransientError(
f"{source.name} 响应缺 choices", source_name=source.name, operation="chat"
)
message = choices[0].get("message") or {}
content, thinking = self._finalize_text(
[message.get("content") or ""], [message.get("reasoning_content") or ""], profile
)
prompt, completion, usage_source = _resolve_usage(body.get("usage") or {}, source)
return TransportResult(
content=content, thinking=thinking, prompt_tokens=prompt,
completion_tokens=completion, usage_source=usage_source,
ttft_ms=None, max_inter_token_ms=None, raw={"usage": body.get("usage")},
)
async def aclose(self) -> None:
"""幂等关闭全部源 client。"""
clients, self._clients = self._clients, {}
for client in clients.values():
await client.aclose()
+255
View File
@@ -0,0 +1,255 @@
"""OpenAICompatTransport 测试(M1 设计 §4.4/§6):SSE 解析、错误翻译、缺 DONE 语义。
SSE 帧样本按三项目真实网关响应形态二次构造(OpenAI 兼容 chunk 结构)。
"""
import json
import httpx
import pytest
from polygateway.errors import (
RequestRejectedError,
SourceDeadError,
TransientError,
)
from polygateway.transports.openai_compat import (
OpenAICompatTransport,
_iter_sse_deltas,
_sse_data_payload,
)
from polygateway.types import SourceConfig
def _source(**overrides):
base = dict(
name="qwen_1", provider="qwen", base_url="https://gw.example/v1",
api_key="sk-test", model="qwen-max", timeout_s=5.0,
)
base.update(overrides)
return SourceConfig(**base)
def _chunk(content=None, reasoning=None, usage=None):
delta = {}
if content is not None:
delta["content"] = content
if reasoning is not None:
delta["reasoning_content"] = reasoning
body = {"choices": [{"delta": delta}]} if (delta or usage is None) else {"choices": []}
if usage is not None:
body["usage"] = usage
return f"data: {json.dumps(body)}\n\n"
_USAGE = {"prompt_tokens": 11, "completion_tokens": 7}
def _sse_stream(*frames, done=True):
text = "".join(frames) + ("data: [DONE]\n\n" if done else "")
return httpx.Response(
200, content=text.encode(), headers={"content-type": "text/event-stream"}
)
def _transport_for(handler):
mock = httpx.MockTransport(handler)
return OpenAICompatTransport(
client_factory=lambda src: httpx.AsyncClient(base_url=src.base_url, transport=mock)
)
async def _complete(transport, source, *, stream=True, overlay=None):
return await transport.complete(
messages=[{"role": "user", "content": "hi"}], source=source,
stream=stream, overlay=overlay or {}, call_id="cid-1",
)
class TestSsePureFunctions:
def test_data_payload_filters_noise(self):
assert _sse_data_payload("") is None
assert _sse_data_payload(": ping") is None
assert _sse_data_payload("event: x") is None
assert _sse_data_payload("data: {}") == "{}"
assert _sse_data_payload("data: [DONE]") == "[DONE]"
async def test_iter_deltas_yields_and_flags_done(self):
async def lines():
for raw in [_chunk(content="he"), ": ping", _chunk(reasoning="think"),
_chunk(usage=_USAGE), "data: [DONE]"]:
for line in raw.splitlines():
yield line
sink = {}
out = [d async for d in _iter_sse_deltas(lines(), sink)]
assert out == [(True, "he"), (False, "think")]
assert sink["done"] is True and sink["usage"] == _USAGE
async def test_malformed_frame_raises_transient(self):
async def lines():
yield "data: {not-json"
with pytest.raises(TransientError, match="malformed"):
_ = [d async for d in _iter_sse_deltas(lines(), {})]
class TestStreamHappyPath:
async def test_full_stream_with_usage(self):
def handler(request):
return _sse_stream(_chunk(reasoning="ponder"), _chunk(content="hello"),
_chunk(content=" world"), _chunk(usage=_USAGE))
result = await _complete(_transport_for(handler), _source())
assert result.content == "hello world"
assert result.thinking == "ponder"
assert result.prompt_tokens == 11 and result.completion_tokens == 7
assert result.usage_source == "measured"
assert result.ttft_ms is not None and result.ttft_ms >= 0
async def test_qwen_think_tag_stripped(self):
def handler(request):
return _sse_stream(_chunk(content="<think>hmm</think>answer"), _chunk(usage=_USAGE))
result = await _complete(_transport_for(handler), _source())
assert result.content == "answer"
assert result.thinking == "hmm"
async def test_usage_missing_falls_back_to_est(self):
def handler(request):
return _sse_stream(_chunk(content="ok"))
result = await _complete(_transport_for(handler), _source(tpm=1000, est_tokens=333))
assert result.usage_source == "estimated"
assert result.prompt_tokens == 0 and result.completion_tokens == 333
