feat(translate): language detection, async HTTP LLM, history model, agent improvements

- Add async HTTP-based LLM transport (_llm_async_http.py)
- Add orthogonal LLM call tests
- Improve language detection (_lang_detect.py) and batch insert
- Update translate schemas, service utils, preview constants/prompts
- Add TranslateHistoryModel with pagination and filtering
- Update agent confirmation, persistence, langgraph setup, run, tools
- Improve LLM health checking in shared module
- Update translate runs API, history route
This commit is contained in:
2026-07-08 19:35:49 +03:00
parent 97a64ce056
commit c488a63dc0
30 changed files with 1148 additions and 550 deletions

View File

@@ -17,9 +17,9 @@ from unittest.mock import AsyncMock, MagicMock, patch
import httpx
from src.plugins.translate._llm_async_http import (
_get_verify,
_get_http_client,
from ss_tools.shared.ssl import httpx_verify, system_ssl_context
from ss_tools.shared._llm_http import get_shared_http_client
from ss_tools.shared._llm_http import (
call_openai_compatible,
_do_http_request,
_handle_response_format_fallback,
@@ -27,34 +27,37 @@ from src.plugins.translate._llm_async_http import (
class TestGetVerify:
"""_get_verify — SSL verification context."""
"""SSL verification context (now from shared module)."""
def test_returns_ssl_context(self):
"""Always returns SSLContext (no env-based disable)."""
result = _get_verify()
"""httpx_verify returns SSLContext (no env-based disable)."""
result = httpx_verify()
assert isinstance(result, ssl.SSLContext)
def test_system_context_not_false(self):
"""system_ssl_context never returns False."""
result = system_ssl_context()
assert result is not False
def test_ignores_llm_ssl_verify_env(self):
"""LLM_SSL_VERIFY env is no longer read — central ssl helper is used."""
with patch.dict(os.environ, {"LLM_SSL_VERIFY": "false"}):
result = _get_verify()
result = system_ssl_context()
assert isinstance(result, ssl.SSLContext), "must not return False"
class TestGetHttpClient:
"""_get_http_client — module-level singleton."""
"""get_shared_http_client — module-level singleton."""
@pytest.mark.asyncio
async def test_creates_client(self):
def test_creates_client(self):
"""Creates client on first call."""
client = await _get_http_client()
client = get_shared_http_client(timeout=180)
assert isinstance(client, httpx.AsyncClient)
@pytest.mark.asyncio
async def test_reuses_client(self):
def test_reuses_client(self):
"""Returns same client on second call."""
client1 = await _get_http_client()
client2 = await _get_http_client()
client1 = get_shared_http_client(timeout=180)
client2 = get_shared_http_client(timeout=180)
assert client1 is client2
@@ -84,8 +87,8 @@ class TestCallOpenaiCompatible:
}
mock_response.text = '{"choices": [{"finish_reason": "stop", "message": {"content": "Hello, world!"}}]}'
with patch("src.plugins.translate._llm_async_http._do_http_request", AsyncMock(return_value=(mock_response, mock_response.text))):
with patch("src.plugins.translate._llm_async_http._handle_response_format_fallback", AsyncMock()):
with patch("ss_tools.shared._llm_http._do_http_request", AsyncMock(return_value=(mock_response, mock_response.text))):
with patch("ss_tools.shared._llm_http._handle_response_format_fallback", AsyncMock()):
content, finish_reason = await call_openai_compatible(
"https://api.openai.com",
"sk-test",
@@ -104,8 +107,8 @@ class TestCallOpenaiCompatible:
mock_response.json.return_value = {"choices": []}
mock_response.text = '{"choices": []}'
with patch("src.plugins.translate._llm_async_http._do_http_request", AsyncMock(return_value=(mock_response, mock_response.text))):
with patch("src.plugins.translate._llm_async_http._handle_response_format_fallback", AsyncMock()):
with patch("ss_tools.shared._llm_http._do_http_request", AsyncMock(return_value=(mock_response, mock_response.text))):
with patch("ss_tools.shared._llm_http._handle_response_format_fallback", AsyncMock()):
with pytest.raises(ValueError, match="no choices"):
await call_openai_compatible(
"https://api.openai.com",
@@ -130,8 +133,8 @@ class TestCallOpenaiCompatible:
}
mock_response.text = '{"choices": [{"finish_reason": "stop", "message": {"content": ""}}]}'
with patch("src.plugins.translate._llm_async_http._do_http_request", AsyncMock(return_value=(mock_response, mock_response.text))):
with patch("src.plugins.translate._llm_async_http._handle_response_format_fallback", AsyncMock()):
with patch("ss_tools.shared._llm_http._do_http_request", AsyncMock(return_value=(mock_response, mock_response.text))):
with patch("ss_tools.shared._llm_http._handle_response_format_fallback", AsyncMock()):
with pytest.raises(ValueError, match="empty content"):
