{"id":113411,"date":"2026-10-06T09:58:09","date_gmt":"2026-10-06T01:58:09","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/113411.html"},"modified":"2026-10-06T09:58:09","modified_gmt":"2026-10-06T01:58:09","slug":"%e7%ac%ac40%e7%ab%a0%ef%bc%9a%e3%80%90%e9%ab%98%e7%ba%a7%e7%af%87%e7%bb%bc%e5%90%88%e5%ae%9e%e6%88%98%e3%80%91%e4%bb%8e%e6%ba%90%e7%a0%81%e5%88%b0%e7%94%9f%e4%ba%a7%e6%89%93%e9%80%a0%e5%8f%af%e8%a7%82","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/113411.html","title":{"rendered":"\u7b2c40\u7ae0\uff1a\u3010\u9ad8\u7ea7\u7bc7\u7efc\u5408\u5b9e\u6218\u3011\u4ece\u6e90\u7801\u5230\u751f\u4ea7\u6253\u9020\u53ef\u89c2\u6d4b\u9ad8\u53ef\u7528 RAGFlow \u5e73\u53f0"},"content":{"rendered":"

1 \u9879\u76ee\u80cc\u666f<\/h3>\n

\u4e1a\u52a1\u573a\u666f<\/h4>\n

\u300c\u4e91\u5e06\u79d1\u6280\u300d\u7684 RAGFlow \u5e73\u53f0\u7ecf\u8fc7\u4e00\u5e74\u7684\u8fd0\u8425，\u5df2\u4ece\u5355\u673a\u623f Docker Compose \u6f14\u8fdb\u5230\u591a\u673a\u623f K8s \u96c6\u7fa4，\u670d\u52a1 20 \u4e2a\u4e1a\u52a1\u90e8\u95e8\u30012000 \u540d\u5458\u5de5\u300120 \u4e07\u4efd\u6587\u6863\u30011000 \u4e07 Chunk\u3002\u968f\u7740\u89c4\u6a21\u589e\u957f\u548c\u67b6\u6784\u590d\u6742\u5ea6\u63d0\u5347，\u56e2\u961f\u9762\u4e34\u4e00\u7cfb\u5217"\u5e73\u53f0\u7ea7"\u6311\u6218：<\/p>\n

  • \u767e\u4e07\u7ea7\u6587\u6863\u7684\u68c0\u7d22\u5ef6\u8fdf\u4ece 500ms \u6da8\u5230 3 \u79d2\u2014\u2014\u9700\u8981\u5bf9 Dealer \u68c0\u7d22\u94fe\u8def\u505a\u6e90\u7801\u7ea7\u4f18\u5316\u3002<\/li>\n
  • \u81ea\u5b9a\u4e49\u6269\u5c55\u6563\u843d\u5404\u5904：5 \u4e2a\u90e8\u95e8\u7684 WPS Parser\u3001Jira Tool\u3001\u5ba1\u6279\u7ec4\u4ef6\u7f3a\u4e4f\u7edf\u4e00\u7ba1\u7406\u2014\u2014\u5347\u7ea7 RAGFlow \u65f6\u517c\u5bb9\u6027\u672a\u77e5\u3002<\/li>\n
  • \u8de8\u673a\u623f\u90e8\u7f72\u7684\u6570\u636e\u540c\u6b65：\u5317\u4eac\u548c\u4e0a\u6d77\u4e24\u4e2a\u673a\u623f，\u7528\u6237\u5c31\u8fd1\u8bbf\u95ee\u4f46\u6570\u636e\u9700\u8981\u4fdd\u6301\u4e00\u81f4\u3002<\/li>\n
  • SRE \u843d\u5730：\u9700\u8981 SLO \u5b9a\u4e49\u3001\u5bb9\u91cf\u89c4\u5212\u3001\u7070\u5ea6\u53d1\u5e03\u3001\u5907\u4efd\u6062\u590d\u3001\u6210\u672c\u76d1\u63a7\u3001\u4e8b\u6545\u6f14\u7ec3\u2014\u2014\u5168\u5957\u8fd0\u7ef4\u4f53\u7cfb\u3002<\/li>\n

    CTO \u8981\u6c42：\u5c06\u8fc7\u53bb 39 \u7ae0\u7684\u77e5\u8bc6\u878d\u4f1a\u8d2f\u901a，\u4ea4\u4ed8\u4e00\u4e2a"\u4ece\u6e90\u7801\u5230\u751f\u4ea7"\u7684\u5b8c\u6574\u5e73\u53f0\u65b9\u6848，\u5305\u542b\u67b6\u6784\u56fe\u3001\u6e90\u7801\u6269\u5c55\u3001\u90e8\u7f72\u6e05\u5355\u3001\u76d1\u63a7\u5927\u76d8\u3001\u538b\u6d4b\u62a5\u544a\u548c\u4e8b\u6545\u9884\u6848\u3002<\/p>\n

    \u75db\u70b9<\/h4>\n

    \u5e73\u53f0\u7ea7\u4ea4\u4ed8 vs \u5355\u70b9\u4f18\u5316\u7684\u6838\u5fc3\u5dee\u5f02：<\/p>\n

  • \u5355\u70b9\u4f18\u5316 vs \u5168\u94fe\u8def\u6700\u4f18：\u4f18\u5316\u4e86\u68c0\u7d22\u4f46\u5ffd\u7565\u4e86 LLM \u8c03\u7528，\u4f53\u9a8c\u6539\u5584\u6709\u9650\u3002<\/li>\n
  • \u4e00\u6b21\u6027\u90e8\u7f72 vs \u6301\u7eed\u6f14\u8fdb：\u5e73\u53f0\u4e0d\u662f"\u642d\u597d\u5c31\u5b8c\u4e86"\u2014\u2014\u9700\u8981\u7070\u5ea6\u53d1\u5e03\u3001\u56de\u6eda\u3001A\/B \u6d4b\u8bd5\u80fd\u529b\u3002<\/li>\n
  • \u4e2a\u4eba\u82f1\u96c4 vs \u56e2\u961f SOP：\u51fa\u95ee\u9898\u4e0d\u80fd\u518d\u9760"\u5c0f\u674e\u770b\u4e00\u773c"\u2014\u2014\u9700\u8981\u76d1\u63a7\u3001\u544a\u8b66\u3001Runbook\u3002<\/li>\n
  • \u529f\u80fd\u53ef\u7528 vs \u54c1\u8d28\u53ef\u91cf\u5316：\u9700\u8981 SLO（P95<8s, \u53ef\u7528\u7387>99.9%），\u800c\u4e0d\u53ea\u662f"\u611f\u89c9\u8fd8\u884c"\u3002<\/li>\n

    2 \u9879\u76ee\u8bbe\u8ba1<\/h3>\n

    \u5c0f\u80d6：（\u628a 39 \u7ae0\u7684\u7b14\u8bb0\u94fa\u4e86\u4e00\u684c\u5b50）\u201c\u5927\u5e08！39 \u7ae0\u5b66\u5b8c\u4e86，\u6bcf\u4e00\u7ae0\u7684\u77e5\u8bc6\u70b9\u6211\u90fd\u7406\u89e3，\u4f46\u600e\u4e48\u628a\u8fd9\u4e9b\u4e1c\u897f\u4e32\u6210\u4e00\u4e2a\u751f\u4ea7\u7ea7\u5e73\u53f0？\u6211\u611f\u89c9\u81ea\u5df1\u50cf\u6512\u4e86\u4e00\u5806\u4e50\u9ad8\u79ef\u6728，\u4f46\u4e0d\u77e5\u9053\u600e\u4e48\u62fc\u6210\u4e00\u5ea7\u57ce\u5821\u3002\u201d<\/p>\n

    \u5927\u5e08：\u201c\u8fd9\u5c31\u662f\u9ad8\u7ea7\u7bc7\u7efc\u5408\u5b9e\u6218\u7684\u610f\u4e49\u2014\u2014\u628a\u524d\u9762\u6240\u6709\u7684\u77e5\u8bc6\u70b9\u7f16\u7ec7\u6210\u4e00\u5f20\u5b8c\u6574\u7684\u7f51\u3002\u6211\u4eec\u6309\u2019\u5e73\u53f0\u4ea4\u4ed8\u516d\u5927\u9886\u57df\u2019\u6765\u7ec4\u7ec7：\u201d<\/p>\n

    \u5e73\u53f0\u4ea4\u4ed8\u516d\u5927\u9886\u57df：<\/p>\n

    \u9886\u57df1：\u67b6\u6784\u8bbe\u8ba1
    \n – \u591a\u673a\u623f\u90e8\u7f72\u62d3\u6251\u56fe
    \n – \u6d41\u91cf\u8def\u7531\u4e0e\u5c31\u8fd1\u63a5\u5165
    \n – \u6570\u636e\u540c\u6b65\u7b56\u7565（MySQL \u4e3b\u4ece\u3001MinIO \u8de8\u533a\u57df\u590d\u5236）<\/p>\n

    \u9886\u57df2：\u6e90\u7801\u6269\u5c55
    \n – \u81ea\u5b9a\u4e49 Parser（WPS \u683c\u5f0f）
    \n – \u81ea\u5b9a\u4e49 Tool（Jira \u5de5\u5355\u96c6\u6210）
    \n – \u68c0\u7d22\u8017\u65f6\u57cb\u70b9\u4e0e\u5206\u6bb5\u76d1\u63a7<\/p>\n

    \u9886\u57df3：\u53ef\u89c2\u6d4b\u6027
    \n – Prometheus + Grafana \u79df\u6237\u7ea7\u4eea\u8868\u76d8
    \n – Jaeger \u5168\u94fe\u8def\u8ffd\u8e2a
    \n – 5 \u6761\u6838\u5fc3\u544a\u8b66 + \u5206\u7ea7\u901a\u77e5<\/p>\n

    \u9886\u57df4：SRE \u843d\u5730
    \n – SLO \u5b9a\u4e49\u4e0e\u76d1\u63a7（P95<8s \/ \u53ef\u7528\u7387 99.9%）
    \n – \u5bb9\u91cf\u89c4\u5212\u6a21\u578b
    \n – \u7070\u5ea6\u53d1\u5e03 + \u91d1\u4e1d\u96c0\u6d41\u91cf<\/p>\n

    \u9886\u57df5：\u8fd0\u7ef4\u6cbb\u7406
    \n – \u5907\u4efd\u6062\u590d（MySQL binlog + MinIO mirror）
    \n – \u4e8b\u6545\u9884\u6848\u4e0e\u6f14\u7ec3
    \n – \u6210\u672c\u76d1\u63a7（LLM \u8d39\u7528 \/ \u57fa\u7840\u8bbe\u65bd）<\/p>\n

    \u9886\u57df6：\u9a8c\u6536\u6807\u51c6
    \n – \u767e\u4e07\u7ea7\u6587\u6863 P95 \u68c0\u7d22 < 500ms
    \n – \u5343\u4e07\u7ea7 Chunk \u7d22\u5f15\u4e0d\u9000\u5316
    \n – \u6838\u5fc3\u94fe\u8def 100% \u53ef\u8ffd\u8e2a
    \n – \u89e3\u6790\u5931\u8d25\u7387 < 1%<\/p>\n

    \u5c0f\u80d6：\u201c\u591a\u673a\u623f\u90e8\u7f72\u2014\u2014\u5317\u4eac\u548c\u4e0a\u6d77\u4e24\u4e2a\u673a\u623f\u2014\u2014\u6587\u6863\u5b58\u54ea？\u7528\u6237\u600e\u4e48\u5c31\u8fd1\u8bbf\u95ee？\u201d<\/p>\n

    \u5927\u5e08：\u201c\u591a\u673a\u623f\u7684\u5173\u952e\u7b56\u7565\u662f\u2019\u8bfb\u5199\u5206\u79bb + \u5c31\u8fd1\u63a5\u5165\u2019：\u201d<\/p>\n

    \u591a\u673a\u623f\u67b6\u6784：<\/p>\n

    \u7528\u6237\u8bf7\u6c42
    \n \u2502
    \n \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2534\u2500\u2500\u2500\u2500\u2500\u2500\u2510
    \n \u2502 GeoDNS \u2502 (\u6839\u636e\u6765\u6e90 IP \u8def\u7531\u5230\u6700\u8fd1\u673a\u623f)
    \n \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u252c\u2500\u2500\u2500\u2500\u2500\u2500\u2518
    \n \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2534\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510
    \n \u25bc \u25bc
    \n \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510 \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510
    \n \u2502 \u5317\u4eac\u673a\u623f (\u4e3b) \u2502 \u2502 \u4e0a\u6d77\u673a\u623f (\u4ece) \u2502
    \n \u2502 \u2502 \u2502 \u2502
    \n \u2502 MySQL (\u4e3b, \u8bfb\u5199) \u2502\u25c0\u2500\u2500\u25b6\u2502 MySQL (\u4ece, \u53ea\u8bfb) \u2502 binlog \u5f02\u6b65\u590d\u5236
    \n \u2502 MinIO (\u4e3b) \u2502\u25c0\u2500\u2500\u25b6\u2502 MinIO (\u4ece) \u2502 mirror \u540c\u6b65
    \n \u2502 Infinity (\u4e3b) \u2502 \u2502 Infinity (\u53ea\u8bfb) \u2502 \u7d22\u5f15\u72ec\u7acb
    \n \u2502 API Server \u00d73 \u2502 \u2502 API Server \u00d73 \u2502
    \n \u2502 Task Executor \u00d72 \u2502 \u2502 Task Executor \u00d71 \u2502 (\u89e3\u6790\u4ec5\u5728\u5317\u4eac\u6267\u884c)
    \n \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518 \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518<\/p>\n

    \u7b56\u7565：
    \n – \u5199\u64cd\u4f5c（\u4e0a\u4f20\u6587\u6863\u3001\u521b\u5efa\u6570\u636e\u96c6）\u2192 \u5168\u90e8\u8def\u7531\u5230\u5317\u4eac
    \n – \u8bfb\u64cd\u4f5c（\u68c0\u7d22\u3001\u95ee\u7b54）\u2192 \u5c31\u8fd1\u673a\u623f\u5904\u7406
    \n – \u6587\u6863\u89e3\u6790 \u2192 \u4ec5\u5728\u5317\u4eac（Task Executor \u5728\u4e0a\u6d77\u90e8\u7f72\u4f46\u4e0d\u6d88\u8d39\u961f\u5217）<\/p>\n

    \u5c0f\u767d：\u201c\u90a3\u5982\u4f55\u4fdd\u8bc1\u6838\u5fc3\u94fe\u8def 100% \u53ef\u8ffd\u8e2a？\u201d<\/p>\n

    \u5927\u5e08：\u201cTraceID \u8d2f\u7a7f\u6574\u4e2a\u8bf7\u6c42\u94fe\u8def\u2014\u2014\u4ece API \u5165\u53e3\u5230 LLM \u8fd4\u56de：\u201d<\/p>\n

    # \u5e73\u53f0\u7ea7 TraceID \u6ce8\u5165\u2014\u2014\u8986\u76d6\u6240\u6709\u6838\u5fc3\u94fe\u8def<\/span>
    \nimport<\/span> uuid
    \nfrom<\/span> contextvars import<\/span> ContextVar<\/p>\n

    trace_id_var =<\/span> ContextVar(<\/span>"trace_id"<\/span>,<\/span> default=<\/span>"no-trace"<\/span>)<\/span><\/p>\n

    class<\/span> PlatformTracer<\/span>:<\/span>
    \n """\u5e73\u53f0\u7ea7\u5168\u94fe\u8def\u8ffd\u8e2a"""<\/span><\/p>\n

