[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"blog-strapi-article:zh:ai-api-latency-by-region-measurement":3,"blog-article:\u002Fblog\u002Fai-api-latency-by-region-measurement":1073,"blog-related:\u002Fblog\u002Fai-api-latency-by-region-measurement":1074,"blog-translations:\u002Fblog\u002Fai-api-latency-by-region-measurement":1075},{"available":4,"article":5,"relatedArticles":174,"translations":962},true,{"source":6,"id":7,"documentId":8,"translationKey":8,"locale":9,"slug":10,"path":11,"_path":11,"title":12,"description":13,"excerpt":13,"image":14,"publishedAt":15,"updatedAt":16,"author":17,"tags":18,"featured":19,"sortOrder":20,"readingTime":21,"bodyType":22,"bodyBlocks":23,"bodyHtml":173},"strapi",3932,"pyz67ezhnjjjyowo5nnv7aqa","zh","ai-api-latency-by-region-measurement","\u002Fblog\u002Fai-api-latency-by-region-measurement","如何测量不同地区经代理IP 访问 AI API 的实际延迟","比较不同地区访问 AI API 的延迟时，应先区分连接阶段、首响应和生成完成时间，避免把 TTFB 与其包含的阶段重复相加。文章说明计时点、冷暖与复用分组、出口识别边界，并给出逐请求记录和分位数统计口径。","https:\u002F\u002Fcms.quanqiudaili.com\u002Fuploads\u002Flarge_AI_API_cover_b818e4c729.webp","2026-11-04T06:48:52.455Z","2026-11-04T06:48:52.317Z","Global Proxies Team",[],false,0,2,"blocks",[24,30,35,39,43,47,51,64,68,72,76,80,84,88,100,104,108,118,122,126,130,134,138,142,149,153,159,163,169],{"type":25,"children":26},"paragraph",[27],{"text":28,"type":29},"同一 AI API 在不同地区的响应时间会受 DNS、网络连接、服务端排队和输出长度共同影响。比较前先定义计时点，分别记录客户端可见的端到端时间和能够取得的连接阶段数据；出口识别只说明所用地址的识别结果，不等同于目标服务实际调度或物理路径。","text",{"type":31,"level":21,"children":32},"heading",[33],{"text":34,"type":29},"AI API 的实际延迟应该从哪一段开始测",{"type":25,"children":36},[37],{"text":38,"type":29},"先记录一组单调时钟时间点，再计算相邻阶段差值：调用开始、DNS 完成、TCP 连接完成、代理隧道可用、TLS 完成、首个响应字节、首个内容事件（流式接口）和响应完成。工具不支持的时间点记为不可观测，不要补猜值。TTFB 若定义为请求发出到首个响应字节，已包含此前建立连接的时间，不能再与 DNS\u002FTCP\u002FTLS 相加；若只测请求发送后的等待时间，应使用不同字段名。TLS 终止点及计时器的累计\u002F差值口径必须注明。",{"type":25,"children":40},[41],{"text":42,"type":29},"非流式请求里，后两段通常合在一起表现为\"总时长\"，其中大部分由服务端生成决定；流式请求则要把首 token 时间和完整输出时间分开。只报总时长，会让\"模型慢\"和\"链路慢\"无法区分。",{"type":31,"level":21,"children":44},[45],{"text":46,"type":29},"为什么流式和非流式请求要分开测",{"type":25,"children":48},[49],{"text":50,"type":29},"两者测的不是同一件事：",{"type":52,"format":53,"children":54},"list","unordered",[55,60],{"type":56,"children":57},"list-item",[58],{"text":59,"type":29},"非流式：一次请求返回完整结果，总时长 = 排队 + 生成 + 传输。生成 token 数不同，时长可以差数倍，所以必须固定 `max_tokens` 和提示词。",{"type":56,"children":61},[62],{"text":63,"type":29},"流式：关注首 token 时间（体感上\"多久开始出字\"）和完成时间（\"多久出完\"）。