zhizhi/server/age-os/mcp-server
copilot-swe-agent[bot] ba4c233509
fix: COS训练Agent大文件崩溃 — 添加Buffer安全处理、HEAD预检、分块提取策略
修复两个训练失败根因:
1. conversations.json (665MB) 超过Node.js字符串限制导致ERR_STRING_TOO_LONG崩溃
2. 单文件崩溃导致整个训练管线中断,后续文件无法处理

- cos.js: 添加readBuffer/head/readPartial方法,cosRequest增加try/catch防止未捕获异常
- corpus-extractor-ops.js: 超大文件(>200MB)采用分块采样策略,ZIP文件提前检测避免无效读取
- cos-training-trigger.js: 增加文件大小预警日志,分类计数(成功/跳过/失败)

Agent-Logs-Url: https://github.com/qinfendebingshuo/guanghulab/sessions/f10f1b02-ed4e-4ee8-ac27-0580c1cf8a7e

Co-authored-by: qinfendebingshuo <207279273+qinfendebingshuo@users.noreply.github.com>
2026-04-11 00:55:39 +00:00
..
tools fix: COS训练Agent大文件崩溃 — 添加Buffer安全处理、HEAD预检、分块提取策略 2026-04-11 00:55:39 +00:00
cos-watcher-state.json feat: COS桶轮询守护进程 · SCF事件替代方案 (cos-watcher.js) 2026-04-10 12:43:16 +00:00
cos-watcher.js fix: address code review feedback on cos-watcher 2026-04-10 12:46:52 +00:00
cos.js fix: COS训练Agent大文件崩溃 — 添加Buffer安全处理、HEAD预检、分块提取策略 2026-04-11 00:55:39 +00:00
db.js 修复代码审查问题: SQL参数化/关系查询逻辑/数据库密码校验 2026-04-03 07:36:33 +00:00
github-client.js Fix: GitHub path encoding, MCP gateway write-tool security, content preview constant 2026-04-04 15:24:09 +00:00
notion-client.js D51 S6+S7: Notion client, GitHub client, 11 new MCP tools, 3800→3100 gateway 2026-04-04 15:21:18 +00:00
server.js fix: address code review feedback on cos-watcher 2026-04-10 12:46:52 +00:00