    @classmethod<\/span>
    \n def<\/span> inject_trace_id<\/span>(<\/span>cls,<\/span> request)<\/span>:<\/span>
    \n """\u5728 API \u5165\u53e3\u6ce8\u5165 TraceID"""<\/span>
    \n trace_id =<\/span> request.<\/span>headers.<\/span>get(<\/span>"X-Trace-ID"<\/span>)<\/span> or<\/span> str<\/span>(<\/span>uuid.<\/span>uuid4(<\/span>)<\/span>)<\/span>[<\/span>:<\/span>12<\/span>]<\/span>
    \n trace_id_var.<\/span>set<\/span>(<\/span>trace_id)<\/span><\/p>\n

    # \u6ce8\u5165\u5230\u65e5\u5fd7<\/span>
    \n logger.<\/span>info(<\/span>f"[trace=<\/span>{<\/span>trace_id}<\/span><\/span>] <\/span>{<\/span>request.<\/span>method}<\/span><\/span> <\/span>{<\/span>request.<\/span>path}<\/span><\/span>"<\/span><\/span>)<\/span><\/p>\n

    return<\/span> trace_id<\/p>\n

    @classmethod<\/span>
    \n def<\/span> trace_span<\/span>(<\/span>cls,<\/span> span_name,<\/span> **<\/span>attrs)<\/span>:<\/span>
    \n """\u521b\u5efa\u4e00\u4e2a\u8ffd\u8e2a Span，\u81ea\u52a8\u643a\u5e26 trace_id"""<\/span>
    \n trace_id =<\/span> trace_id_var.<\/span>get(<\/span>)<\/span>
    \n span =<\/span> tracer.<\/span>start_span(<\/span>span_name)<\/span>
    \n span.<\/span>set_attribute(<\/span>"trace_id"<\/span>,<\/span> trace_id)<\/span>
    \n for<\/span> k,<\/span> v in<\/span> attrs.<\/span>items(<\/span>)<\/span>:<\/span>
    \n span.<\/span>set_attribute(<\/span>k,<\/span> str<\/span>(<\/span>v)<\/span>[<\/span>:<\/span>200<\/span>]<\/span>)<\/span>
    \n return<\/span> span<\/p>\n

    # \u5728\u5173\u952e\u8282\u70b9\u57cb\u70b9<\/span>
    \n# 1. API \u5165\u53e3<\/span>
    \n# 2. Dealer \u68c0\u7d22（\u5411\u91cf\u641c\u7d22\u3001\u5173\u952e\u8bcd\u641c\u7d22\u3001RRF \u878d\u5408）<\/span>
    \n# 3. Rerank \u8c03\u7528<\/span>
    \n# 4. LLM \u8c03\u7528<\/span>
    \n# 5. \u6587\u6863\u89e3\u6790（\u4ec5 Task Executor）<\/span><\/p>\n

    3 \u9879\u76ee\u5b9e\u6218<\/h3>\n

    \u73af\u5883\u51c6\u5907<\/h4>\n

    \u76ee\u6807：\u57fa\u4e8e\u524d 39 \u7ae0\u77e5\u8bc6，\u5728 K8s \u96c6\u7fa4\u4e0a\u4ea4\u4ed8\u5b8c\u6574\u7684"\u53ef\u89c2\u6d4b\u9ad8\u53ef\u7528 RAGFlow \u5e73\u53f0"\u3002<\/p>\n

    \u524d\u63d0：K8s \u96c6\u7fa4（\u81f3\u5c11 6 \u8282\u70b9）\u3001Prometheus + Grafana + Jaeger + Loki \u53ef\u89c2\u6d4b\u6027\u5957\u4ef6\u5df2\u90e8\u7f72\u3002<\/p>\n

    \u5206\u6b65\u5b9e\u73b0<\/h4>\n

    \u6b65\u9aa41：\u591a\u673a\u623f K8s \u90e8\u7f72\u67b6\u6784<\/h5>\n

    # platform-production.yaml – \u751f\u4ea7\u7ea7\u90e8\u7f72\u6e05\u5355<\/span><\/p>\n

    # \u5317\u4eac\u673a\u623f：\u4e3b\u96c6\u7fa4<\/span>
    \n—<\/span>
    \napiVersion<\/span>:<\/span> apps\/v1
    \nkind<\/span>:<\/span> Deployment
    \nmetadata<\/span>:<\/span>
    \n name<\/span>:<\/span> ragflow–<\/span>api–<\/span>beijing
    \n namespace<\/span>:<\/span> ragflow–<\/span>prod
    \nspec<\/span>:<\/span>
    \n replicas<\/span>:<\/span> 5<\/span>
    \n template<\/span>:<\/span>
    \n spec<\/span>:<\/span>
    \n affinity<\/span>:<\/span>
    \n podAntiAffinity<\/span>:<\/span>
    \n requiredDuringSchedulingIgnoredDuringExecution<\/span>:<\/span>
    \n –<\/span> labelSelector<\/span>:<\/span>
    \n matchLabels<\/span>:<\/span>
    \n app<\/span>:<\/span> ragflow–<\/span>api
    \n topologyKey<\/span>:<\/span> kubernetes.io\/hostname # \u5206\u6563\u5230\u4e0d\u540c\u7269\u7406\u8282\u70b9<\/span>
    \n containers<\/span>:<\/span>
    \n –<\/span> name<\/span>:<\/span> api
    \n image<\/span>:<\/span> infiniflow\/ragflow:<\/span>v0.27.0
    \n env<\/span>:<\/span>
    \n –<\/span> name<\/span>:<\/span> DEPLOY_REGION
    \n value<\/span>:<\/span> "beijing"<\/span>
    \n –<\/span> name<\/span>:<\/span> MYSQL_HOST
    \n value<\/span>:<\/span> mysql–<\/span>primary.beijing.svc.cluster.local
    \n –<\/span> name<\/span>:<\/span> TRACE_ENABLED
    \n value<\/span>:<\/span> "true"<\/span>
    \n resources<\/span>:<\/span>
    \n requests<\/span>:<\/span> {<\/span>memory<\/span>:<\/span> "2Gi"<\/span>,<\/span> cpu<\/span>:<\/span> "1"<\/span>}<\/span>
    \n limits<\/span>:<\/span> {<\/span>memory<\/span>:<\/span> "4Gi"<\/span>,<\/span> cpu<\/span>:<\/span> "2"<\/span>}<\/span><\/p>\n

    —<\/span>
    \n# \u4e0a\u6d77\u673a\u623f：\u4ece\u96c6\u7fa4（\u53ea\u8bfb）<\/span>
    \napiVersion<\/span>:<\/span> apps\/v1
    \nkind<\/span>:<\/span> Deployment
    \nmetadata<\/span>:<\/span>
    \n name<\/span>:<\/span> ragflow–<\/span>api–<\/span>shanghai
    \n namespace<\/span>:<\/span> ragflow–<\/span>prod
    \nspec<\/span>:<\/span>
    \n replicas<\/span>:<\/span> 3<\/span>
    \n template<\/span>:<\/span>
    \n spec<\/span>:<\/span>
    \n containers<\/span>:<\/span>
    \n –<\/span> name<\/span>:<\/span> api
    \n image<\/span>:<\/span> infiniflow\/ragflow:<\/span>v0.27.0
    \n env<\/span>:<\/span>
    \n –<\/span> name<\/span>:<\/span> DEPLOY_REGION
    \n value<\/span>:<\/span> "shanghai"<\/span>
    \n –<\/span> name<\/span>:<\/span> MYSQL_HOST
    \n value<\/span>:<\/span> mysql–<\/span>replica.shanghai.svc.cluster.local
    \n –<\/span> name<\/span>:<\/span> READ_ONLY_MODE
    \n value<\/span>:<\/span> "true"<\/span> # \u53ea\u8bfb\u6a21\u5f0f（\u7981\u6b62\u5199\u64cd\u4f5c）<\/span>
    \n –<\/span> name<\/span>:<\/span> TRACE_ENABLED
    \n value<\/span>:<\/span> "true"<\/span><\/p>\n

    # GeoDNS \u914d\u7f6e（Cloudflare \/ Route53 \u793a\u4f8b）<\/span>
    \n# \u5317\u4eac\u7528\u6237 \u2192 \u89e3\u6790\u5230 beijing.ragflow.yunfan.com<\/span>
    \n# \u4e0a\u6d77\u7528\u6237 \u2192 \u89e3\u6790\u5230 shanghai.ragflow.yunfan.com<\/span>
    \n# \u5176\u4ed6\u7528\u6237 \u2192 \u5c31\u8fd1\u6216\u8f6e\u8be2<\/span><\/p>\n

    \u6b65\u9aa42：\u6e90\u7801\u6269\u5c55\u96c6\u6210\u2014\u2014\u68c0\u7d22\u8017\u65f6\u57cb\u70b9<\/h5>\n

    # rag\/nlp\/search.py \u589e\u52a0\u68c0\u7d22\u8017\u65f6\u57cb\u70b9（\u6574\u5408\u7b2c36\u7ae0\u77e5\u8bc6）<\/span><\/p>\n

    class<\/span> Dealer<\/span>:<\/span>
    \n async<\/span> def<\/span> search_with_trace<\/span>(<\/span>self,<\/span> question,<\/span> **<\/span>kwargs)<\/span>:<\/span>
    \n """\u5e26\u5168\u94fe\u8def\u8ffd\u8e2a\u7684\u68c0\u7d22"""<\/span>
    \n with<\/span> PlatformTracer.<\/span>trace_span(<\/span>"dealer.search"<\/span>,<\/span>
    \n question=<\/span>question[<\/span>:<\/span>100<\/span>]<\/span>,<\/span>
    \n dataset_count=<\/span>len<\/span>(<\/span>self.<\/span>dataset_ids)<\/span>)<\/span> as<\/span> span:<\/span><\/p>\n

    # \u5b50 Span: \u5411\u91cf\u68c0\u7d22<\/span>
    \n with<\/span> PlatformTracer.<\/span>trace_span(<\/span>"dealer.vector_search"<\/span>)<\/span> as<\/span> vs_span:<\/span>
    \n t0 =<\/span> time.<\/span>time(<\/span>)<\/span>
    \n vector_results =<\/span> await<\/span> self.<\/span>vector_search(<\/span>query_vector)<\/span>
    \n vs_span.<\/span>set_attribute(<\/span>"results_count"<\/span>,<\/span> len<\/span>(<\/span>vector_results)<\/span>)<\/span>
    \n vs_span.<\/span>set_attribute(<\/span>"latency_ms"<\/span>,<\/span> (<\/span>time.<\/span>time(<\/span>)<\/span> –<\/span> t0)<\/span> *<\/span> 1000<\/span>)<\/span><\/p>\n

    # \u5b50 Span: \u5173\u952e\u8bcd\u68c0\u7d22<\/span>
    \n with<\/span> PlatformTracer.<\/span>trace_span(<\/span>"dealer.keyword_search"<\/span>)<\/span> as<\/span> ks_span:<\/span>
    \n t0 =<\/span> time.<\/span>time(<\/span>)<\/span>
    \n keyword_results =<\/span> await<\/span> self.<\/span>keyword_search(<\/span>question)<\/span>
    \n ks_span.<\/span>set_attribute(<\/span>"results_count"<\/span>,<\/span> len<\/span>(<\/span>keyword_results)<\/span>)<\/span>
    \n ks_span.<\/span>set_attribute(<\/span>"latency_ms"<\/span>,<\/span> (<\/span>time.<\/span>time(<\/span>)<\/span> –<\/span> t0)<\/span> *<\/span> 1000<\/span>)<\/span><\/p>\n

    # \u5b50 Span: RRF \u878d\u5408<\/span>
    \n with<\/span> PlatformTracer.<\/span>trace_span(<\/span>"dealer.fusion"<\/span>)<\/span>:<\/span>
    \n fused =<\/span> self.<\/span>fusion(<\/span>vector_results,<\/span> keyword_results)<\/span><\/p>\n

    # \u5b50 Span: Rerank<\/span>
    \n if<\/span> self.<\/span>config.<\/span>get(<\/span>"rerank_model"<\/span>)<\/span>:<\/span>
    \n with<\/span> PlatformTracer.<\/span>trace_span(<\/span>"dealer.rerank"<\/span>)<\/span> as<\/span> r_span:<\/span>
    \n t0 =<\/span> time.<\/span>time(<\/span>)<\/span>
    \n fused =<\/span> await<\/span> self.<\/span>rerank(<\/span>question,<\/span> fused)<\/span>
    \n r_span.<\/span>set_attribute(<\/span>"latency_ms"<\/span>,<\/span> (<\/span>time.<\/span>time(<\/span>)<\/span> –<\/span> t0)<\/span> *<\/span> 1000<\/span>)<\/span><\/p>\n

    span.<\/span>set_attribute(<\/span>"total_results"<\/span>,<\/span> len<\/span>(<\/span>fused)<\/span>)<\/span>
    \n return<\/span> fused[<\/span>:<\/span>self.<\/span>config.<\/span>get(<\/span>"top_k"<\/span>,<\/span> 10<\/span>)<\/span>]<\/span><\/p>\n