首 token 是端到端指标，同时受排队、调度和生成影响；解释时要结合 DNS\u002FTCP\u002FTLS 各段、服务端时间戳和重复样本。",{"type":25,"children":65},[66],{"text":67,"type":29},"同一地区、同一提示词下，流式首内容时间与非流式完整响应时间不能互相替代。报告应分栏记录，并明确各计时点是否包含连接建立阶段。",{"type":31,"level":21,"children":69},[70],{"text":71,"type":29},"如何预热请求并排除冷启动样本",{"type":25,"children":73},[74],{"text":75,"type":29},"冷态与暖态回答的是不同问题，应分别报告；连接新建和连接复用也要分组。3–5 次可作为试跑示例，不能保证连接、DNS 缓存或模型已预热。若只研究暖态，应依据连接复用、缓存命中等实际事件定义排除规则，同时保留冷态结果或说明其未纳入的理由。各组固定提示词、参数和采样时间窗，并记录请求间隔与限流状态。",{"type":25,"children":77},[78],{"text":79,"type":29},"固定请求还有一个容易忽略的点：随机性。温度不为 0 时输出长度会波动，间接影响总时长，所以比对链路时优先看首字节和首 token，而不是总时长。",{"type":31,"level":21,"children":81},[82],{"text":83,"type":29},"如何逐条确认请求使用了预期出口",{"type":25,"children":85},[86],{"text":87,"type":29},"每条请求记录配置地区、实际出口 IP、所用 IP 地理库及版本、识别结果，以及目标服务端可取得的调度或区域信息。IP 库结果不等同于物理路径或目标服务实际调度；ASN 也不能单独证明住宅或机房属性。同一测试出现多个出口时按出口分组，先说明配置与观测差异，再解释延迟结果。",{"type":25,"children":89},[90,92,98],{"text":91,"type":29},"连接阶段计时可参考 ",{"text":93,"type":29,"marks":94},"curl write-out 变量文档",[95],{"href":96,"type":97},"https:\u002F\u002Fcurl.se\u002Fdocs\u002Fmanpage.html#-w","link",{"text":99,"type":29},"；它只适用于 curl 能提供的指标，首内容事件仍需应用埋点。",{"type":31,"level":21,"children":101},[102],{"text":103,"type":29},"测量时应该记录哪些字段并如何统计",{"type":25,"children":105},[106],{"text":107,"type":29},"先按地区把原始记录填全，再谈统计口径。",{"type":52,"format":53,"children":109},[110,114],{"type":56,"children":111},[112],{"text":113,"type":29},"地区\u002F分组：A-暖态示例 ｜ 出口及识别来源：示例出口 \u002F 库版本待填 ｜ 连接新建\u002F复用：复用 ｜ DNS\u002FTCP\u002FTLS\u002F隧道（可观测差值）：按实际工具填 ｜ TTFB 或首内容时间：按实际埋点填 ｜ 总时长：按实际记录填 ｜ 状态：示例 ｜ 成功样本数：试跑样本 ｜ p50 \u002F p95 对应字段：明确指标后计算",{"type":56,"children":115},[116],{"text":117,"type":29},"地区\u002F分组：B-冷态示例 ｜ 出口及识别来源：示例出口 \u002F 库版本待填 ｜ 连接新建\u002F复用：新建 ｜ DNS\u002FTCP\u002FTLS\u002F隧道（可观测差值）：按实际工具填 ｜ TTFB 或首内容时间：按实际埋点填 ｜ 总时长：按实际记录填 ｜ 状态：示例 ｜ 成功样本数：试跑样本 ｜ p50 \u002F p95 对应字段：明确指标后计算",{"type":25,"children":119},[120],{"text":121,"type":29},"表格展示结构而非固定样本门槛。原始逐请求记录与地区汇总分开保存；每组报告成功、失败和超时数量及采样时间窗。按明确字段计算 p50\u002Fp95，并给出成功样本数；平均值可作为辅助统计，不替代分位数。429 与 5xx 单独分组，避免将限流或服务端错误误判为网络延迟。样本量按差异大小、波动程度、成功率精度及时间覆盖确定，不以达到某个示例数就判充分。",{"type":31,"level":21,"children":123},[124],{"text":125,"type":29},"如何解释测量结果并进行长期跟踪",{"type":25,"children":127},[128],{"text":129,"type":29},"一次完整测量记录时间窗、客户端版本、提示词与参数、配置地区、出口及识别来源、可观测计时点、状态码和各分组样本数。关键字段缺失时，将结论限定为本次观察，不概括为某地区普遍更快。",{"type":25,"children":131},[132],{"text":133,"type":29},"结论边界包括：只覆盖当前服务端环境与当前时段；模型版本变化后需重测；跨地区差异可能来自 DNS 调度而非物理路径。