feat: 知识批量沉淀改为后台任务,避免 nginx 504

- /ingest/threads/batch 新增 background 参数(默认 true),立即返回 {queued:true}
- 抽取循环抽出 _run_batch_ingest,后台路径显式传入 created_by
- 前端知识库页与输入框、线程 hooks/types 配套调整
- config.yaml 上下文压缩测试值

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
吉鑫升andClaude Opus 4.8 committed 2026-06-16 21:07:17 +08:00
1 parent b380575f16
commit 342fa8265a
12 files changed
+170 -22

No files matched your search

@@ -65,6 +65,10 @@ class BatchIngestRequest(BaseModel):
include_sources: bool = True
skip_existing: bool = True
mode: str = "summary"
background: bool = Field(
default=True,
description="When true (default), the per-thread extraction loop runs in a background task and the endpoint returns immediately ({queued:true}). Batch capture of many threads is far too slow to finish inside one HTTP request — running it inline makes nginx time out (504), which the frontend surfaces as a generic 'request failed'.",
)
model_name: str | None = Field(default=None, description="可选:指定用于知识提炼的大模型名称;留空使用默认配置")
template_id: str | None = Field(default=None, description="可选:抽取提示词模板 id;留空使用对话默认模板")
folder: str | None = Field(default=None, description="可选:沉淀结果归入的目录(斜杠分隔路径)")
@@ -285,15 +289,14 @@ async def ingest_thread(thread_id: str, body: IngestThreadRequest, request: Requ
}
@router.post("/ingest/threads/batch")
@require_auth
async def ingest_threads_batch(body: BatchIngestRequest, request: Request) -> dict[str, Any]:
"""Batch-sediment the caller's recent threads into knowledge notes."""
service = get_knowledge_service(request)
checkpointer = get_checkpointer(request)
thread_store = get_thread_store(request)
async def _run_batch_ingest(service, checkpointer, threads: list[dict[str, Any]], body: BatchIngestRequest, created_by: str | None) -> dict[str, Any]:
"""Run the per-thread sediment loop and tally the outcome.
threads = await thread_store.search(limit=body.limit, offset=body.offset)
Shared by the inline and background paths; ``created_by`` is passed in
explicitly because the user contextvar is gone once a background task runs.
Each thread's transcript is truncated by the extractor (``max_input_chars``)
before the LLM call, so a single huge conversation can't blow the context.
"""
created = skipped = failed = 0
# Notes that landed via the rule-based fallback because the chosen model
# produced nothing usable (timeout / API error / non-JSON). Surfaced so the
@@ -319,7 +322,7 @@ async def ingest_threads_batch(body: BatchIngestRequest, request: Request) -> di
status=body.status,
include_sources=body.include_sources,
reference_batches=batches,
created_by=get_effective_user_id(),
created_by=created_by,
skip_existing=body.skip_existing,
model_name=body.model_name,
template_id=body.template_id,
@@ -347,6 +350,49 @@ async def ingest_threads_batch(body: BatchIngestRequest, request: Request) -> di
return {"ok": True, "processed": len(items), "created": created, "skipped": skipped, "failed": failed, "llm_fallback": llm_fallback, "items": items}
async def _ingest_threads_batch_background(service, checkpointer, threads: list[dict[str, Any]], body: BatchIngestRequest, created_by: str | None) -> None:
"""Background variant of the batch loop (failures swallowed/logged)."""
try:
result = await _run_batch_ingest(service, checkpointer, threads, body, created_by)
logger.info(
"Background batch ingest finished: processed=%d created=%d skipped=%d failed=%d llm_fallback=%d",
result["processed"], result["created"], result["skipped"], result["failed"], result["llm_fallback"],
)
except Exception:
logger.exception("Background batch ingest failed")
@router.post("/ingest/threads/batch")
@require_auth
async def ingest_threads_batch(body: BatchIngestRequest, request: Request, background_tasks: BackgroundTasks) -> dict[str, Any]:
"""Batch-sediment the caller's recent threads into knowledge notes.
Defaults to ``background=true``: the (slow) per-thread extraction runs in a
background task and the endpoint returns ``{"queued": true, "processed": N}``
immediately, so the request never outlives the nginx gateway timeout. Pass
``background=false`` to block until every thread is processed (small batches /
scripts that want the full per-thread tally).
"""
service = get_knowledge_service(request)
checkpointer = get_checkpointer(request)
thread_store = get_thread_store(request)
threads = await thread_store.search(limit=body.limit, offset=body.offset)
if body.background:
# Capture the user id now — the contextvar is gone once the task runs.
background_tasks.add_task(
_ingest_threads_batch_background,
service,
checkpointer,
threads,
body,
get_effective_user_id(),
)
return {"ok": True, "queued": True, "processed": len(threads), "created": 0, "skipped": 0, "failed": 0, "llm_fallback": 0, "items": []}
return await _run_batch_ingest(service, checkpointer, threads, body, get_effective_user_id())
async def _extract_upload_text(file: UploadFile) -> tuple[str, str]:
"""Read an uploaded file and return ``(filename, markdown_text)``.
@@ -476,7 +522,7 @@ async def list_notes(
offset: int = 0,
) -> dict[str, Any]:
service = get_knowledge_service(request)
limit = max(1, min(limit, 100))
limit = max(1, min(limit, 1000))
offset = max(0, offset)
return await service.list_notes(keyword=keyword, tag=tag, entity=entity, source_type=source_type, status=status, folder=folder, limit=limit, offset=offset)
@@ -243,6 +243,9 @@ _CONTEXT_CONFIGURABLE_KEYS: frozenset[str] = frozenset(
# 关掉个人记忆召回 / builtin 记忆注入(见 multi_agent._run_payload)。
"memory_recall_disabled",
"memory_injection_enabled",
# 上下文压缩开关:由聊天框下方的按钮控制(默认关闭),透传到 run config 后
# 在 _build_middlewares 里决定是否挂载 SummarizationMiddleware。
"summarization_enabled",
}
)