    \u6b65\u9aa43：Grafana \u5e73\u53f0\u7ea7\u76d1\u63a7\u5927\u5c4f<\/h5>\n

    {<\/span>
    \n "dashboard"<\/span>:<\/span> {<\/span>
    \n "title"<\/span>:<\/span> "RAGFlow \u5e73\u53f0\u603b\u89c8 – \u751f\u4ea7"<\/span>,<\/span>
    \n "rows"<\/span>:<\/span> [<\/span>
    \n {<\/span>
    \n "title"<\/span>:<\/span> "\u5168\u5e73\u53f0\u6838\u5fc3\u6307\u6807"<\/span>,<\/span>
    \n "panels"<\/span>:<\/span> [<\/span>
    \n {<\/span>
    \n "title"<\/span>:<\/span> "\u603b QPS (\u6309\u673a\u623f)"<\/span>,<\/span>
    \n "targets"<\/span>:<\/span> [<\/span>
    \n {<\/span>"expr"<\/span>:<\/span> "sum(rate(chat_requests_total{region='beijing'}[1m]))"<\/span>,<\/span> "legendFormat"<\/span>:<\/span> "\u5317\u4eac"<\/span>}<\/span>,<\/span>
    \n {<\/span>"expr"<\/span>:<\/span> "sum(rate(chat_requests_total{region='shanghai'}[1m]))"<\/span>,<\/span> "legendFormat"<\/span>:<\/span> "\u4e0a\u6d77"<\/span>}<\/span>
    \n ]<\/span>
    \n }<\/span>,<\/span>
    \n {<\/span>
    \n "title"<\/span>:<\/span> "P95 \u5ef6\u8fdf (\u6309\u673a\u623f)"<\/span>,<\/span>
    \n "targets"<\/span>:<\/span> [<\/span>
    \n {<\/span>"expr"<\/span>:<\/span> "histogram_quantile(0.95, chat_latency_seconds_bucket{region='beijing'})"<\/span>,<\/span> "legendFormat"<\/span>:<\/span> "\u5317\u4eac"<\/span>}<\/span>,<\/span>
    \n {<\/span>"expr"<\/span>:<\/span> "histogram_quantile(0.95, chat_latency_seconds_bucket{region='shanghai'})"<\/span>,<\/span> "legendFormat"<\/span>:<\/span> "\u4e0a\u6d77"<\/span>}<\/span>
    \n ]<\/span>
    \n }<\/span>,<\/span>
    \n {<\/span>
    \n "title"<\/span>:<\/span> "LLM \u9519\u8bef\u7387 (\u6309\u4f9b\u5e94\u5546)"<\/span>,<\/span>
    \n "targets"<\/span>:<\/span> [<\/span>
    \n {<\/span>"expr"<\/span>:<\/span> "rate(llm_errors_total{factory='YunFan'}[5m]) \/ rate(llm_calls_total{factory='YunFan'}[5m])"<\/span>}<\/span>,<\/span>
    \n {<\/span>"expr"<\/span>:<\/span> "rate(llm_errors_total{factory='OpenAI'}[5m]) \/ rate(llm_calls_total{factory='OpenAI'}[5m])"<\/span>}<\/span>
    \n ]<\/span>
    \n }<\/span>,<\/span>
    \n {<\/span>
    \n "title"<\/span>:<\/span> "\u79df\u6237 SLO \u8fbe\u6807\u7387"<\/span>,<\/span>
    \n "targets"<\/span>:<\/span> [<\/span>
    \n {<\/span>"expr"<\/span>:<\/span> "sum(tenant_slo_compliant) \/ count(tenant_slo_compliant)"<\/span>}<\/span>
    \n ]<\/span>,<\/span>
    \n "thresholds"<\/span>:<\/span> [<\/span>{<\/span>"value"<\/span>:<\/span> 0.95<\/span>,<\/span> "color"<\/span>:<\/span> "green"<\/span>}<\/span>,<\/span> {<\/span>"value"<\/span>:<\/span> 0.9<\/span>,<\/span> "color"<\/span>:<\/span> "yellow"<\/span>}<\/span>,<\/span> {<\/span>"value"<\/span>:<\/span> 0<\/span>,<\/span> "color"<\/span>:<\/span> "red"<\/span>}<\/span>]<\/span>
    \n }<\/span>
    \n ]<\/span>
    \n }<\/span>,<\/span>
    \n {<\/span>
    \n "title"<\/span>:<\/span> "\u6210\u672c\u4e2d\u5fc3"<\/span>,<\/span>
    \n "panels"<\/span>:<\/span> [<\/span>
    \n {<\/span>
    \n "title"<\/span>:<\/span> "LLM \u8d39\u7528\u8d8b\u52bf (\u65e5)"<\/span>,<\/span>
    \n "targets"<\/span>:<\/span> [<\/span>{<\/span>"expr"<\/span>:<\/span> "llm_cost_daily_total"<\/span>}<\/span>]<\/span>
    \n }<\/span>,<\/span>
    \n {<\/span>
    \n "title"<\/span>:<\/span> "\u57fa\u7840\u8bbe\u65bd\u6210\u672c (\u6708)"<\/span>,<\/span>
    \n "targets"<\/span>:<\/span> [<\/span>{<\/span>"expr"<\/span>:<\/span> "kube_cost_total"<\/span>}<\/span>]<\/span>
    \n }<\/span>
    \n ]<\/span>
    \n }<\/span>
    \n ]<\/span>
    \n }<\/span>
    \n}<\/span><\/p>\n

    \u6b65\u9aa44：\u7070\u5ea6\u53d1\u5e03\u4e0e\u91d1\u4e1d\u96c0\u90e8\u7f72<\/h5>\n

    # canary-deployment.yaml<\/span>
    \n# \u91d1\u4e1d\u96c0\u53d1\u5e03：5% \u6d41\u91cf\u8d70\u65b0\u7248\u672c，95% \u8d70\u65e7\u7248\u672c<\/span><\/p>\n

    apiVersion<\/span>:<\/span> v1
    \nkind<\/span>:<\/span> Service
    \nmetadata<\/span>:<\/span>
    \n name<\/span>:<\/span> ragflow–<\/span>api–<\/span>canary
    \nspec<\/span>:<\/span>
    \n selector<\/span>:<\/span>
    \n app<\/span>:<\/span> ragflow–<\/span>api
    \n version<\/span>:<\/span> canary<\/p>\n

    —<\/span>
    \napiVersion<\/span>:<\/span> networking.k8s.io\/v1
    \nkind<\/span>:<\/span> Ingress
    \nmetadata<\/span>:<\/span>
    \n name<\/span>:<\/span> ragflow–<\/span>canary
    \n annotations<\/span>:<\/span>
    \n nginx.ingress.kubernetes.io\/canary<\/span>:<\/span> "true"<\/span>
    \n nginx.ingress.kubernetes.io\/canary-weight<\/span>:<\/span> "5"<\/span> # 5% \u6d41\u91cf<\/span>
    \nspec<\/span>:<\/span>
    \n rules<\/span>:<\/span>
    \n –<\/span> host<\/span>:<\/span> ragflow.yunfan.com
    \n http<\/span>:<\/span>
    \n paths<\/span>:<\/span>
    \n –<\/span> backend<\/span>:<\/span>
    \n service<\/span>:<\/span>
    \n name<\/span>:<\/span> ragflow–<\/span>api–<\/span>canary
    \n port<\/span>:<\/span>
    \n number<\/span>:<\/span> 9380<\/span><\/p>\n

    # \u7070\u5ea6\u53d1\u5e03\u6d41\u7a0b<\/span>
    \n# Step 1: \u90e8\u7f72\u91d1\u4e1d\u96c0\u7248\u672c<\/span>
    \nkubectl apply -f<\/span> canary-deployment.yaml<\/p>\n

    # Step 2: \u89c2\u5bdf\u91d1\u4e1d\u96c0\u6307\u6807（\u9519\u8bef\u7387\u3001\u5ef6\u8fdf）<\/span>
    \n# \u82e5\u91d1\u4e1d\u96c0\u7248\u672c\u9519\u8bef\u7387 > 2%，\u7acb\u5373\u56de\u6eda<\/span><\/p>\n

    # Step 3: \u9010\u6b65\u63d0\u9ad8\u91d1\u4e1d\u96c0\u6743\u91cd 5% \u2192 20% \u2192 50% \u2192 100%<\/span>
    \nkubectl annotate ingress ragflow-canary \\\\<\/span>
    \n nginx.ingress.kubernetes.io\/canary-weight=<\/span>"20"<\/span><\/p>\n

    # Step 4: \u5168\u91cf\u5207\u6362\u540e\u5c06\u91d1\u4e1d\u96c0\u7248\u672c\u63d0\u5347\u4e3a\u7a33\u5b9a\u7248<\/span>
    \nkubectl set<\/span> image deployment\/ragflow-api-stable \\\\<\/span>
    \n api<\/span>=<\/span>infiniflow\/ragflow:v0.27.0<\/p>\n

    # Step 5: \u5220\u9664\u91d1\u4e1d\u96c0\u90e8\u7f72<\/span>
    \nkubectl delete -f<\/span> canary-deployment.yaml<\/p>\n

    \u6b65\u9aa45：\u5907\u4efd\u6062\u590d\u4e0e\u4e8b\u6545\u9884\u6848<\/h5>\n

    # disaster_recovery.sh – \u707e\u96be\u6062\u590d\u811a\u672c<\/span>
    \n#!\/bin\/bash<\/span>
    \necho<\/span> "=== RAGFlow \u707e\u96be\u6062\u590d\u9884\u6848 ==="<\/span><\/p>\n

    # === \u5907\u4efd\u7b56\u7565 ===<\/span>
    \n# MySQL: \u6bcf\u65e5\u5168\u91cf + binlog \u5b9e\u65f6\u5907\u4efd<\/span>
    \nmysqldump -h<\/span> mysql-primary –all-databases –single-transaction \\\\<\/span>
    \n |<\/span> gzip<\/span> ><\/span> \/backup\/mysql\/ragflow_$(<\/span>date<\/span> +%Y%m%d)<\/span><\/span>.sql.gz<\/p>\n

    # MinIO: \u5b9e\u65f6 mirror \u5230\u5907\u7528\u8282\u70b9<\/span>
    \nmc<\/span> mirror –watch<\/span> ragflow-minio\/ragflow backup-minio\/ragflow-backup &<\/span><\/p>\n

    # Infinity\/ES: \u6bcf\u65e5\u5feb\u7167<\/span>
    \ncurl<\/span> -X<\/span> PUT "http:\/\/infinity:23820\/db\/ragflow\/_snapshot\/daily_$(<\/span>date<\/span> +%Y%m%d)<\/span><\/span>"<\/span><\/p>\n

    # === \u6062\u590d\u6d41\u7a0b ===<\/span>
    \n# \u573a\u666f1: MySQL \u6570\u636e\u635f\u574f<\/span>
    \n# 1. \u505c\u6b62 API Server (\u9632\u6b62\u5199\u5165\u810f\u6570\u636e)<\/span>
    \n# 2. \u4ece\u6700\u8fd1\u5907\u4efd\u6062\u590d<\/span>
    \n# zcat \/backup\/mysql\/ragflow_20250615.sql.gz | mysql -h mysql-primary<\/span>
    \n# 3. \u5e94\u7528 binlog \u5230\u6545\u969c\u524d\u4e00\u523b<\/span>
    \n# 4. \u91cd\u542f API Server<\/span><\/p>\n

    # \u573a\u666f2: \u6574\u4e2a\u5317\u4eac\u673a\u623f\u5b95\u673a<\/span>
    \n# 1. GeoDNS \u5c06\u6240\u6709\u6d41\u91cf\u5207\u6362\u5230\u4e0a\u6d77\u673a\u623f<\/span>
    \n# 2. \u4e0a\u6d77 MySQL \u4ece\u5e93\u63d0\u5347\u4e3a\u4e3b\u5e93<\/span>
    \n# mysql -h mysql-replica.shanghai -e "STOP SLAVE; RESET SLAVE ALL;"<\/span>
    \n# 3. \u4e0a\u6d77 Task Executor \u5f00\u59cb\u6d88\u8d39 Redis \u961f\u5217<\/span>
    \n# 4. \u901a\u77e5\u7528\u6237\u670d\u52a1\u964d\u7ea7（\u4ec5\u8bfb\u53ef\u7528，\u5199\u64cd\u4f5c\u6682\u4e0d\u53ef\u7528）<\/span><\/p>\n

    # \u573a\u666f3: LLM API \u5168\u90e8\u4e0d\u53ef\u7528<\/span>
    \n# 1. \u81ea\u52a8\u5207\u6362\u5230\u672c\u5730 Ollama \u6a21\u578b（\u5df2\u5728 LLMBundle \u964d\u7ea7\u94fe\u4e2d\u914d\u7f6e）<\/span>
    \n# 2. \u524d\u7aef\u5c55\u793a"\u5f53\u524d\u4f7f\u7528\u5907\u7528\u6a21\u578b，\u56de\u7b54\u8d28\u91cf\u53ef\u80fd\u4e0b\u964d"<\/span>
    \n# 3. \u4e3b\u6a21\u578b\u6062\u590d\u540e\u81ea\u52a8\u5207\u56de<\/span><\/p>\n

    echo<\/span> "\u9884\u6848\u68c0\u67e5\u5b8c\u6210"<\/span><\/p>\n

    \u6b65\u9aa46：\u4e8b\u6545\u6f14\u7ec3\u2014\u2014\u6df7\u6c8c\u5de5\u7a0b<\/h5>\n

    # chaos_test.py – \u6df7\u6c8c\u5de5\u7a0b\u6d4b\u8bd5<\/span>
    \nimport<\/span> random
    \nimport<\/span> subprocess
    \nimport<\/span> time<\/p>\n

    class<\/span> ChaosTest<\/span>:<\/span>
    \n """RAGFlow \u5e73\u53f0\u6df7\u6c8c\u5de5\u7a0b\u6d4b\u8bd5"""<\/span><\/p>\n

    @classmethod<\/span>
    \n def<\/span> kill_random_api_pod<\/span>(<\/span>cls)<\/span>:<\/span>
    \n """\u968f\u673a\u6740\u6b7b\u4e00\u4e2a API Server Pod\u2014\u2014\u9a8c\u8bc1\u81ea\u52a8\u6062\u590d"""<\/span>
    \n pods =<\/span> subprocess.<\/span>run(<\/span>
    \n [<\/span>"kubectl"<\/span>,<\/span> "get"<\/span>,<\/span> "pods"<\/span>,<\/span> "-n"<\/span>,<\/span> "ragflow-prod"<\/span>,<\/span> "-l"<\/span>,<\/span> "app=ragflow-api"<\/span>,<\/span>
    \n "-o"<\/span>,<\/span> "jsonpath={.items[*].metadata.name}"<\/span>]<\/span>,<\/span>
    \n capture_output=<\/span>True<\/span>,<\/span> text=<\/span>True<\/span>
    \n )<\/span>.<\/span>stdout.<\/span>split(<\/span>)<\/span>
    \n victim =<\/span> random.<\/span>choice(<\/span>pods)<\/span>
    \n print<\/span>(<\/span>f"💣 \u6740\u6b7b Pod: <\/span>{<\/span>victim}<\/span><\/span>"<\/span><\/span>)<\/span>
    \n subprocess.<\/span>run(<\/span>[<\/span>"kubectl"<\/span>,<\/span> "delete"<\/span>,<\/span> "pod"<\/span>,<\/span> victim,<\/span> "-n"<\/span>,<\/span> "ragflow-prod"<\/span>]<\/span>)<\/span><\/p>\n

    @classmethod<\/span>
    \n def<\/span> inject_network_delay<\/span>(<\/span>cls,<\/span> target,<\/span> delay_ms=<\/span>2000<\/span>)<\/span>:<\/span>
    \n """\u6ce8\u5165\u7f51\u7edc\u5ef6\u8fdf\u2014\u2014\u6d4b\u8bd5\u8d85\u65f6\u5904\u7406"""<\/span>
    \n print<\/span>(<\/span>f"🐌 \u6ce8\u5165 <\/span>{<\/span>delay_ms}<\/span><\/span>ms \u5ef6\u8fdf\u5230 <\/span>{<\/span>target}<\/span><\/span>"<\/span><\/span>)<\/span>
    \n subprocess.<\/span>run(<\/span>[<\/span>
    \n "kubectl"<\/span>,<\/span> "exec"<\/span>,<\/span> target,<\/span> "-n"<\/span>,<\/span> "ragflow-prod"<\/span>,<\/span> "–"<\/span>,<\/span>
    \n "tc"<\/span>,<\/span> "qdisc"<\/span>,<\/span> "add"<\/span>,<\/span> "dev"<\/span>,<\/span> "eth0"<\/span>,<\/span> "root"<\/span>,<\/span> "netem"<\/span>,<\/span>
    \n f"delay"<\/span><\/span>,<\/span> f"<\/span>{<\/span>delay_ms}<\/span><\/span>ms"<\/span><\/span>
    \n ]<\/span>)<\/span><\/p>\n

    @classmethod<\/span>
    \n def<\/span> drain_redis_memory<\/span>(<\/span>cls)<\/span>:<\/span>
    \n """\u586b\u6ee1 Redis \u5185\u5b58\u2014\u2014\u6d4b\u8bd5 OOM \u6062\u590d"""<\/span>
    \n print<\/span>(<\/span>"💧 \u586b\u6ee1 Redis \u5185\u5b58…"<\/span>)<\/span>
    \n subprocess.<\/span>run(<\/span>[<\/span>
    \n "kubectl"<\/span>,<\/span> "exec"<\/span>,<\/span> "redis-0"<\/span>,<\/span> "-n"<\/span>,<\/span> "ragflow-prod"<\/span>,<\/span> "–"<\/span>,<\/span>
    \n "redis-cli"<\/span>,<\/span> "DEBUG"<\/span>,<\/span> "POPULATE"<\/span>,<\/span> "1000000"<\/span>,<\/span> "testkey"<\/span>,<\/span> "1000"<\/span>
    \n ]<\/span>)<\/span><\/p>\n