把这些写明，读者才知道数字能用到哪一步。",{"type":25,"children":135},[136],{"text":137,"type":29},"长期跟踪时保持计时定义、客户端版本和分组规则一致；连接阶段细节不足时，将其标为不可观测并结合服务端时间戳，不据端到端时长单独归因网络段。",{"type":31,"level":21,"children":139},[140],{"text":141,"type":29},"FAQ",{"type":25,"children":143},[144],{"text":145,"type":29,"marks":146},"Q：冷启动样本应该丢弃吗？",[147],{"type":148},"bold",{"type":25,"children":150},[151],{"text":152,"type":29},"A：不应默认丢弃。冷态和暖态代表不同使用场景，应分别统计；只有研究暖态时，才按连接复用或缓存事件定义排除规则，并保留排除前后的样本数。",{"type":25,"children":154},[155],{"text":156,"type":29,"marks":157},"Q：流式接口该看首 token 还是总时长？",[158],{"type":148},{"type":25,"children":160},[161],{"text":162,"type":29},"A：两个都记。首 token 是端到端指标，受链路、排队与生成共同影响；总时长更受输出长度影响，单看一个容易得出错误结论。",{"type":25,"children":164},[165],{"text":166,"type":29,"marks":167},"Q：p95 比平均延迟高很多，说明什么？",[168],{"type":148},{"type":25,"children":170},[171],{"text":172,"type":29},"A：说明尾部请求明显更慢。应同时检查样本数、状态码、重试和采样时段；平均值可辅助判断，但不能替代分位数或单独说明根因。","",[175,418,689],{"source":6,"id":176,"documentId":177,"translationKey":177,"locale":9,"slug":178,"path":179,"_path":179,"title":180,"description":181,"excerpt":181,"image":182,"publishedAt":183,"updatedAt":184,"author":17,"tags":185,"featured":19,"sortOrder":20,"readingTime":186,"bodyType":22,"bodyBlocks":187,"bodyHtml":173},3937,"eg98712cpvp1nwo9plj025g4","websocket-reconnection-session-recovery-testing","\u002Fblog\u002Fwebsocket-reconnection-session-recovery-testing","WebSocket 断线重连：代理IP 出口切换后如何验证会话恢复","客户端显示已重连，用户却丢了消息或看到重复下单，是实时应用常见的假恢复。文章按客户端正常关闭、网络中断与服务端重启三类断线分别制造中断，比较恢复耗时、丢失与重复条数，再用消息序号、去重窗口和业务状态核对，确认恢复的是业务而不只是连接。","https:\u002F\u002Fcms.quanqiudaili.com\u002Fuploads\u002Flarge_Web_Socket_cover_2052233c88.webp","2026-11-04T06:50:15.056Z","2026-11-04T06:50:15.027Z",[],3,[188,192,196,206,210,214,218,222,226,237,241,263,267,271,285,289,293,297,315,319,323,327,331,335,339,343,357,361,365,375,385,388,394,398,404,408,414],{"type":25,"children":189},[190],{"text":191,"type":29},"日志里出现\"重连成功\"，并不代表业务恢复完成：消息可能缺失、重复处理或状态未同步。应按受控断线场景检查重连与业务状态，并根据协议契约验证恢复位置。本篇不预设断线一定发生在代理出口，也不把单一关闭码当作根因。",{"type":31,"level":21,"children":193},[194],{"text":195,"type":29},"怎样判断连接恢复是否已经变成业务恢复",{"type":52,"format":53,"children":197},[198,202],{"type":56,"children":199},[200],{"text":201,"type":29},"连接恢复：新的 WebSocket 完成握手，收到 101，可以收发消息。",{"type":56,"children":203},[204],{"text":205,"type":29},"业务恢复：断线期间产生的消息被补齐或明确丢弃，恢复后不重复、不乱序，业务状态与服务端一致。",{"type":25,"children":207},[208],{"text":209,"type":29},"连接恢复是必要条件，不是充分条件。客户端重连后如果没有带上恢复凭据，服务端可能按游标、时间窗或自身策略补发部分或全部消息，也可能直接拒绝恢复；策略不同，出问题的形态也不同：全量重放会重复，拒绝恢复会缺失。