await call_openai_compatible(
"https://api.openai.com",
@@ -151,8 +154,8 @@ class TestCallOpenaiCompatible:
}
mock_response.text = '{"choices": [null]}'
with patch("src.plugins.translate._llm_async_http._do_http_request", AsyncMock(return_value=(mock_response, mock_response.text))):
with patch("src.plugins.translate._llm_async_http._handle_response_format_fallback", AsyncMock()):
with patch("ss_tools.shared._llm_http._do_http_request", AsyncMock(return_value=(mock_response, mock_response.text))):
with patch("ss_tools.shared._llm_http._handle_response_format_fallback", AsyncMock()):
with pytest.raises(ValueError, match="LLM response processing failed"):
await call_openai_compatible(
"https://api.openai.com",
@@ -177,8 +180,8 @@ class TestCallOpenaiCompatible:
}
mock_response.text = '{"choices": [{"finish_reason": "stop", "message": {"refusal": "I cannot answer that", "content": null}}]}'
with patch("src.plugins.translate._llm_async_http._do_http_request", AsyncMock(return_value=(mock_response, mock_response.text))):
with patch("src.plugins.translate._llm_async_http._handle_response_format_fallback", AsyncMock()):
with patch("ss_tools.shared._llm_http._do_http_request", AsyncMock(return_value=(mock_response, mock_response.text))):
with patch("ss_tools.shared._llm_http._handle_response_format_fallback", AsyncMock()):
with pytest.raises(ValueError, match="refused"):
await call_openai_compatible(
"https://api.openai.com",
@@ -200,8 +203,8 @@ class TestCallOpenaiCompatible:
)
mock_response.text = "Internal Server Error"
with patch("src.plugins.translate._llm_async_http._do_http_request", AsyncMock(return_value=(mock_response, mock_response.text))):
with patch("src.plugins.translate._llm_async_http._handle_response_format_fallback", AsyncMock()):
with patch("ss_tools.shared._llm_http._do_http_request", AsyncMock(return_value=(mock_response, mock_response.text))):
with patch("ss_tools.shared._llm_http._handle_response_format_fallback", AsyncMock()):
with pytest.raises(httpx.HTTPStatusError):
await call_openai_compatible(
"https://api.openai.com",
@@ -222,8 +225,8 @@ class TestCallOpenaiCompatible:
}
mock_response.text = '{"choices": [{"finish_reason": "stop", "message": {"content": "hi"}}]}'
with patch("src.plugins.translate._llm_async_http._do_http_request", AsyncMock(return_value=(mock_response, mock_response.text))):
with patch("src.plugins.translate._llm_async_http._handle_response_format_fallback", AsyncMock()):
with patch("ss_tools.shared._llm_http._do_http_request", AsyncMock(return_value=(mock_response, mock_response.text))):
with patch("ss_tools.shared._llm_http._handle_response_format_fallback", AsyncMock()):
# disable_reasoning=False, so response_format should be set
content, _ = await call_openai_compatible(
"https://api.openai.com",
@@ -247,8 +250,8 @@ class TestCallOpenaiCompatible:
}
mock_response.text = '{"choices": [{"finish_reason": "stop", "message": {"content": "hi"}}]}'
with patch("src.plugins.translate._llm_async_http._do_http_request", AsyncMock(return_value=(mock_response, mock_response.text))):
with patch("src.plugins.translate._llm_async_http._handle_response_format_fallback", AsyncMock()):
with patch("ss_tools.shared._llm_http._do_http_request", AsyncMock(return_value=(mock_response, mock_response.text))):
with patch("ss_tools.shared._llm_http._handle_response_format_fallback", AsyncMock()):
content, _ = await call_openai_compatible(
"https://api.openai.com",
"sk-test",
@@ -262,28 +265,29 @@ class TestCallOpenaiCompatible:
class TestDoHttpRequest:
"""_do_http_request — async HTTP POST with rate-limit retry."""
@pytest.fixture
def mock_client(self):
return AsyncMock(spec=httpx.AsyncClient)
@pytest.mark.asyncio
async def test_success(self):
async def test_success(self, mock_client):
"""Single successful request."""
mock_response = MagicMock(spec=httpx.Response)
mock_response.status_code = 200
mock_response.text = "OK"
mock_client.post.return_value = mock_response
with patch("src.plugins.translate._llm_async_http._get_http_client", AsyncMock()) as mock_get:
mock_client = AsyncMock()
mock_get.return_value = mock_client
mock_client.post.return_value = mock_response
response, text = await _do_http_request(
"https://api.openai.com/chat/completions",
{"Authorization": "Bearer test"},
{"model": "gpt-4o"},
)
assert response.status_code == 200
assert text == "OK"
response, text = await _do_http_request(
mock_client,
"https://api.openai.com/chat/completions",
{"Authorization": "Bearer test"},
{"model": "gpt-4o"},
)
assert response.status_code == 200
assert text == "OK"
@pytest.mark.asyncio
async def test_rate_limit_retry(self):
async def test_rate_limit_retry(self, mock_client):
"""429 triggers retry."""