    @classmethod<\/span>
    \n async<\/span> def<\/span> run_chaos_experiment<\/span>(<\/span>cls)<\/span>:<\/span>
    \n """\u8fd0\u884c\u6df7\u6c8c\u5b9e\u9a8c\u2014\u2014\u540c\u65f6\u76d1\u63a7\u4e1a\u52a1\u6307\u6807"""<\/span>
    \n # 1. \u8bb0\u5f55\u5b9e\u9a8c\u524d\u7684\u57fa\u51c6\u6307\u6807<\/span>
    \n baseline =<\/span> cls.<\/span>_get_current_metrics(<\/span>)<\/span><\/p>\n

    # 2. \u6ce8\u5165\u6545\u969c<\/span>
    \n cls.<\/span>kill_random_api_pod(<\/span>)<\/span>
    \n await<\/span> asyncio.<\/span>sleep(<\/span>10<\/span>)<\/span><\/p>\n

    # 3. \u68c0\u67e5\u662f\u5426\u89e6\u53d1\u544a\u8b66<\/span>
    \n alerts =<\/span> cls.<\/span>_get_firing_alerts(<\/span>)<\/span><\/p>\n

    # 4. \u7b49\u5f85\u7cfb\u7edf\u81ea\u52a8\u6062\u590d<\/span>
    \n await<\/span> asyncio.<\/span>sleep(<\/span>30<\/span>)<\/span><\/p>\n

    # 5. \u68c0\u67e5\u6062\u590d\u540e\u7684\u6307\u6807<\/span>
    \n recovered =<\/span> cls.<\/span>_get_current_metrics(<\/span>)<\/span><\/p>\n

    # 6. \u8f93\u51fa\u5b9e\u9a8c\u62a5\u544a<\/span>
    \n print<\/span>(<\/span>"\\\\n=== \u6df7\u6c8c\u5b9e\u9a8c\u62a5\u544a ==="<\/span>)<\/span>
    \n print<\/span>(<\/span>f"\u6545\u969c\u6ce8\u5165: \u6740\u6b7b\u968f\u673a API Pod"<\/span><\/span>)<\/span>
    \n print<\/span>(<\/span>f"\u89e6\u53d1\u544a\u8b66: <\/span>{<\/span>len<\/span>(<\/span>alerts)<\/span>}<\/span><\/span> \u6761 (<\/span>{<\/span>[<\/span>a[<\/span>'name'<\/span>]<\/span> for<\/span> a in<\/span> alerts]<\/span>}<\/span><\/span>)"<\/span><\/span>)<\/span>
    \n print<\/span>(<\/span>f"\u6062\u590d\u65f6\u95f4: <\/span>{<\/span>recovered[<\/span>'timestamp'<\/span>]<\/span> –<\/span> baseline[<\/span>'timestamp'<\/span>]<\/span>:<\/span>.0f<\/span>}<\/span><\/span>s"<\/span><\/span>)<\/span>
    \n print<\/span>(<\/span>f"\u53ef\u7528\u7387\u5f71\u54cd: <\/span>{<\/span>recovered[<\/span>'availability'<\/span>]<\/span> –<\/span> baseline[<\/span>'availability'<\/span>]<\/span>:<\/span>.2%<\/span>}<\/span><\/span>"<\/span><\/span>)<\/span>
    \n print<\/span>(<\/span>f"\u7ed3\u8bba: <\/span>{<\/span>'\u2705 \u901a\u8fc7'<\/span> if<\/span> recovered[<\/span>'availability'<\/span>]<\/span> ><\/span> 0.995<\/span> else<\/span> '\u274c \u672a\u901a\u8fc7'<\/span>}<\/span><\/span>"<\/span><\/span>)<\/span><\/p>\n

    \u6b65\u9aa47：\u9a8c\u6536\u81ea\u68c0\u6e05\u5355<\/h5>\n

    # acceptance_checklist.py<\/span>
    \ndef<\/span> run_acceptance_checks<\/span>(<\/span>)<\/span>:<\/span>
    \n """\u5e73\u53f0\u4ea4\u4ed8\u9a8c\u6536\u81ea\u68c0"""<\/span>
    \n checks =<\/span> [<\/span>]<\/span><\/p>\n

    # 1. \u6838\u5fc3\u94fe\u8def\u53ef\u8ffd\u8e2a<\/span>
    \n trace_id =<\/span> str<\/span>(<\/span>uuid.<\/span>uuid4(<\/span>)<\/span>)<\/span>
    \n resp =<\/span> requests.<\/span>get(<\/span>f"<\/span>{<\/span>BASE}<\/span><\/span>\/api\/v1\/chats\/xxx"<\/span><\/span>,<\/span> headers=<\/span>{<\/span>"X-Trace-ID"<\/span>:<\/span> trace_id}<\/span>)<\/span>
    \n jaeger_trace =<\/span> get_jaeger_trace(<\/span>trace_id)<\/span>
    \n checks.<\/span>append(<\/span>(<\/span>"\u6838\u5fc3\u94fe\u8def TraceID \u8ffd\u8e2a"<\/span>,<\/span> jaeger_trace is<\/span> not<\/span> None<\/span>)<\/span>)<\/span><\/p>\n

    # 2. \u767e\u4e07\u7ea7\u6587\u6863 P95 \u68c0\u7d22 < 500ms<\/span>
    \n p95 =<\/span> benchmark_search_p95(<\/span>doc_count=<\/span>1000000<\/span>)<\/span>
    \n checks.<\/span>append(<\/span>(<\/span>"\u767e\u4e07\u6587\u6863 P95 \u68c0\u7d22"<\/span>,<\/span> p95 <<\/span> 500<\/span>)<\/span>)<\/span><\/p>\n

    # 3. \u81ea\u52a8\u6545\u969c\u6062\u590d<\/span>
    \n kill_api_pod(<\/span>)<\/span>
    \n time.<\/span>sleep(<\/span>30<\/span>)<\/span>
    \n pod_recovered =<\/span> count_running_api_pods(<\/span>)<\/span> >=<\/span> 3<\/span>
    \n checks.<\/span>append(<\/span>(<\/span>"API Pod \u81ea\u52a8\u6062\u590d"<\/span>,<\/span> pod_recovered)<\/span>)<\/span><\/p>\n

    # 4. \u8de8\u673a\u623f\u5c31\u8fd1\u63a5\u5165<\/span>
    \n beijing_latency =<\/span> measure_latency_from(<\/span>"beijing"<\/span>)<\/span>
    \n shanghai_latency =<\/span> measure_latency_from(<\/span>"shanghai"<\/span>)<\/span>
    \n checks.<\/span>append(<\/span>(<\/span>"\u5c31\u8fd1\u63a5\u5165\u5ef6\u8fdf"<\/span>,<\/span> beijing_latency <<\/span> 100<\/span> and<\/span> shanghai_latency <<\/span> 100<\/span>)<\/span>)<\/span><\/p>\n

    # 5. \u89e3\u6790\u5931\u8d25\u7387 < 1%<\/span>
    \n failure_rate =<\/span> get_parse_failure_rate_last_24h(<\/span>)<\/span>
    \n checks.<\/span>append(<\/span>(<\/span>"\u89e3\u6790\u5931\u8d25\u7387"<\/span>,<\/span> failure_rate <<\/span> 0.01<\/span>)<\/span>)<\/span><\/p>\n

    # \u8f93\u51fa\u9a8c\u6536\u62a5\u544a<\/span>
    \n print<\/span>(<\/span>"\\\\n=== \u5e73\u53f0\u9a8c\u6536\u62a5\u544a ==="<\/span>)<\/span>
    \n all_pass =<\/span> True<\/span>
    \n for<\/span> name,<\/span> result in<\/span> checks:<\/span>
    \n status =<\/span> "\u2705"<\/span> if<\/span> result else<\/span> "\u274c"<\/span>
    \n print<\/span>(<\/span>f" <\/span>{<\/span>status}<\/span><\/span> <\/span>{<\/span>name}<\/span><\/span>: <\/span>{<\/span>'\u901a\u8fc7'<\/span> if<\/span> result else<\/span> '\u672a\u901a\u8fc7'<\/span>}<\/span><\/span>"<\/span><\/span>)<\/span>
    \n if<\/span> not<\/span> result:<\/span>
    \n all_pass =<\/span> False<\/span>
    \n print<\/span>(<\/span>f"\\\\n\u603b\u4f53: <\/span>{<\/span>'\u2705 \u9a8c\u6536\u901a\u8fc7'<\/span> if<\/span> all_pass else<\/span> '\u274c \u9a8c\u6536\u672a\u901a\u8fc7'<\/span>}<\/span><\/span>"<\/span><\/span>)<\/span><\/p>\n

    \u5b8c\u6574\u4ee3\u7801\u6e05\u5355<\/h4>\n\n\u8def\u5f84\u8bf4\u660e<\/tr>\n\n\n\n\n\n
    column\/chapter40\/<\/td>\n\u672c\u7ae0\u90e8\u7f72\u6e05\u5355\u3001\u811a\u672c\u3001\u76d1\u63a7\u914d\u7f6e<\/td>\n<\/tr>\n
    api\/ragflow_server.py<\/td>\n\u5e73\u53f0\u5165\u53e3（TraceID \u6ce8\u5165）<\/td>\n<\/tr>\n
    rag\/nlp\/search.py<\/td>\nDealer \u68c0\u7d22\u57cb\u70b9\u7248<\/td>\n<\/tr>\n
    agent\/canvas.py<\/td>\nCanvas \u8ffd\u8e2a\u7248<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

    4 \u9879\u76ee\u603b\u7ed3<\/h3>\n

    \u4f18\u70b9 & \u7f3a\u70b9<\/h4>\n\n\u7ef4\u5ea6RAGFlow \u81ea\u5efa\u5e73\u53f0\u4e91\u5382\u5546 RAG \u670d\u52a1\u5f00\u6e90 + \u81ea\u8fd0\u7ef4\u5546\u4e1a RAG \u5e73\u53f0<\/tr>\n\n\n\n\n\n\n\n
    \u5b9a\u5236\u5316<\/td>\n\u2605\u2605\u2605 \u5b8c\u5168\u53ef\u63a7<\/td>\n\u2605\u2605\u2606 \u6709\u9650<\/td>\n\u2605\u2605\u2605 \u5b8c\u5168\u53ef\u63a7<\/td>\n\u2605\u2606\u2606 \u9ed1\u76d2<\/td>\n<\/tr>\n
    \u8fd0\u7ef4\u6210\u672c<\/td>\n\u2605\u2605\u2606 \u9700\u8981SRE\u56e2\u961f<\/td>\n\u2605\u2605\u2605 \u5e73\u53f0\u627f\u62c5<\/td>\n\u2605\u2605\u2606 \u9700\u8981\u56e2\u961f<\/td>\n\u2605\u2605\u2605 \u5e73\u53f0\u627f\u62c5<\/td>\n<\/tr>\n
    \u6269\u5c55\u6027<\/td>\n\u2605\u2605\u2605 \u6e90\u7801\u7ea7\u6269\u5c55<\/td>\n\u2605\u2605\u2606 \u63d2\u4ef6\u6709\u9650<\/td>\n\u2605\u2605\u2605 \u6e90\u7801\u7ea7<\/td>\n\u2605\u2606\u2606 \u53d7\u9650<\/td>\n<\/tr>\n
    \u53ef\u89c2\u6d4b\u6027<\/td>\n\u2605\u2605\u2605 \u5b8c\u5168\u5b9a\u5236<\/td>\n\u2605\u2605\u2606 \u5e73\u53f0\u63d0\u4f9b<\/td>\n\u2605\u2605\u2605 \u5b8c\u5168\u5b9a\u5236<\/td>\n\u2605\u2605\u2605 \u5185\u7f6e<\/td>\n<\/tr>\n
    \u8de8\u673a\u623f<\/td>\n\u2605\u2605\u2606 \u9700\u81ea\u5efa<\/td>\n\u2605\u2605\u2605 \u5e73\u53f0\u63d0\u4f9b<\/td>\n\u2605\u2605\u2606 \u9700\u81ea\u5efa<\/td>\n\u2605\u2605\u2605 \u5e73\u53f0\u63d0\u4f9b<\/td>\n<\/tr>\n
    \u6210\u672c<\/td>\n\u2605\u2605\u2605 \u53ef\u63a7<\/td>\n\u2605\u2605\u2606 \u6309\u91cf<\/td>\n\u2605\u2605\u2605 \u53ef\u63a7<\/td>\n\u2605\u2606\u2606 \u6602\u8d35<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

    \u9002\u7528\u573a\u666f<\/h4>\n
  • \u4f01\u4e1a\u7ea7\u77e5\u8bc6\u7ba1\u7406\u5e73\u53f0：20+ \u90e8\u95e8\u5171\u4eab，\u767e\u4e07\u7ea7\u6587\u6863，\u5343\u7ea7\u5e76\u53d1\u7528\u6237\u3002<\/li>\n
  • \u591a\u673a\u623f\u90e8\u7f72：\u8de8\u5730\u57df\u56e2\u961f，\u5c31\u8fd1\u8bbf\u95ee+\u6570\u636e\u540c\u6b65\u3002<\/li>\n
  • \u9700\u8981\u6e90\u7801\u7ea7\u5b9a\u5236：WPS Parser\u3001Jira Tool\u3001\u5ba1\u6279\u7ec4\u4ef6\u7b49\u5df2\u5f00\u53d1\u7684\u81ea\u5b9a\u4e49\u6269\u5c55\u3002<\/li>\n
  • SRE \u4f53\u7cfb\u5b8c\u5584\u7684\u7ec4\u7ec7：\u6709\u76d1\u63a7\u3001\u544a\u8b66\u3001\u4e8b\u6545\u9884\u6848\u3001\u6df7\u6c8c\u5de5\u7a0b\u7684\u56e2\u961f\u3002<\/li>\n
  • \u6210\u672c\u654f\u611f\u4e14\u5bf9\u6570\u636e\u5b89\u5168\u8981\u6c42\u9ad8：\u5168\u90e8\u79c1\u6709\u5316\u90e8\u7f72，\u65e0\u6570\u636e\u51fa\u5883\u3002<\/li>\n

    \u4e0d\u9002\u7528\u573a\u666f：<\/p>\n

  • 3 \u4eba\u5c0f\u56e2\u961f：\u5168\u5957\u5e73\u53f0\u8fd0\u7ef4（K8s+Prometheus+Jaeger+\u591a\u673a\u623f）\u9700\u8981\u81f3\u5c11 2 \u4e2a\u4e13\u804c SRE\u3002<\/li>\n
  • \u4e0d\u9700\u8981\u5b9a\u5236\u5316：\u5982\u679c\u6807\u51c6 RAGFlow \u529f\u80fd\u5b8c\u5168\u6ee1\u8db3\u9700\u6c42，\u4e0d\u9700\u8981\u9ad8\u7ea7\u7bc7\u7684\u6e90\u7801\u6269\u5c55\u3002<\/li>\n