class TestMissingDoneSemantics:
def _no_done_handler(self, request):
return _sse_stream(_chunk(content="partial"), _chunk(usage=_USAGE), done=False)
async def test_default_retry_policy_raises_transient(self):
with pytest.raises(TransientError, match="missing_done|truncated"):
await _complete(_transport_for(self._no_done_handler), _source())
async def test_salvage_policy_keeps_content_as_estimated(self):
result = await _complete(
_transport_for(self._no_done_handler), _source(missing_done="salvage")
)
assert result.content == "partial"
assert result.usage_source == "estimated" # 打捞路径强制 estimated
async def test_early_eof_always_transient_even_under_salvage(self):
def handler(request):
return _sse_stream(done=False) # 零内容提前断流
with pytest.raises(TransientError, match="early_eof"):
await _complete(_transport_for(handler), _source(missing_done="salvage"))
class TestNonStreamFastPath:
async def test_non_stream_parses_message(self):
def handler(request):
body = json.loads(request.content)
assert body.get("stream") is False and "stream_options" not in body
return httpx.Response(200, json={
"choices": [{"message": {"content": "42", "reasoning_content": "count"}}],
"usage": _USAGE,
})
result = await _complete(_transport_for(handler), _source(), stream=False)
assert result.content == "42" and result.thinking == "count"
assert result.usage_source == "measured"
assert result.ttft_ms is None
class TestRequestShaping:
@pytest.mark.parametrize(
("enable_thinking", "expected"),
[(True, {"enable_thinking": True}), (False, {"enable_thinking": False}), (None, {})],
)
async def test_thinking_tri_state_injection(self, enable_thinking, expected):
seen = {}
def handler(request):
seen.update(json.loads(request.content))
return _sse_stream(_chunk(content="x"), _chunk(usage=_USAGE))
await _complete(_transport_for(handler), _source(enable_thinking=enable_thinking))
assert {k: seen[k] for k in expected} == expected
if enable_thinking is None:
assert "enable_thinking" not in seen
assert seen["stream_options"] == {"include_usage": True}
async def test_overlay_merged_into_payload(self):
seen = {}
def handler(request):
seen.update(json.loads(request.content))
return _sse_stream(_chunk(content="x"), _chunk(usage=_USAGE))
await _complete(
_transport_for(handler), _source(),
overlay={"response_format": {"type": "json_object"}},
)
assert seen["response_format"] == {"type": "json_object"}
class TestErrorTranslation:
@pytest.mark.parametrize(
("status", "body", "headers", "exc", "match"),
[
(401, "{}", {}, SourceDeadError, None),
(403, "{}", {}, SourceDeadError, None),
(400, "{}", {}, RequestRejectedError, None),
(418, "{}", {}, RequestRejectedError, None),
(500, "{}", {}, TransientError, None),
(503, "{}", {}, TransientError, None),
(429, json.dumps({"error": {"type": "insufficient_quota"}}), {}, SourceDeadError, None),
(429, "{}", {"retry-after": "2.5"}, TransientError, None),
],
)
async def test_status_translation(self, status, body, headers, exc, match):
def handler(request):
return httpx.Response(status, content=body.encode(), headers=headers)
with pytest.raises(exc) as ei:
await _complete(_transport_for(handler), _source())
assert ei.value.status_code == status
assert ei.value.source_name == "qwen_1"
if status == 429 and exc is TransientError:
assert ei.value.retry_after_s == 2.5
async def test_retry_after_http_date_ignored(self):
def handler(request):
return httpx.Response(429, content=b"{}",
headers={"retry-after": "Wed, 21 Oct 2026 07:28:00 GMT"})
with pytest.raises(TransientError) as ei:
await _complete(_transport_for(handler), _source())
assert ei.value.retry_after_s is None # 仅支持秒数形态(CHS 同款)
async def test_httpx_timeout_becomes_transient(self):
def handler(request):
raise httpx.ConnectTimeout("boom")
with pytest.raises(TransientError):
await _complete(_transport_for(handler), _source())
async def test_httpx_transport_error_becomes_transient(self):
def handler(request):
raise httpx.RemoteProtocolError("broken pipe")
with pytest.raises(TransientError):
await _complete(_transport_for(handler), _source())
class TestLifecycle:
async def test_aclose_idempotent(self):
def handler(request):
return _sse_stream(_chunk(content="x"), _chunk(usage=_USAGE))
transport = _transport_for(handler)
await _complete(transport, _source())
await transport.aclose()
await transport.aclose()