mock_429 = MagicMock(spec=httpx.Response)
mock_429.status_code = 429
@@ -294,23 +298,21 @@ class TestDoHttpRequest:
mock_200.status_code = 200
mock_200.text = "OK"
with patch("src.plugins.translate._llm_async_http._get_http_client", AsyncMock()) as mock_get:
mock_client = AsyncMock()
mock_get.return_value = mock_client
mock_client.post.side_effect = [mock_429, mock_200]
mock_client.post.side_effect = [mock_429, mock_200]
with patch("src.plugins.translate._llm_async_http.asyncio.sleep", AsyncMock()):
response, text = await _do_http_request(
"https://api.openai.com/chat/completions",
{"Authorization": "Bearer test"},
{"model": "gpt-4o"},
)
assert response.status_code == 200
assert text == "OK"
assert mock_client.post.call_count == 2
with patch("ss_tools.shared._llm_http.asyncio.sleep", AsyncMock()):
response, text = await _do_http_request(
mock_client,
"https://api.openai.com/chat/completions",
{"Authorization": "Bearer test"},
{"model": "gpt-4o"},
)
assert response.status_code == 200
assert text == "OK"
assert mock_client.post.call_count == 2
@pytest.mark.asyncio
async def test_rate_limit_with_retry_after(self):
async def test_rate_limit_with_retry_after(self, mock_client):
"""Retry-After header is respected."""
mock_429 = MagicMock(spec=httpx.Response)
mock_429.status_code = 429
@@ -321,46 +323,41 @@ class TestDoHttpRequest:
mock_200.status_code = 200
mock_200.text = "OK"
with patch("src.plugins.translate._llm_async_http._get_http_client", AsyncMock()) as mock_get:
mock_client = AsyncMock()
mock_get.return_value = mock_client
mock_client.post.side_effect = [mock_429, mock_200]
mock_client.post.side_effect = [mock_429, mock_200]
with patch("src.plugins.translate._llm_async_http.asyncio.sleep", AsyncMock()) as mock_sleep:
response, _ = await _do_http_request(
"https://api.openai.com/chat/completions",
{"Authorization": "Bearer test"},
{"model": "gpt-4o"},
)
assert response.status_code == 200
# Verify sleep was called with retry-after value
mock_sleep.assert_called_once_with(2)
with patch("ss_tools.shared._llm_http.asyncio.sleep", AsyncMock()) as mock_sleep:
response, _ = await _do_http_request(
mock_client,
"https://api.openai.com/chat/completions",
{"Authorization": "Bearer test"},
{"model": "gpt-4o"},
)
assert response.status_code == 200
mock_sleep.assert_called_once_with(2)
@pytest.mark.asyncio
async def test_rate_limit_exhausted(self):
async def test_rate_limit_exhausted(self, mock_client):
"""All 429 retries exhausted, returns last 429."""
mock_429 = MagicMock(spec=httpx.Response)
mock_429.status_code = 429
mock_429.headers = {}
mock_429.text = "Rate limited"
with patch("src.plugins.translate._llm_async_http._get_http_client", AsyncMock()) as mock_get:
mock_client = AsyncMock()
mock_get.return_value = mock_client
mock_client.post.return_value = mock_429 # Always 429
mock_client.post.return_value = mock_429 # Always 429
with patch("src.plugins.translate._llm_async_http.asyncio.sleep", AsyncMock()):
response, _ = await _do_http_request(
"https://api.openai.com/chat/completions",
{"Authorization": "Bearer test"},
{"model": "gpt-4o"},
)
assert response.status_code == 429
assert mock_client.post.call_count == 3 # max 3 retries
with patch("ss_tools.shared._llm_http.asyncio.sleep", AsyncMock()):
response, _ = await _do_http_request(
mock_client,
"https://api.openai.com/chat/completions",
{"Authorization": "Bearer test"},
{"model": "gpt-4o"},
)
assert response.status_code == 429
assert mock_client.post.call_count == 3 # max 3 retries
@pytest.mark.asyncio
async def test_rate_limit_invalid_retry_after(self):
"""Retry-After with non-int value falls back to exponential backoff (lines 215-216)."""
async def test_rate_limit_invalid_retry_after(self, mock_client):
"""Retry-After with non-int value falls back to exponential backoff."""