    \u6ce8\u610f\u4e8b\u9879<\/h4>\n
  • \u7070\u5ea6\u53d1\u5e03\u5fc5\u987b\u6709\u56de\u6eda\u9884\u6848：\u91d1\u4e1d\u96c0\u7248\u672c\u7684\u65b0\u529f\u80fd\u53ef\u80fd\u5f15\u5165"\u9759\u9ed8\u9519\u8bef"（\u4e0d\u62a5\u9519\u4f46\u7ed3\u679c\u9519\u8bef）\u2014\u2014\u9700\u8981\u4e1a\u52a1\u6307\u6807\u76d1\u63a7\u800c\u4e0d\u4ec5\u662f\u6280\u672f\u6307\u6807\u3002<\/li>\n
  • \u8de8\u673a\u623f\u7684\u6570\u636e\u540c\u6b65\u5ef6\u8fdf：MySQL binlog \u4e3b\u4ece\u590d\u5236\u901a\u5e38\u6709 50-200ms \u5ef6\u8fdf\u3002\u5982\u679c\u7528\u6237\u5728 A \u673a\u623f\u4e0a\u4f20\u6587\u6863\u540e\u7acb\u5373\u5728 B \u673a\u623f\u68c0\u7d22\u2014\u2014\u53ef\u80fd\u67e5\u4e0d\u5230\u3002<\/li>\n
  • LLM \u8d39\u7528\u5931\u63a7：\u5343\u4e07\u7ea7 Chunk \u7684 Embedding \u8d39\u7528\u53ef\u80fd\u9ad8\u8fbe\u6570\u5343\u7f8e\u5143\u2014\u2014\u9700\u8981\u8d39\u7528\u9884\u7b97\u544a\u8b66\u3002<\/li>\n
  • \u6df7\u6c8c\u5b9e\u9a8c\u5fc5\u987b\u5728\u975e\u4e1a\u52a1\u9ad8\u5cf0\u671f\u6267\u884c：\u5b9e\u9a8c\u672c\u8eab\u4f1a\u5bfc\u81f4\u77ed\u6682\u7684\u670d\u52a1\u52a3\u5316\u2014\u2014\u63d0\u524d\u901a\u77e5\u7528\u6237\u3002<\/li>\n

    \u4ea4\u4ed8\u7269\u6e05\u5355（\u672c\u7ae0\u4ea7\u51fa）：<\/p>\n\n\u4ea4\u4ed8\u7269\u683c\u5f0f\u7528\u9014<\/tr>\n\n\n\n\n\n\n\n\n\n\n\n
    \u591a\u673a\u623f\u67b6\u6784\u56fe<\/td>\ndraw.io\/Mermaid<\/td>\n\u6280\u672f\u8bc4\u5ba1\u3001\u5411\u9886\u5bfc\u6c47\u62a5<\/td>\n<\/tr>\n
    K8s \u90e8\u7f72\u6e05\u5355<\/td>\nYAML<\/td>\n\u4e00\u952e\u90e8\u7f72\u5230\u65b0\u96c6\u7fa4<\/td>\n<\/tr>\n
    \u6e90\u7801\u6269\u5c55\u8865\u4e01<\/td>\nPython diff<\/td>\n\u68c0\u7d22\u57cb\u70b9\u3001TraceID \u6ce8\u5165<\/td>\n<\/tr>\n
    \u76d1\u63a7\u5927\u5c4f JSON<\/td>\nGrafana JSON<\/td>\n\u5bfc\u5165\u5373\u7528\u7684\u5e73\u53f0\u4eea\u8868\u76d8<\/td>\n<\/tr>\n
    SLO \u5b9a\u4e49\u6587\u6863<\/td>\nMarkdown<\/td>\n\u5bf9\u9f50\u4e1a\u52a1\u4e0e\u6280\u672f\u7684\u627f\u8bfa<\/td>\n<\/tr>\n
    \u5bb9\u91cf\u89c4\u5212\u62a5\u544a<\/td>\nCSV + \u56fe\u8868<\/td>\n3\/6\/12 \u4e2a\u6708\u6269\u5bb9\u8ba1\u5212<\/td>\n<\/tr>\n
    \u5907\u4efd\u6062\u590d Runbook<\/td>\nShell + \u6587\u6863<\/td>\n\u8fd0\u7ef4\u503c\u73ed\u7684\u6807\u51c6\u64cd\u4f5c\u6d41\u7a0b<\/td>\n<\/tr>\n
    \u6df7\u6c8c\u5b9e\u9a8c\u62a5\u544a<\/td>\nMarkdown<\/td>\n\u8bc1\u660e\u5e73\u53f0\u97e7\u6027<\/td>\n<\/tr>\n
    \u9a8c\u6536\u81ea\u68c0\u811a\u672c<\/td>\nPython<\/td>\n\u6bcf\u6b21\u4e0a\u7ebf\u524d\u7684\u8d28\u91cf\u95e8<\/td>\n<\/tr>\n
    \u6210\u672c\u76d1\u63a7\u65e5\u62a5<\/td>\nPython + Cron<\/td>\n\u6309\u90e8\u95e8\/\u4f9b\u5e94\u5546\u7684\u6bcf\u65e5\u8d39\u7528<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

    \u4e0e\u57fa\u7840\u7bc7\u7efc\u5408\u5b9e\u6218（\u7b2c16\u7ae0）\u7684\u5bf9\u6bd4：<\/p>\n\n\u7ef4\u5ea6\u7b2c16\u7ae0（\u57fa\u7840\u7bc7\u4ea4\u4ed8）\u7b2c40\u7ae0（\u9ad8\u7ea7\u7bc7\u4ea4\u4ed8）<\/tr>\n\n\n\n\n\n\n\n\n\n\n\n
    \u90e8\u7f72\u65b9\u5f0f<\/td>\nDocker Compose \u5355\u673a<\/td>\nK8s \u591a\u673a\u623f\u96c6\u7fa4<\/td>\n<\/tr>\n
    \u7528\u6237\u89c4\u6a21<\/td>\n500 \u4eba \/ 1 \u4e2a\u90e8\u95e8<\/td>\n2000 \u4eba \/ 20 \u4e2a\u90e8\u95e8<\/td>\n<\/tr>\n
    \u6587\u6863\u89c4\u6a21<\/td>\n50 \u4efd \/ 5000 Chunk<\/td>\n20 \u4e07\u4efd \/ 1000 \u4e07 Chunk<\/td>\n<\/tr>\n
    \u53ef\u7528\u6027<\/td>\n\u5355\u70b9\u6545\u969c<\/td>\n99.9% SLO<\/td>\n<\/tr>\n
    \u53ef\u89c2\u6d4b\u6027<\/td>\ndocker logs<\/td>\nPrometheus + Grafana + Jaeger + Loki<\/td>\n<\/tr>\n
    \u6269\u5c55\u7ba1\u7406<\/td>\n\u65e0<\/td>\n\u4f01\u4e1a\u6269\u5c55\u6ce8\u518c\u4e2d\u5fc3<\/td>\n<\/tr>\n
    \u5b89\u5168<\/td>\n\u57fa\u7840\u6743\u9650<\/td>\n\u4e94\u5c42\u9694\u79bb + \u6c99\u7bb1 + \u5ba1\u8ba1<\/td>\n<\/tr>\n
    \u8fd0\u7ef4<\/td>\n\u4eba\u5de5\u91cd\u542f<\/td>\n\u81ea\u52a8\u6062\u590d + \u7070\u5ea6\u53d1\u5e03 + \u6df7\u6c8c\u5de5\u7a0b<\/td>\n<\/tr>\n
    \u4ea4\u4ed8\u5468\u671f<\/td>\n2 \u5468（\u5b66\u4e60+\u642d\u5efa）<\/td>\n3 \u4e2a\u6708（\u9010\u6b65\u6f14\u8fdb）<\/td>\n<\/tr>\n
    5. \u5907\u4efd\u6062\u590d\u5fc5\u987b\u5b9a\u671f\u6f14\u7ec3：\u4e0d\u80fd\u7b49\u5230\u771f\u51fa\u4e8b\u4e86\u624d\u7b2c\u4e00\u6b21\u6267\u884c\u6062\u590d\u6d41\u7a0b\u2014\u2014\u5efa\u8bae\u6bcf\u5b63\u5ea6\u505a\u4e00\u6b21\u6062\u590d\u6f14\u7ec3\u3002<\/td>\n<\/td>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

    \u5e38\u89c1\u8e29\u5751\u7ecf\u9a8c<\/h4>\n\n\u6545\u969c\u73b0\u8c61\u6839\u56e0\u89e3\u51b3\u65b9\u6cd5<\/tr>\n\n\n\n\n\n\n\n\n\n
    \u91d1\u4e1d\u96c0\u7248\u672c\u9519\u8bef\u7387\u4f4e\u4f46\u4e1a\u52a1\u6307\u6807\u4e0b\u964d<\/td>\n\u65b0\u7248 Retriever \u8fd4\u56de\u4e86\u4e0d\u540c\u6392\u5e8f，\u5bfc\u81f4\u7b54\u6848\u8d28\u91cf\u53d8\u5316<\/td>\n\u91d1\u4e1d\u96c0\u76d1\u63a7\u540c\u65f6\u770b\u6280\u672f\u6307\u6807\u548c\u4e1a\u52a1\u6307\u6807<\/td>\n<\/tr>\n
    \u8de8\u673a\u623f\u67e5\u8be2\u5ef6\u8fdf\u9ad8\u8fbe 500ms<\/td>\n\u5317\u4eac\u7684\u8bf7\u6c42\u9519\u8bef\u8def\u7531\u5230\u4e86\u4e0a\u6d77（\u7ed5\u8def\u4e86）<\/td>\n\u68c0\u67e5 GeoDNS \u914d\u7f6e\u548c CDN \u8282\u70b9<\/td>\n<\/tr>\n
    \u707e\u96be\u6062\u590d\u6f14\u7ec3\u540e\u6570\u636e\u4e0d\u4e00\u81f4<\/td>\nMinIO mirror \u4e0d\u662f\u5b9e\u65f6\u540c\u6b65，\u6709\u5206\u949f\u7ea7\u5ef6\u8fdf<\/td>\n\u6f14\u7ec3\u540e\u6267\u884c\u6570\u636e\u4e00\u81f4\u6027\u6821\u9a8c\u811a\u672c<\/td>\n<\/tr>\n
    \u6df7\u6c8c\u5b9e\u9a8c\u89e6\u53d1\u771f\u5b9e P0 \u544a\u8b66<\/td>\n\u544a\u8b66\u6ca1\u6709\u533a\u5206"\u6df7\u6c8c\u5b9e\u9a8c"\u548c"\u771f\u5b9e\u6545\u969c"<\/td>\n\u6df7\u6c8c\u5b9e\u9a8c\u5f00\u59cb\u65f6\u5728\u544a\u8b66\u7cfb\u7edf\u4e2d\u6807\u8bb0\u7ef4\u62a4\u7a97\u53e3<\/td>\n<\/tr>\n
    LLM \u8d39\u7528\u65e5\u9884\u7b97\u9891\u7e41\u89e6\u53d1<\/td>\n\u67d0\u90e8\u95e8\u7684 Prompt \u4e2d\u5305\u542b\u4e86"\u8bf7\u8be6\u7ec6\u56de\u7b54，\u4e0d\u5c11\u4e8e 2000 \u5b57"<\/td>\n\u5ba1\u8ba1 Prompt \u957f\u5ea6，\u8bbe\u7f6e max_tokens \u786c\u9650\u5236<\/td>\n<\/tr>\n
    \u7070\u5ea6\u53d1\u5e03\u540e\u65e7 Pod \u672a\u5b8c\u5168\u7ec8\u6b62<\/td>\nHPA \u5728\u7070\u5ea6\u671f\u95f4\u6269\u5bb9\u4e86\u65e7\u7248\u672c Pod<\/td>\n\u7070\u5ea6\u671f\u95f4\u6682\u505c HPA，\u6216\u5bf9\u65e7\u7248\u672c Deployment \u8bbe\u7f6e replicas=0<\/td>\n<\/tr>\n
    SLO \u8fbe\u6807\u4f46\u7528\u6237\u6295\u8bc9"\u592a\u6162"<\/td>\nP95 < 8s \u4f46\u90e8\u5206\u957f\u5c3e\u7528\u6237（P99+）\u53ef\u80fd\u7b49\u5230 20s+<\/td>\n\u589e\u52a0 P99 \u76d1\u63a7\u6307\u6807，\u5bf9\u957f\u5c3e\u8bf7\u6c42\u505a\u8d85\u65f6\u964d\u7ea7<\/td>\n<\/tr>\n
    \u6269\u5c55\u6ce8\u518c\u4e2d\u5fc3\u7684\u7248\u672c\u68c0\u67e5\u8bef\u62a5<\/td>\n\u7248\u672c\u53f7\u6bd4\u8f83\u903b\u8f91\u8fc7\u4e8e\u4e25\u683c（\u5982 \u201c>=0.26\u201d \u88ab\u89e3\u91ca\u4e3a \u201c\u4ec5 0.26\u201d）<\/td>\n\u4f7f\u7528 semver \u5e93\u505a\u8bed\u4e49\u5316\u7248\u672c\u6bd4\u8f83<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

    \u5e73\u53f0\u6301\u7eed\u6f14\u8fdb\u7684\u5e74\u5ea6\u8def\u7ebf\u56fe<\/h4>\n

    \u672c\u7ae0\u662f\u5168\u4e13\u680f\u7684\u6536\u5b98，\u4f46\u5e73\u53f0\u5efa\u8bbe\u6c38\u8fdc\u5728\u8def\u4e0a\u3002\u5efa\u8bae\u7684\u5e74\u5ea6\u6f14\u8fdb\u8def\u7ebf：<\/p>\n

    Q1（\u5f53\u524d\u5b63\u5ea6）：\u7a33\u5b9a
    \n – \u5b8c\u6210\u7070\u5ea6\u53d1\u5e03\u6d41\u6c34\u7ebf
    \n – \u5efa\u7acb SLO \u76d1\u63a7\u4e0e\u6708\u5ea6\u62a5\u544a
    \n – \u9996\u6b21\u6df7\u6c8c\u5b9e\u9a8c + \u4e8b\u6545\u6f14\u7ec3
    \n – \u5236\u5b9a LLM \u8d39\u7528\u9884\u7b97\u548c\u544a\u8b66<\/p>\n

    Q2：\u6269\u5c55
    \n – \u65b0\u63a5 5 \u4e2a\u4e1a\u52a1\u90e8\u95e8
    \n – \u4e0a\u7ebf\u4f01\u4e1a\u6269\u5c55\u5e02\u573a
    \n – \u591a\u673a\u623f\u5199\u5165\u80fd\u529b\u8bd5\u70b9（\u53cc\u4e3b\u65b9\u6848）
    \n – \u5f15\u5165 LLM \u7f13\u5b58\u5c42（\u76f8\u540c\u95ee\u9898\u76f4\u63a5\u8fd4\u56de\u7f13\u5b58\u7ed3\u679c，\u964d\u4f4e\u6210\u672c 30%+）<\/p>\n

