8 Commits
Author SHA1 Message Date
QihuAI Design 9a43ee244d fix(im_llm): 代理绕过Docker内部hostname 2026-07-31 08:51:15 +08:00
QihuAI Design a2504bce34 fix(chat): resolve Auto-route failover event-loop freeze (#812)
Root cause: ensure_structured_column() executed a synchronous blocking
engine.connect().execute(SELECT information_schema.columns ...) inside the
async request path (_build_chat_history / _maybe_compress_history), freezing
the uvicorn event loop and hanging all endpoints (~200s, recovered only by
docker restart).

Fix:
- chat_context.py: remove per-request ensure_structured_column() calls; run
  once idempotently at app startup (main.py lifespan). Use CURRENT_DATABASE()
  for PostgreSQL compatibility.
- main.py: add startup-only ensure_structured_column() hook (try/except).
- llm_client.py: trust_env=False + bounded httpx.Timeout/wait_for so broken
  endpoints fail fast instead of hanging the event loop.
- chat_service.py: run media generation in a background daemon thread with its
  own event loop (_launch_media_task_bg) so SSE end does not cancel image/video
  tasks (#811); remove TEMP faulthandler diagnostic.

Safe-align: persist accumulated live container changes (admin/*, knowledge,
market, models, config, observability, mcp_client, platform_knowledge,
im_bot_platform, chat_media, capability_registry, requirements) and add
workflow_templates.

Verified: live failover test (model-006 -> conn-refused) returns in ~1.9s with
concurrent login 0.83s/200; event-loop responsiveness restored (sub-second).
2026-07-26 00:35:35 +08:00
QihuAI Design d3d8d62b94 fix(media): repair JSON_CONTAINS auto-routing + reconcile stale backend tests
CI / gate (push) Has been cancelled
Real bug:
- capability_registry.resolve_auto_model and chat_media._build_media_route used
  `Column.op("JSON_CONTAINS")(col, val)`, but SQLAlchemy's op() accepts only ONE
  argument -> TypeError on every image/video/audio auto-route, silently failing
  media generation (caught as "后台生成异常"). Replaced with func.json_contains().

Test reconciliation (no production behavior change):
- tests/test_context_engine.py: tiktoken upgrade made the hardcoded len//2 token
  estimates wrong; relax to tokenizer-agnostic assertions and add
  structured_memory=None to the _Summary mock (code now branches on it).
- tests/test_regression_r1r2r3.py: R3 FakeModel lacked `name` (code reads m.name);
  also the assertion assumed a literal '超时' string, but the failover architecture
  catches model-level exceptions and writes a user-understandable error marker
  instead. Align the assertion to the real contract: a media failure/timeout must
  be written back as a user-understandable error message, never left as '生成中…'.

CI gate: 28 backend pytest + 15 frontend vitest all green.
2026-07-23 09:16:31 +08:00
QihuAI Design a08ec87b22 chore(align): sync Git with running container as authoritative source
One-time alignment to close drift between the running container
(qihu-aiflow / image qihuaiflow:with-deps) and this Git repository.
Container hotfixes previously persisted only via `docker commit` are now
captured in Git so Git becomes the single source of truth.

Backend:
- chat_service.py: accumulated fixes incl. EigenFlux README guidance
- chat_context.py / chat_media.py: URL/web-repo context injection
- api/url_reader.py: NEW SSRF-safe web/README reader (Issue #725)
- skills_admin.py & Skills.tsx: 14-scene capability unification
- pgvector_search.py, orchestrator.py, platform_knowledge.py,
  market.py, admin/models.py: drift sync

Frontend:
- ChatPage.tsx: collapsible AI think/act block (Issue #724)
- MarketPage / EvolutionDashboard / TokenDashboard styling sync

Migrations:
- 018_normalize_marketapp_scene.sql: fixed capability allowlist
  (prevents capability wipe on restart)

Alignment performed 2026-07-23 05:42:47. No deletions; secrets/venv/node_modules/
caches were not touched (tracked-source-only copy).
2026-07-23 05:42:48 +08:00
QihuAI Design 0b00d6da45 fix: 文生图连接测试 size 1024x1024 + 媒体端点智能推导 + 图像/视频模型 failover 路由(默认模型第一顺位) 2026-07-20 22:11:53 +08:00
QihuAI Design 5def2697de fix(chat): image send 500 + table/newline formatting (v2026.07.20.1708)
CI / gate (push) Has been cancelled
2026-07-20 17:08:20 +08:00
QihuAI Design a7a57f9a03 refactor(observability): Phase2.2 成本归因 trace_id列+token日志写入+LLM/MCP/媒体子span [v2026.07.19.3]
CI / gate (push) Has been cancelled
2026-07-19 08:51:28 +08:00
QihuAI Design 1b6a184d27 refactor(chat): split 10070-line monolith into 8 service submodules (Phase1) [v2026.07.19.1]
CI / gate (push) Has been cancelled
2026-07-19 00:59:03 +08:00