mock_429 = MagicMock(spec=httpx.Response)
mock_429.status_code = 429
mock_429.headers = {"Retry-After": "abc"}
@@ -370,27 +367,29 @@ class TestDoHttpRequest:
mock_200.status_code = 200
mock_200.text = "OK"
with patch("src.plugins.translate._llm_async_http._get_http_client", AsyncMock()) as mock_get:
mock_client = AsyncMock()
mock_get.return_value = mock_client
mock_client.post.side_effect = [mock_429, mock_200]
mock_client.post.side_effect = [mock_429, mock_200]
with patch("src.plugins.translate._llm_async_http.asyncio.sleep", AsyncMock()) as mock_sleep:
response, _ = await _do_http_request(
"https://api.openai.com/chat/completions",
{"Authorization": "Bearer test"},
{"model": "gpt-4o"},
)
assert response.status_code == 200
# Should fall back to exponential backoff: 2^1 = 2
mock_sleep.assert_called_once_with(2)
with patch("ss_tools.shared._llm_http.asyncio.sleep", AsyncMock()) as mock_sleep:
response, _ = await _do_http_request(
mock_client,
"https://api.openai.com/chat/completions",
{"Authorization": "Bearer test"},
{"model": "gpt-4o"},
)
assert response.status_code == 200
# Should fall back to exponential backoff: 2^1 = 2
mock_sleep.assert_called_once_with(2)
class TestHandleResponseFormatFallback:
"""_handle_response_format_fallback — structured output fallback."""
@pytest.fixture
def mock_client(self):
return AsyncMock(spec=httpx.AsyncClient)
@pytest.mark.asyncio
async def test_400_response_format_error(self):
async def test_400_response_format_error(self, mock_client):
"""400 error with response_format pattern triggers retry."""
mock_400 = MagicMock(spec=httpx.Response)
mock_400.status_code = 400
@@ -404,27 +403,25 @@ class TestHandleResponseFormatFallback:
mock_200.encoding = "utf-8"
mock_200.headers = {}
mock_client.post.return_value = mock_200
payload = {"model": "gpt-4o", "response_format": {"type": "json_object"}}
with patch("src.plugins.translate._llm_async_http._get_http_client", AsyncMock()) as mock_get:
mock_client = AsyncMock()
mock_get.return_value = mock_client
mock_client.post.return_value = mock_200
await _handle_response_format_fallback(
mock_400,
mock_400.text,
payload,
"https://api.openai.com",
{},
)
# Verify response_format was popped
assert "response_format" not in payload
# Verify status_code was updated
assert mock_400.status_code == 200
await _handle_response_format_fallback(
mock_client,
mock_400,
mock_400.text,
payload,
"https://api.openai.com",
{},
)
# Verify response_format was popped
assert "response_format" not in payload
# Verify status_code was updated
assert mock_400.status_code == 200
@pytest.mark.asyncio
async def test_non_400_noop(self):
async def test_non_400_noop(self, mock_client):
"""Non-400 error does nothing."""
mock_500 = MagicMock(spec=httpx.Response)
mock_500.status_code = 500
@@ -433,24 +430,21 @@ class TestHandleResponseFormatFallback:
payload = {"model": "gpt-4o", "response_format": {"type": "json_object"}}
with patch("src.plugins.translate._llm_async_http._get_http_client", AsyncMock()) as mock_get:
mock_client = AsyncMock()
mock_get.return_value = mock_client
await _handle_response_format_fallback(
mock_500,
mock_500.text,
payload,
"https://api.openai.com",
{},
)
# Payload unchanged
assert "response_format" in payload
# No retry call
mock_client.post.assert_not_called()
await _handle_response_format_fallback(
mock_client,
mock_500,
mock_500.text,
payload,
"https://api.openai.com",
{},
)
# Payload unchanged
assert "response_format" in payload
# No retry call
mock_client.post.assert_not_called()
@pytest.mark.asyncio
async def test_400_no_pattern_noop(self):
async def test_400_no_pattern_noop(self, mock_client):
"""400 error without response_format pattern does nothing."""
mock_400 = MagicMock(spec=httpx.Response)
mock_400.status_code = 400
@@ -459,19 +453,16 @@ class TestHandleResponseFormatFallback:
payload = {"model": "gpt-4o", "response_format": {"type": "json_object"}}
with patch("src.plugins.translate._llm_async_http._get_http_client", AsyncMock()) as mock_get:
mock_client = AsyncMock()
mock_get.return_value = mock_client
await _handle_response_format_fallback(
mock_400,
mock_400.text,
payload,
"https://api.openai.com",
{},
)
assert "response_format" in payload
mock_client.post.assert_not_called()
await _handle_response_format_fallback(
mock_client,
mock_400,
mock_400.text,
payload,
"https://api.openai.com",
{},
)
assert "response_format" in payload
mock_client.post.assert_not_called()
# #endregion Test.LLMAsyncHttpClient