    Q3：\u4f18\u5316
    \n – \u5343\u4e07\u7ea7 Chunk \u68c0\u7d22\u6027\u80fd\u4f18\u5316（\u51b7\u70ed\u5206\u79bb \/ \u5411\u91cf\u91cf\u5316）
    \n – \u81ea\u7814 embedding \u6a21\u578b\u5fae\u8c03（\u57fa\u4e8e\u516c\u53f8\u6587\u6863\u9886\u57df\u6570\u636e）
    \n – \u5168\u94fe\u8def P99 < 5s
    \n – \u65e5\u6210\u672c\u964d\u4f4e 20%（\u901a\u8fc7\u7f13\u5b58 + Prompt \u4f18\u5316 + \u6a21\u578b\u964d\u7ea7）<\/p>\n

    Q4：\u667a\u80fd
    \n – \u7528\u6237\u884c\u4e3a\u5206\u6790\u9a71\u52a8\u7684\u81ea\u52a8\u4f18\u5316（\u9ad8\u9891\u95ee\u9898\u7f13\u5b58\u3001\u81ea\u52a8\u53d1\u73b0\u4f4e\u8d28\u91cf\u5207\u7247）
    \n – \u591a\u6a21\u6001 RAG（\u56fe\u7247+\u8868\u683c+\u89c6\u9891\u5b57\u5e55\u7684\u7edf\u4e00\u68c0\u7d22）
    \n – Agent \u81ea\u4e3b\u51b3\u7b56\u7684\u590d\u6742\u5de5\u4f5c\u6d41（\u96f6\u4eba\u5de5\u5e72\u9884\u7684\u62a5\u9500\u5ba1\u6279）
    \n – \u5e73\u53f0 AIOps（\u81ea\u52a8\u6839\u56e0\u5206\u6790\u3001\u81ea\u52a8\u6269\u5bb9\u3001\u81ea\u52a8\u964d\u7ea7）<\/p>\n

    \u6b65\u9aa48：SLO \u5b9a\u4e49\u4e0e\u76d1\u63a7\u2014\u2014\u5e73\u53f0\u53ef\u7528\u6027\u627f\u8bfa<\/h5>\n

    \u76ee\u6807：\u5b9a\u4e49\u5e73\u53f0\u7ea7\u7684 SLO，\u5e76\u5efa\u7acb\u5b9e\u65f6\u76d1\u63a7\u4e0e\u6708\u5ea6\u62a5\u544a\u3002<\/p>\n

    # platform_slo.py – \u5e73\u53f0 SLO \u5b9a\u4e49\u4e0e\u76d1\u63a7<\/span>
    \nfrom<\/span> datetime import<\/span> datetime,<\/span> timedelta<\/p>\n

    class<\/span> PlatformSLO<\/span>:<\/span>
    \n """\u5e73\u53f0\u7ea7\u670d\u52a1\u6c34\u5e73\u76ee\u6807"""<\/span><\/p>\n

    # SLO \u5b9a\u4e49<\/span>
    \n SLOS =<\/span> {<\/span>
    \n "chat_availability"<\/span>:<\/span> {<\/span>
    \n "target"<\/span>:<\/span> 0.999<\/span>,<\/span> # 99.9% \u53ef\u7528\u6027<\/span>
    \n "window"<\/span>:<\/span> "30d"<\/span>,<\/span> # \u6eda\u52a8 30 \u5929\u7a97\u53e3<\/span>
    \n "description"<\/span>:<\/span> "\u95ee\u7b54 API \u6210\u529f\u7387"<\/span>,<\/span>
    \n }<\/span>,<\/span>
    \n "chat_p95_latency"<\/span>:<\/span> {<\/span>
    \n "target"<\/span>:<\/span> 8000<\/span>,<\/span> # P95 < 8 \u79d2<\/span>
    \n "window"<\/span>:<\/span> "7d"<\/span>,<\/span>
    \n "description"<\/span>:<\/span> "\u95ee\u7b54 P95 \u5ef6\u8fdf (ms)"<\/span>,<\/span>
    \n }<\/span>,<\/span>
    \n "parse_success_rate"<\/span>:<\/span> {<\/span>
    \n "target"<\/span>:<\/span> 0.99<\/span>,<\/span> # 99% \u89e3\u6790\u6210\u529f\u7387<\/span>
    \n "window"<\/span>:<\/span> "7d"<\/span>,<\/span>
    \n "description"<\/span>:<\/span> "\u6587\u6863\u89e3\u6790\u6210\u529f\u7387"<\/span>,<\/span>
    \n }<\/span>,<\/span>
    \n "search_p95_latency"<\/span>:<\/span> {<\/span>
    \n "target"<\/span>:<\/span> 500<\/span>,<\/span> # P95 < 500ms（\u542b Rerank）<\/span>
    \n "window"<\/span>:<\/span> "7d"<\/span>,<\/span>
    \n "description"<\/span>:<\/span> "\u68c0\u7d22 P95 \u5ef6\u8fdf（\u542b Rerank）(ms)"<\/span>,<\/span>
    \n }<\/span>,<\/span>
    \n }<\/span><\/p>\n

    @classmethod<\/span>
    \n def<\/span> calculate_compliance<\/span>(<\/span>cls,<\/span> metrics)<\/span>:<\/span>
    \n """\u8ba1\u7b97 SLO \u8fbe\u6807\u60c5\u51b5"""<\/span>
    \n report =<\/span> {<\/span>}<\/span>
    \n for<\/span> slo_name,<\/span> slo_def in<\/span> cls.<\/span>SLOS.<\/span>items(<\/span>)<\/span>:<\/span>
    \n actual =<\/span> metrics.<\/span>get(<\/span>slo_name,<\/span> 0<\/span>)<\/span>
    \n target =<\/span> slo_def[<\/span>"target"<\/span>]<\/span>
    \n compliant =<\/span> actual <=<\/span> target if<\/span> "latency"<\/span> in<\/span> slo_name else<\/span> actual >=<\/span> target
    \n burn_rate =<\/span> cls.<\/span>_calculate_burn_rate(<\/span>actual,<\/span> target)<\/span><\/p>\n

    report[<\/span>slo_name]<\/span> =<\/span> {<\/span>
    \n "target"<\/span>:<\/span> target,<\/span>
    \n "actual"<\/span>:<\/span> actual,<\/span>
    \n "compliant"<\/span>:<\/span> compliant,<\/span>
    \n "burn_rate"<\/span>:<\/span> burn_rate,<\/span> # \u9884\u7b97\u6d88\u8017\u901f\u7387<\/span>
    \n "remaining_budget"<\/span>:<\/span> 1<\/span> –<\/span> burn_rate,<\/span>
    \n "status"<\/span>:<\/span> "🟢"<\/span> if<\/span> compliant else<\/span> (<\/span>"🟡"<\/span> if<\/span> burn_rate <<\/span> 0.8<\/span> else<\/span> "🔴"<\/span>)<\/span>,<\/span>
    \n }<\/span>
    \n return<\/span> report<\/p>\n

    @classmethod<\/span>
    \n def<\/span> _calculate_burn_rate<\/span>(<\/span>cls,<\/span> actual,<\/span> target)<\/span>:<\/span>
    \n """\u8ba1\u7b97\u9519\u8bef\u9884\u7b97\u6d88\u8017\u901f\u7387"""<\/span>
    \n # \u5982\u679c SLO \u662f 99.9%，\u5141\u8bb8 0.1% \u7684\u9519\u8bef<\/span>
    \n # \u5982\u679c\u5f53\u524d\u9519\u8bef\u7387\u662f 0.15%，burn_rate = 0.15\/0.1 = 1.5（\u8d85\u9884\u7b97 50%）<\/span>
    \n if<\/span> isinstance<\/span>(<\/span>target,<\/span> float<\/span>)<\/span> and<\/span> target <<\/span> 1<\/span>:<\/span>
    \n error_budget =<\/span> 1<\/span> –<\/span> target
    \n current_error =<\/span> 1<\/span> –<\/span> actual
    \n return<\/span> current_error \/<\/span> max<\/span>(<\/span>error_budget,<\/span> 0.0001<\/span>)<\/span>
    \n return<\/span> 0<\/span><\/p>\n

    @classmethod<\/span>
    \n def<\/span> generate_monthly_slo_report<\/span>(<\/span>cls)<\/span>:<\/span>
    \n """\u751f\u6210\u6708\u5ea6 SLO \u62a5\u544a"""<\/span>
    \n metrics =<\/span> cls.<\/span>_fetch_30d_metrics(<\/span>)<\/span>
    \n report =<\/span> cls.<\/span>calculate_compliance(<\/span>metrics)<\/span><\/p>\n

    print<\/span>(<\/span>"="<\/span> *<\/span> 70<\/span>)<\/span>
    \n print<\/span>(<\/span>f" RAGFlow \u5e73\u53f0 SLO \u6708\u5ea6\u62a5\u544a – <\/span>{<\/span>datetime.<\/span>now(<\/span>)<\/span>:<\/span>%Y\u5e74%m\u6708<\/span>}<\/span><\/span>"<\/span><\/span>)<\/span>
    \n print<\/span>(<\/span>"="<\/span> *<\/span> 70<\/span>)<\/span>
    \n print<\/span>(<\/span>f"<\/span>{<\/span>'SLO\u6307\u6807'<\/span>:<\/span><24<\/span>}<\/span><\/span> <\/span>{<\/span>'\u76ee\u6807'<\/span>:<\/span><12<\/span>}<\/span><\/span> <\/span>{<\/span>'\u5b9e\u9645'<\/span>:<\/span><12<\/span>}<\/span><\/span> <\/span>{<\/span>'\u72b6\u6001'<\/span>:<\/span><6<\/span>}<\/span><\/span> <\/span>{<\/span>'\u9884\u7b97\u6d88\u8017'<\/span>:<\/span><10<\/span>}<\/span><\/span>"<\/span><\/span>)<\/span>
    \n print<\/span>(<\/span>"-"<\/span> *<\/span> 70<\/span>)<\/span><\/p>\n

    all_compliant =<\/span> True<\/span>
    \n for<\/span> name,<\/span> data in<\/span> report.<\/span>items(<\/span>)<\/span>:<\/span>
    \n print<\/span>(<\/span>f"<\/span>{<\/span>data[<\/span>'description'<\/span>]<\/span>:<\/span><24<\/span>}<\/span><\/span> <\/span>{<\/span>data[<\/span>'target'<\/span>]<\/span>:<\/span><12<\/span>}<\/span><\/span> <\/span>{<\/span>data[<\/span>'actual'<\/span>]<\/span>:<\/span><12.4f<\/span>}<\/span><\/span> "<\/span><\/span>
    \n f"<\/span>{<\/span>data[<\/span>'status'<\/span>]<\/span>:<\/span><6<\/span>}<\/span><\/span> <\/span>{<\/span>data[<\/span>'burn_rate'<\/span>]<\/span>:<\/span>.0%<\/span>}<\/span><\/span>"<\/span><\/span>)<\/span>
    \n if<\/span> not<\/span> data[<\/span>'compliant'<\/span>]<\/span>:<\/span>
    \n all_compliant =<\/span> False<\/span><\/p>\n

    print<\/span>(<\/span>"-"<\/span> *<\/span> 70<\/span>)<\/span>
    \n print<\/span>(<\/span>f"\u603b\u4f53\u8fbe\u6807: <\/span>{<\/span>'\u2705 \u5168\u90e8\u8fbe\u6807'<\/span> if<\/span> all_compliant else<\/span> '\u274c \u5b58\u5728\u672a\u8fbe\u6807\u9879'<\/span>}<\/span><\/span>"<\/span><\/span>)<\/span>
    \n return<\/span> report<\/p>\n

    # \u96c6\u6210\u5230 Grafana \u76d1\u63a7<\/span>
    \n# \u5728 Grafana \u4e2d\u521b\u5efa SLO \u9762\u677f，\u663e\u793a\u6bcf\u4e2a SLO \u7684\u5b9e\u65f6\u8fbe\u6807\u72b6\u6001<\/span>
    \n# \u8bbe\u7f6e\u544a\u8b66：\u5f53 burn_rate > 1.0 \u4e14\u6301\u7eed 1 \u5c0f\u65f6 \u2192 \u89e6\u53d1 P1 \u544a\u8b66<\/span><\/p>\n

    \u6b65\u9aa49：\u6210\u672c\u76d1\u63a7\u4e0e\u4f18\u5316\u2014\u2014LLM \u8d39\u7528\u65e5\u9884\u7b97\u544a\u8b66<\/h5>\n

    \u76ee\u6807：\u5efa\u7acb LLM \u8d39\u7528\u7684\u65e5\u9884\u7b97\u76d1\u63a7，\u8d85\u652f\u81ea\u52a8\u544a\u8b66\u3002<\/p>\n

    # cost_monitor.py – LLM \u8d39\u7528\u76d1\u63a7<\/span>
    \nfrom<\/span> datetime import<\/span> datetime,<\/span> timedelta<\/p>\n

    class<\/span> CostMonitor<\/span>:<\/span>
    \n """LLM \u8d39\u7528\u65e5\u9884\u7b97\u76d1\u63a7"""<\/span><\/p>\n

    DAILY_BUDGETS =<\/span> {<\/span>
    \n "OpenAI"<\/span>:<\/span> 50<\/span>,<\/span> # $50\/\u5929<\/span>
    \n "DeepSeek"<\/span>:<\/span> 20<\/span>,<\/span> # $20\/\u5929<\/span>
    \n "YunFan"<\/span>:<\/span> 0<\/span>,<\/span> # \u81ea\u7814\u6a21\u578b\u514d\u8d39<\/span>
    \n "Ollama"<\/span>:<\/span> 0<\/span>,<\/span> # \u672c\u5730\u6a21\u578b\u514d\u8d39<\/span>
    \n }<\/span><\/p>\n

