fix: stabilize knowledge graph extraction for long documents

Long document imports could send oversized graph prompts, surface empty model responses without useful diagnostics, and fail the entire new document even when its full-text index was usable. Graph extraction now uses bounded token batches with retry and timeout handling, preserves successful partial results, and reports graph failures independently from new-document indexing. Chinese graph retrieval now recognizes bounded CJK query terms, while graph search keeps one-hop relationship context visible around matched entities. Release note: 修复长文档知识图谱抽取易因空响应或超时导致整篇文档导入失败的问题；现在图谱按 Token 分批重试，失败时仍保留新文档全文检索，并改进中文图谱问答和关系上下文展示。
2026-08-25 08:03:15 +08:00
parent 098f21b47a
commit 88658aceaa
12 changed files with 679 additions and 64 deletions
@@ -205,6 +205,7 @@
 - 全文完成但向量失败时，明确显示“全文可用、向量失败”
 - 向量未启用与向量失败是不同状态
 - 图谱按需、未启用和失败是不同状态
+- 图谱模型按有界 Token 批次抽取；部分批次失败时保留成功结果并在任务中显示具体警告，全部批次失败时新文档仍保留可用全文索引
 - 汇总显示全文可用数、向量完成数和失败数

 ### US-D2 修复失败文档