    ALERT_THRESHOLDS =<\/span> {<\/span>
    \n "warning"<\/span>:<\/span> 0.7<\/span>,<\/span> # 70% \u9884\u7b97 \u2192 \u63d0\u9192<\/span>
    \n "critical"<\/span>:<\/span> 0.9<\/span>,<\/span> # 90% \u9884\u7b97 \u2192 \u544a\u8b66<\/span>
    \n "block"<\/span>:<\/span> 1.0<\/span>,<\/span> # 100% \u9884\u7b97 \u2192 \u963b\u6b62（\u5207\u6362\u5230\u5907\u7528\u6a21\u578b）<\/span>
    \n }<\/span><\/p>\n

    def<\/span> __init__<\/span>(<\/span>self)<\/span>:<\/span>
    \n self.<\/span>today_costs =<\/span> defaultdict(<\/span>float<\/span>)<\/span>
    \n self.<\/span>today =<\/span> datetime.<\/span>now(<\/span>)<\/span>.<\/span>date(<\/span>)<\/span><\/p>\n

    def<\/span> record_call<\/span>(<\/span>self,<\/span> factory,<\/span> tokens_in,<\/span> tokens_out)<\/span>:<\/span>
    \n """\u8bb0\u5f55\u4e00\u6b21 LLM \u8c03\u7528\u53ca\u5176\u8d39\u7528"""<\/span>
    \n # \u7b80\u5316\u8ba1\u8d39\u6a21\u578b<\/span>
    \n cost_per_1k_input =<\/span> {<\/span>"OpenAI"<\/span>:<\/span> 0.01<\/span>,<\/span> "DeepSeek"<\/span>:<\/span> 0.001<\/span>,<\/span> "YunFan"<\/span>:<\/span> 0<\/span>,<\/span> "Ollama"<\/span>:<\/span> 0<\/span>}<\/span>
    \n cost_per_1k_output =<\/span> {<\/span>"OpenAI"<\/span>:<\/span> 0.03<\/span>,<\/span> "DeepSeek"<\/span>:<\/span> 0.002<\/span>,<\/span> "YunFan"<\/span>:<\/span> 0<\/span>,<\/span> "Ollama"<\/span>:<\/span> 0<\/span>}<\/span><\/p>\n

    cost =<\/span> (<\/span>tokens_in \/<\/span> 1000<\/span> *<\/span> cost_per_1k_input.<\/span>get(<\/span>factory,<\/span> 0<\/span>)<\/span> +<\/span>
    \n tokens_out \/<\/span> 1000<\/span> *<\/span> cost_per_1k_output.<\/span>get(<\/span>factory,<\/span> 0<\/span>)<\/span>)<\/span><\/p>\n

    self.<\/span>today_costs[<\/span>factory]<\/span> +=<\/span> cost<\/p>\n

    # \u68c0\u67e5\u9884\u7b97<\/span>
    \n budget =<\/span> self.<\/span>DAILY_BUDGETS.<\/span>get(<\/span>factory,<\/span> 10<\/span>)<\/span>
    \n spent_ratio =<\/span> self.<\/span>today_costs[<\/span>factory]<\/span> \/<\/span> max<\/span>(<\/span>budget,<\/span> 1<\/span>)<\/span><\/p>\n

    if<\/span> spent_ratio ><\/span> self.<\/span>ALERT_THRESHOLDS[<\/span>"critical"<\/span>]<\/span>:<\/span>
    \n self.<\/span>_send_alert(<\/span>factory,<\/span> spent_ratio,<\/span> self.<\/span>today_costs[<\/span>factory]<\/span>,<\/span> budget)<\/span>
    \n if<\/span> spent_ratio ><\/span> self.<\/span>ALERT_THRESHOLDS[<\/span>"block"<\/span>]<\/span>:<\/span>
    \n return<\/span> False<\/span> # \u963b\u6b62\u8c03\u7528<\/span>
    \n elif<\/span> spent_ratio ><\/span> self.<\/span>ALERT_THRESHOLDS[<\/span>"warning"<\/span>]<\/span>:<\/span>
    \n self.<\/span>_send_alert(<\/span>factory,<\/span> spent_ratio,<\/span> self.<\/span>today_costs[<\/span>factory]<\/span>,<\/span> budget,<\/span>
    \n severity=<\/span>"warning"<\/span>)<\/span><\/p>\n

    return<\/span> True<\/span><\/p>\n

    def<\/span> get_daily_report<\/span>(<\/span>self)<\/span>:<\/span>
    \n """\u751f\u6210\u6bcf\u65e5\u8d39\u7528\u62a5\u544a"""<\/span>
    \n print<\/span>(<\/span>f"\\\\n=== LLM \u8d39\u7528\u65e5\u62a5 (<\/span>{<\/span>self.<\/span>today}<\/span><\/span>) ==="<\/span><\/span>)<\/span>
    \n total =<\/span> sum<\/span>(<\/span>self.<\/span>today_costs.<\/span>values(<\/span>)<\/span>)<\/span>
    \n for<\/span> factory,<\/span> budget in<\/span> self.<\/span>DAILY_BUDGETS.<\/span>items(<\/span>)<\/span>:<\/span>
    \n cost =<\/span> self.<\/span>today_costs.<\/span>get(<\/span>factory,<\/span> 0<\/span>)<\/span>
    \n pct =<\/span> cost \/<\/span> max<\/span>(<\/span>budget,<\/span> 1<\/span>)<\/span> *<\/span> 100<\/span> if<\/span> budget ><\/span> 0<\/span> else<\/span> 0<\/span>
    \n bar =<\/span> "\u2588"<\/span> *<\/span> int<\/span>(<\/span>pct \/<\/span> 10<\/span>)<\/span> +<\/span> "\u2591"<\/span> *<\/span> (<\/span>10<\/span> –<\/span> int<\/span>(<\/span>pct \/<\/span> 10<\/span>)<\/span>)<\/span>
    \n print<\/span>(<\/span>f" <\/span>{<\/span>factory:<\/span><12<\/span>}<\/span><\/span> $<\/span>{<\/span>cost:<\/span>>8.2f<\/span>}<\/span><\/span> \/ $<\/span>{<\/span>budget:<\/span>>6<\/span>}<\/span><\/span> [<\/span>{<\/span>bar}<\/span><\/span>] <\/span>{<\/span>pct:<\/span>.0f<\/span>}<\/span><\/span>%"<\/span><\/span>)<\/span>
    \n print<\/span>(<\/span>f" <\/span>{<\/span>'\u603b\u8ba1'<\/span>:<\/span><12<\/span>}<\/span><\/span> $<\/span>{<\/span>total:<\/span>>8.2f<\/span>}<\/span><\/span>"<\/span><\/span>)<\/span><\/p>\n

    \u6b65\u9aa410：\u6269\u5c55\u7edf\u4e00\u7ba1\u7406\u2014\u2014\u4f01\u4e1a\u6269\u5c55\u5e02\u573a<\/h5>\n

    \u76ee\u6807：\u89e3\u51b3 5 \u4e2a\u90e8\u95e8\u81ea\u5b9a\u4e49\u6269\u5c55\u5206\u6563\u7ba1\u7406\u7684\u95ee\u9898\u3002<\/p>\n

    # extension_registry.py – \u4f01\u4e1a\u6269\u5c55\u6ce8\u518c\u4e2d\u5fc3<\/span>
    \nimport<\/span> json
    \nimport<\/span> hashlib
    \nfrom<\/span> pathlib import<\/span> Path<\/p>\n

    class<\/span> ExtensionRegistry<\/span>:<\/span>
    \n """\u4f01\u4e1a\u6269\u5c55\u7edf\u4e00\u6ce8\u518c\u4e2d\u5fc3"""<\/span><\/p>\n

    EXTENSIONS_DIR =<\/span> Path(<\/span>"\/opt\/ragflow\/extensions"<\/span>)<\/span>
    \n REGISTRY_FILE =<\/span> EXTENSIONS_DIR \/<\/span> "registry.json"<\/span><\/p>\n

    def<\/span> __init__<\/span>(<\/span>self)<\/span>:<\/span>
    \n self.<\/span>registry =<\/span> self.<\/span>_load_registry(<\/span>)<\/span><\/p>\n

    def<\/span> register_extension<\/span>(<\/span>self,<\/span> name,<\/span> ext_type,<\/span> module_path,<\/span> class_name,<\/span>
    \n version,<\/span> author,<\/span> description)<\/span>:<\/span>
    \n """\u6ce8\u518c\u4e00\u4e2a\u6269\u5c55"""<\/span>
    \n ext_id =<\/span> hashlib.<\/span>md5(<\/span>f"<\/span>{<\/span>name}<\/span><\/span>:<\/span>{<\/span>ext_type}<\/span><\/span>:<\/span>{<\/span>module_path}<\/span><\/span>"<\/span><\/span>.<\/span>encode(<\/span>)<\/span>)<\/span>.<\/span>hexdigest(<\/span>)<\/span>[<\/span>:<\/span>8<\/span>]<\/span><\/p>\n

    self.<\/span>registry[<\/span>"extensions"<\/span>]<\/span>[<\/span>ext_id]<\/span> =<\/span> {<\/span>
    \n "id"<\/span>:<\/span> ext_id,<\/span>
    \n "name"<\/span>:<\/span> name,<\/span>
    \n "type"<\/span>:<\/span> ext_type,<\/span> # parser \/ tool \/ component<\/span>
    \n "module_path"<\/span>:<\/span> module_path,<\/span>
    \n "class_name"<\/span>:<\/span> class_name,<\/span>
    \n "version"<\/span>:<\/span> version,<\/span>
    \n "author"<\/span>:<\/span> author,<\/span>
    \n "description"<\/span>:<\/span> description,<\/span>
    \n "registered_at"<\/span>:<\/span> datetime.<\/span>now(<\/span>)<\/span>.<\/span>isoformat(<\/span>)<\/span>,<\/span>
    \n "status"<\/span>:<\/span> "active"<\/span>,<\/span>
    \n "compatible_ragflow_versions"<\/span>:<\/span> [<\/span>">=0.26.0, <0.28.0"<\/span>]<\/span>,<\/span>
    \n }<\/span>
    \n self.<\/span>_save_registry(<\/span>)<\/span>
    \n return<\/span> ext_id<\/p>\n

    def<\/span> check_compatibility<\/span>(<\/span>self,<\/span> ragflow_version)<\/span>:<\/span>
    \n """\u68c0\u67e5\u6240\u6709\u6269\u5c55\u4e0e\u5f53\u524d RAGFlow \u7248\u672c\u7684\u517c\u5bb9\u6027"""<\/span>
    \n issues =<\/span> [<\/span>]<\/span>
    \n for<\/span> ext_id,<\/span> ext in<\/span> self.<\/span>registry[<\/span>"extensions"<\/span>]<\/span>.<\/span>items(<\/span>)<\/span>:<\/span>
    \n if<\/span> ext[<\/span>"status"<\/span>]<\/span> !=<\/span> "active"<\/span>:<\/span>
    \n continue<\/span>
    \n # \u7b80\u5316\u7248\u672c\u68c0\u67e5（\u5b9e\u9645\u5e94\u7528 semver \u5e93）<\/span>
    \n compatible =<\/span> any<\/span>(<\/span>
    \n check_version(<\/span>ragflow_version,<\/span> vc)<\/span>
    \n for<\/span> vc in<\/span> ext[<\/span>"compatible_ragflow_versions"<\/span>]<\/span>
    \n )<\/span>
    \n if<\/span> not<\/span> compatible:<\/span>
    \n issues.<\/span>append(<\/span>{<\/span>
    \n "extension"<\/span>:<\/span> ext[<\/span>"name"<\/span>]<\/span>,<\/span>
    \n "type"<\/span>:<\/span> ext[<\/span>"type"<\/span>]<\/span>,<\/span>
    \n "compatible_versions"<\/span>:<\/span> ext[<\/span>"compatible_ragflow_versions"<\/span>]<\/span>,<\/span>
    \n "current_version"<\/span>:<\/span> ragflow_version,<\/span>
    \n }<\/span>)<\/span><\/p>\n

    if<\/span> issues:<\/span>
    \n print<\/span>(<\/span>f"\u26a0 \u53d1\u73b0 <\/span>{<\/span>len<\/span>(<\/span>issues)<\/span>}<\/span><\/span> \u4e2a\u517c\u5bb9\u6027\u95ee\u9898:"<\/span><\/span>)<\/span>
    \n for<\/span> i in<\/span> issues:<\/span>
    \n print<\/span>(<\/span>f" – <\/span>{<\/span>i[<\/span>'extension'<\/span>]<\/span>}<\/span><\/span> (<\/span>{<\/span>i[<\/span>'type'<\/span>]<\/span>}<\/span><\/span>): "<\/span><\/span>
    \n f"\u517c\u5bb9 <\/span>{<\/span>i[<\/span>'compatible_versions'<\/span>]<\/span>}<\/span><\/span>, \u5f53\u524d <\/span>{<\/span>i[<\/span>'current_version'<\/span>]<\/span>}<\/span><\/span>"<\/span><\/span>)<\/span><\/p>\n

    return<\/span> issues<\/p>\n

    def<\/span> list_extensions<\/span>(<\/span>self,<\/span> ext_type=<\/span>None<\/span>)<\/span>:<\/span>
    \n """\u5217\u51fa\u6240\u6709\u5df2\u6ce8\u518c\u6269\u5c55"""<\/span>
    \n exts =<\/span> self.<\/span>registry[<\/span>"extensions"<\/span>]<\/span>.<\/span>values(<\/span>)<\/span>
    \n if<\/span> ext_type:<\/span>
    \n exts =<\/span> [<\/span>e for<\/span> e in<\/span> exts if<\/span> e[<\/span>"type"<\/span>]<\/span> ==<\/span> ext_type]<\/span><\/p>\n

    print<\/span>(<\/span>f"\\\\n=== \u4f01\u4e1a\u6269\u5c55\u6e05\u5355 (<\/span>{<\/span>len<\/span>(<\/span>exts)<\/span>}<\/span><\/span> \u4e2a) ==="<\/span><\/span>)<\/span>
    \n print<\/span>(<\/span>f"<\/span>{<\/span>'\u540d\u79f0'<\/span>:<\/span><20<\/span>}<\/span><\/span> <\/span>{<\/span>'\u7c7b\u578b'<\/span>:<\/span><10<\/span>}<\/span><\/span> <\/span>{<\/span>'\u7248\u672c'<\/span>:<\/span><8<\/span>}<\/span><\/span> <\/span>{<\/span>'\u72b6\u6001'<\/span>:<\/span><8<\/span>}<\/span><\/span> <\/span>{<\/span>'\u4f5c\u8005'<\/span>:<\/span><12<\/span>}<\/span><\/span>"<\/span><\/span>)<\/span>
    \n print<\/span>(<\/span>"-"<\/span> *<\/span> 60<\/span>)<\/span>
    \n for<\/span> ext in<\/span> exts:<\/span>
    \n print<\/span>(<\/span>f"<\/span>{<\/span>ext[<\/span>'name'<\/span>]<\/span>:<\/span><20<\/span>}<\/span><\/span> <\/span>{<\/span>ext[<\/span>'type'<\/span>]<\/span>:<\/span><10<\/span>}<\/span><\/span> <\/span>{<\/span>ext[<\/span>'version'<\/span>]<\/span>:<\/span><8<\/span>}<\/span><\/span> "<\/span><\/span>
    \n f"<\/span>{<\/span>ext[<\/span>'status'<\/span>]<\/span>:<\/span><8<\/span>}<\/span><\/span> <\/span>{<\/span>ext[<\/span>'author'<\/span>]<\/span>:<\/span><12<\/span>}<\/span><\/span>"<\/span><\/span>)<\/span><\/p>\n

    \u6b65\u9aa411：\u5168\u94fe\u8def\u538b\u6d4b\u62a5\u544a\u2014\u2014\u9a8c\u8bc1\u5e73\u53f0\u5bb9\u91cf<\/h5>\n

    \u76ee\u6807：\u57fa\u4e8e\u7b2c28\u7ae0\u77e5\u8bc6，\u5bf9\u751f\u4ea7\u7ea7\u5e73\u53f0\u505a\u5168\u94fe\u8def\u538b\u6d4b\u3002<\/p>\n

    # full_chain_benchmark.sh – \u5168\u94fe\u8def\u538b\u6d4b\u811a\u672c<\/span>
    \n#!\/bin\/bash<\/span>
    \necho<\/span> "=== RAGFlow \u5168\u94fe\u8def\u538b\u6d4b ==="<\/span>
    \necho<\/span> "\u573a\u666f: 2000 \u5e76\u53d1\u7528\u6237, \u5317\u4eac+\u4e0a\u6d77\u53cc\u673a\u623f"<\/span>
    \necho<\/span> ""<\/span><\/p>\n

    # \u6d4b\u8bd5\u77e9\u9635<\/span>
    \nUSERS<\/span>=<\/span>(<\/span>100<\/span> 500<\/span> 1000<\/span> 2000<\/span>)<\/span>
    \nDURATION<\/span>=<\/span>300<\/span> # \u6bcf\u8f6e 5 \u5206\u949f<\/span><\/p>\n

    for<\/span> users<\/span> in<\/span> "${USERS[<\/span>@]<\/span>}<\/span>"<\/span>;<\/span> do<\/span>
    \n echo<\/span> "— \u5e76\u53d1: $users<\/span> —"<\/span><\/p>\n

    # \u542f\u52a8 locust (\u5206\u5e03\u5f0f\u6a21\u5f0f, \u5317\u4eac+\u4e0a\u6d77\u5404\u4e00\u4e2a worker)<\/span>
    \n locust -f<\/span> locustfile.py \\\\<\/span>
    \n –headless<\/span> \\\\<\/span>
    \n –users<\/span> $users<\/span> \\\\<\/span>
    \n –spawn-rate 50<\/span> \\\\<\/span>
    \n –run-time ${DURATION}<\/span>s \\\\<\/span>
    \n –master-host locust-master \\\\<\/span>
    \n –host<\/span> http:\/\/ragflow-api:9380 \\\\<\/span>
    \n –csv<\/span> "benchmark_${users}<\/span>users"<\/span><\/p>\n

    # \u7b49\u5f85\u7a33\u5b9a<\/span>
    \n sleep<\/span> 30<\/span><\/p>\n

    # \u6536\u96c6\u6307\u6807<\/span>
    \n python3 collect_metrics.py –users<\/span> $users<\/span> –output<\/span> "metrics_${users}<\/span>users.json"<\/span><\/p>\n

    echo<\/span> " P50: $(<\/span>jq '.chat_latency.p50'<\/span> metrics_${<\/span>users}<\/span>users.json)<\/span><\/span>ms"<\/span>
    \n echo<\/span> " P95: $(<\/span>jq '.chat_latency.p95'<\/span> metrics_${<\/span>users}<\/span>users.json)<\/span><\/span>ms"<\/span>
    \n echo<\/span> " P99: $(<\/span>jq '.chat_latency.p99'<\/span> metrics_${<\/span>users}<\/span>users.json)<\/span><\/span>ms"<\/span>
    \n echo<\/span> " \u6210\u529f\u7387: $(<\/span>jq '.success_rate'<\/span> metrics_${<\/span>users}<\/span>users.json)<\/span><\/span>"<\/span>
    \n echo<\/span> " HPA \u526f\u672c\u6570: $(<\/span>jq '.hpa_replicas'<\/span> metrics_${<\/span>users}<\/span>users.json)<\/span><\/span>"<\/span>
    \n echo<\/span> ""<\/span>
    \ndone<\/span><\/p>\n

    echo<\/span> "=== \u538b\u6d4b\u5b8c\u6210, \u62a5\u544a\u5df2\u4fdd\u5b58 ==="<\/span>
    \n# \u751f\u6210\u538b\u6d4b\u62a5\u544a<\/span>
    \npython3 generate_benchmark_report.py<\/p>\n

    \u601d\u8003\u9898<\/h4>\n
  • \n

    \u5f53\u524d\u5e73\u53f0\u7684\u591a\u673a\u623f\u65b9\u6848\u662f"\u5317\u4eac\u5199\u3001\u4e24\u5730\u8bfb"\u3002\u5982\u679c\u4e1a\u52a1\u8981\u6c42\u4e24\u5730\u90fd\u80fd\u5199（\u5982\u6587\u6863\u53ef\u4ee5\u5728\u4e0a\u6d77\u4e0a\u4f20），\u8be5\u5982\u4f55\u8bbe\u8ba1"\u591a\u4e3b\u5199\u5165 + \u51b2\u7a81\u89e3\u51b3"\u65b9\u6848？MySQL \u7684\u53cc\u4e3b\u590d\u5236\u6709\u4ec0\u4e48\u9677\u9631？<\/p>\n<\/li>\n

  • \n

    \u5343\u4e07\u7ea7 Chunk \u4e0b，Infinity \u7684\u68c0\u7d22 P95 \u5ef6\u8fdf\u53ef\u80fd\u4ece 50ms \u589e\u957f\u5230 500ms\u3002\u4f60\u5df2\u505a\u4e86\u5411\u91cf\u91cf\u5316\u3001\u7d22\u5f15\u5206\u7247\u3001\u67e5\u8be2\u7f13\u5b58\u7b49\u4f18\u5316\u2014\u2014\u4f46\u4ecd\u4e0d\u6ee1\u8db3 SLO（P95<200ms）\u3002\u662f\u5426\u5e94\u8be5\u8003\u8651"\u51b7\u70ed\u5206\u79bb"\u2014\u20146 \u4e2a\u6708\u524d\u7684\u6587\u6863\u81ea\u52a8\u964d\u7ea7\u5230\u6162\u5b58\u50a8（\u5982\u672c\u5730 HD），\u4f46\u68c0\u7d22\u65f6\u53ef\u80fd\u67e5\u4e0d\u5230\u2014\u2014\u5982\u4f55\u5728 Prompt \u4e2d\u544a\u77e5 LLM \u68c0\u7d22\u7ed3\u679c\u4e0d\u5b8c\u6574？<\/p>\n<\/li>\n

    （\u7b54\u6848\u63d0\u793a\u89c1\u9644\u5f55 D\u3002）<\/p>\n

    \u5ef6\u4f38\u9605\u8bfb\u4e0e\u8d44\u6e90<\/h3>\n

    10\u500d\u5f00\u53d1\u8005\u7684 Dify \u9b54\u6cd5\u4e66：\u4ece\u96f6\u6784\u5efa\u5168\u6808 AI \u5e94\u7528 \u540e\u7aef\u5de5\u7a0b\u5e08\u8f6c\u578bAI\u7b2c\u4e00\u8bfe-Ollama \u4e0e\u79c1\u6709\u5316\u5927\u6a21\u578b\u5b9e\u6218 \u5927\u578b\u8bed\u8a00\u6a21\u578b(LLM) vLLM \u9ad8\u6027\u80fd\u63a8\u7406\u843d\u5730\u5b9e\u6218 Agent\u5f00\u53d1\u4e4bLlamaIndex \u5b9e\u6218\u4fee\u70bc\u4e0e\u6e90\u7801\u8fdb\u9636 \u5927\u8bed\u8a00\u6a21\u578bTransformers \u5b9e\u6218\u4fee\u70bc\u4e0e\u6e90\u7801\u5256\u6790<\/p>\n","protected":false},"excerpt":{"rendered":"

    1 \u9879\u76ee\u80cc\u666f
    \n\u4e1a\u52a1\u573a\u666f
    \n\u300c\u4e91\u5e06\u79d1\u6280\u300d\u7684 RAGFlow \u5e73\u53f0\u7ecf\u8fc7\u4e00\u5e74\u7684\u8fd0\u8425，\u5df2\u4ece\u5355\u673a\u623f Docker Compose \u6f14\u8fdb\u5230\u591a\u673a\u623f K8s \u96c6\u7fa4，\u670d\u52a1 20 \u4e2a\u4e1a\u52a1\u90e8\u95e8\u30012000 \u540d\u5458\u5de5\u300120 \u4e07\u4efd\u6587\u6863\u30011000 \u4e07 Chunk\u3002\u968f\u7740\u89c4\u6a21\u589e\u957f\u548c\u67b6\u6784\u590d\u6742\u5ea6\u63d0\u5347，\u56e2\u961f\u9762\u4e34\u4e00\u7cfb\u5217\\”\u5e73\u53f0\u7ea7\\”\u6311\u6218：
    \n\u767e\u4e07\u7ea7\u6587\u6863\u7684\u68c0\u7d22\u5ef6\u8fdf\u4ece 500ms \u6da8\u5230 3 \u79d2\u2014\u2014\u9700\u8981\u5bf9 Dealer \u68c0\u7d22\u94fe\u8def\u505a\u6e90\u7801\u7ea7\u4f18\u5316\u3002\u81ea\u5b9a\u4e49\u6269\u5c55\u6563\u843d\u5404\u5904&#xff1<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[11243,11033,12908,12909,2268],"topic":[],"class_list":["post-113411","post","type-post","status-publish","format-standard","hentry","category-server","tag-slo","tag-11033","tag-12908","tag-12909","tag-ragflow"],"yoast_head":"\n\u7b2c40\u7ae0\uff1a\u3010\u9ad8\u7ea7\u7bc7\u7efc\u5408\u5b9e\u6218\u3011\u4ece\u6e90\u7801\u5230\u751f\u4ea7\u6253\u9020\u53ef\u89c2\u6d4b\u9ad8\u53ef\u7528 RAGFlow \u5e73\u53f0 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.wsisp.com\/helps\/113411.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"\u7b2c40\u7ae0\uff1a\u3010\u9ad8\u7ea7\u7bc7\u7efc\u5408\u5b9e\u6218\u3011\u4ece\u6e90\u7801\u5230\u751f\u4ea7\u6253\u9020\u53ef\u89c2\u6d4b\u9ad8\u53ef\u7528 RAGFlow \u5e73\u53f0 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"1 \u9879\u76ee\u80cc\u666f \u4e1a\u52a1\u573a\u666f \u300c\u4e91\u5e06\u79d1\u6280\u300d\u7684 RAGFlow \u5e73\u53f0\u7ecf\u8fc7\u4e00\u5e74\u7684\u8fd0\u8425，\u5df2\u4ece\u5355\u673a\u623f Docker Compose \u6f14\u8fdb\u5230\u591a\u673a\u623f K8s \u96c6\u7fa4，\u670d\u52a1 20 \u4e2a\u4e1a\u52a1\u90e8\u95e8\u30012000 \u540d\u5458\u5de5\u300120 \u4e07\u4efd\u6587\u6863\u30011000 \u4e07 Chunk\u3002\u968f\u7740\u89c4\u6a21\u589e\u957f\u548c\u67b6\u6784\u590d\u6742\u5ea6\u63d0\u5347，\u56e2\u961f\u9762\u4e34\u4e00\u7cfb\u5217"\u5e73\u53f0\u7ea7"\u6311\u6218： \u767e\u4e07\u7ea7\u6587\u6863\u7684\u68c0\u7d22\u5ef6\u8fdf\u4ece 500ms \u6da8\u5230 3 \u79d2\u2014\u2014\u9700\u8981\u5bf9 Dealer \u68c0\u7d22\u94fe\u8def\u505a\u6e90\u7801\u7ea7\u4f18\u5316\u3002\u81ea\u5b9a\u4e49\u6269\u5c55\u6563\u843d\u5404\u5904&#xff1\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/113411.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2026-10-06T01:58:09+00:00\" \/>\n<meta name=\"author\" content=\"admin\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"admin\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"13 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/113411.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/113411.html\",\"name\":\"\u7b2c40\u7ae0\uff1a\u3010\u9ad8\u7ea7\u7bc7\u7efc\u5408\u5b9e\u6218\u3011\u4ece\u6e90\u7801\u5230\u751f\u4ea7\u6253\u9020\u53ef\u89c2\u6d4b\u9ad8\u53ef\u7528 RAGFlow \u5e73\u53f0 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2026-10-06T01:58:09+00:00\",\"dateModified\":\"2026-10-06T01:58:09+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/113411.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/113411.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/113411.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"\u7b2c40\u7ae0\uff1a\u3010\u9ad8\u7ea7\u7bc7\u7efc\u5408\u5b9e\u6218\u3011\u4ece\u6e90\u7801\u5230\u751f\u4ea7\u6253\u9020\u53ef\u89c2\u6d4b\u9ad8\u53ef\u7528 RAGFlow \u5e73\u53f0\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\",\"url\":\"https:\/\/www.wsisp.com\/helps\/\",\"name\":\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"description\":\"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\",\"name\":\"admin\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"contentUrl\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"caption\":\"admin\"},\"sameAs\":[\"http:\/\/wp.wsisp.com\"],\"url\":\"https:\/\/www.wsisp.com\/helps\/author\/admin\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"\u7b2c40\u7ae0\uff1a\u3010\u9ad8\u7ea7\u7bc7\u7efc\u5408\u5b9e\u6218\u3011\u4ece\u6e90\u7801\u5230\u751f\u4ea7\u6253\u9020\u53ef\u89c2\u6d4b\u9ad8\u53ef\u7528 RAGFlow \u5e73\u53f0 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.wsisp.com\/helps\/113411.html","og_locale":"zh_CN","og_type":"article","og_title":"\u7b2c40\u7ae0\uff1a\u3010\u9ad8\u7ea7\u7bc7\u7efc\u5408\u5b9e\u6218\u3011\u4ece\u6e90\u7801\u5230\u751f\u4ea7\u6253\u9020\u53ef\u89c2\u6d4b\u9ad8\u53ef\u7528 RAGFlow \u5e73\u53f0 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"1 \u9879\u76ee\u80cc\u666f \u4e1a\u52a1\u573a\u666f \u300c\u4e91\u5e06\u79d1\u6280\u300d\u7684 RAGFlow \u5e73\u53f0\u7ecf\u8fc7\u4e00\u5e74\u7684\u8fd0\u8425，\u5df2\u4ece\u5355\u673a\u623f Docker Compose \u6f14\u8fdb\u5230\u591a\u673a\u623f K8s \u96c6\u7fa4，\u670d\u52a1 20 \u4e2a\u4e1a\u52a1\u90e8\u95e8\u30012000 \u540d\u5458\u5de5\u300120 \u4e07\u4efd\u6587\u6863\u30011000 \u4e07 Chunk\u3002\u968f\u7740\u89c4\u6a21\u589e\u957f\u548c\u67b6\u6784\u590d\u6742\u5ea6\u63d0\u5347，\u56e2\u961f\u9762\u4e34\u4e00\u7cfb\u5217\"\u5e73\u53f0\u7ea7\"\u6311\u6218： \u767e\u4e07\u7ea7\u6587\u6863\u7684\u68c0\u7d22\u5ef6\u8fdf\u4ece 500ms \u6da8\u5230 3 \u79d2\u2014\u2014\u9700\u8981\u5bf9 Dealer \u68c0\u7d22\u94fe\u8def\u505a\u6e90\u7801\u7ea7\u4f18\u5316\u3002\u81ea\u5b9a\u4e49\u6269\u5c55\u6563\u843d\u5404\u5904&#xff1","og_url":"https:\/\/www.wsisp.com\/helps\/113411.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2026-10-06T01:58:09+00:00","author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"13 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/113411.html","url":"https:\/\/www.wsisp.com\/helps\/113411.html","name":"\u7b2c40\u7ae0\uff1a\u3010\u9ad8\u7ea7\u7bc7\u7efc\u5408\u5b9e\u6218\u3011\u4ece\u6e90\u7801\u5230\u751f\u4ea7\u6253\u9020\u53ef\u89c2\u6d4b\u9ad8\u53ef\u7528 RAGFlow \u5e73\u53f0 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2026-10-06T01:58:09+00:00","dateModified":"2026-10-06T01:58:09+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/113411.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/113411.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/113411.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"\u7b2c40\u7ae0\uff1a\u3010\u9ad8\u7ea7\u7bc7\u7efc\u5408\u5b9e\u6218\u3011\u4ece\u6e90\u7801\u5230\u751f\u4ea7\u6253\u9020\u53ef\u89c2\u6d4b\u9ad8\u53ef\u7528 RAGFlow \u5e73\u53f0"}]},{"@type":"WebSite","@id":"https:\/\/www.wsisp.com\/helps\/#website","url":"https:\/\/www.wsisp.com\/helps\/","name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","description":"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41","name":"admin","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/","url":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","contentUrl":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","caption":"admin"},"sameAs":["http:\/\/wp.wsisp.com"],"url":"https:\/\/www.wsisp.com\/helps\/author\/admin"}]}},"_links":{"self":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/113411","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/comments?post=113411"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/113411\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=113411"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=113411"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=113411"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=113411"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}