具体按哪种策略，以协议文档和测试结果为准。",{"type":31,"level":21,"children":211},[212],{"text":213,"type":29},"如何分别制造三种可控断线",{"type":25,"children":215},[216],{"text":217,"type":29},"不同断线方式会触发不同的关闭码与恢复路径，必须分开测：",{"type":25,"children":219},[220],{"text":221,"type":29},"1. 客户端主动关闭：调用 `close(1000)`，按应用预定策略验证停止或重新连接及状态处理；不自动重连不必然表示故障。\n2. 网络层中断：在自建测试网络中丢弃连接、发送 RST 或停止客户端进程等，记录故障点和方向。若客户端报告 1006，将其视为本地未收到有效 Close 帧的异常状态，不预设所有中断都会触发该码。\n3. 服务端主动断开：服务端关闭连接或重启，验证重连后的鉴权与恢复流程。",{"type":25,"children":223},[224],{"text":225,"type":29},"在自建测试环境中制造中断，不对正常业务链路注入故障。先做少量试跑检查测试流程，再根据业务 SLA、风险和观测波动确定正式覆盖次数。断线时机可覆盖空闲期、消息传输中和鉴权刚完成时，并记录 TCP RST、丢包、进程退出、心跳超时等实际证据。",{"type":25,"children":227},[228,230,235],{"text":229,"type":29},"网络层中断的定位可参考",{"text":231,"type":29,"marks":232},"代理 IP 故障排查中心",[233],{"href":234,"type":97},"https:\u002F\u002Fhelp.quanqiudaili.com\u002Fdocs\u002FSmart-check",{"text":236,"type":29},"。",{"type":31,"level":21,"children":238},[239],{"text":240,"type":29},"重连策略应该观察哪些指标",{"type":52,"format":53,"children":242},[243,247,251,255,259],{"type":56,"children":244},[245],{"text":246,"type":29},"首次重连延迟：从检测到断线到发出重连请求的时间。",{"type":56,"children":248},[249],{"text":250,"type":29},"重连间隔序列：是否指数退避，是否带随机抖动。",{"type":56,"children":252},[253],{"text":254,"type":29},"重连次数上限：达到上限后是放弃还是降级。",{"type":56,"children":256},[257],{"text":258,"type":29},"鉴权是否重做：令牌过期时是否重新获取。",{"type":56,"children":260},[261],{"text":262,"type":29},"是否携带恢复凭据：resume token、最后一条消息的序号或事件 ID。",{"type":25,"children":264},[265],{"text":266,"type":29},"退避没有抖动时，大量客户端会在同一时刻重连，形成二次冲击。验证时观察间隔序列是否递增且带随机量。",{"type":31,"level":21,"children":268},[269],{"text":270,"type":29},"常见的三种会话恢复机制怎么验证",{"type":52,"format":53,"children":272},[273,277,281],{"type":56,"children":274},[275],{"text":276,"type":29},"服务端保留游标：客户端带上最后确认的序号，服务端从下一条开始推送。",{"type":56,"children":278},[279],{"text":280,"type":29},"客户端拉取历史：重连后主动请求断线时间窗内的消息，再切回实时流。",{"type":56,"children":282},[283],{"text":284,"type":29},"无状态重放：服务端按时间窗重放，客户端负责去重。",{"type":25,"children":286},[287],{"text":288,"type":29},"序号的单调范围可能限于某个分区、订阅或过滤条件；先确认序号作用域、客户端确认点、乱序窗口和过滤规则。三种机制按产品契约分别验证：游标是否被接受、历史拉取与实时流是否衔接、重放窗口是否被正确去重。缺少客户端 token 不代表服务端没有已保存的恢复位置。",{"type":31,"level":21,"children":290},[291],{"text":292,"type":29},"如何检查补发消息是否重复或乱序",{"type":25,"children":294},[295],{"text":296,"type":29},"恢复阶段最容易出现\"补多了\"和\"补乱了\"。校验方法：",{"type":52,"format":53,"children":298},[299,303,307,311],{"type":56,"children":300},[301],{"text":302,"type":29},"每条业务消息带唯一 ID 与单调递增序号。",{"type":56,"children":304},[305],{"text":306,"type":29},"客户端记录已处理 ID 的窗口，重复 ID 直接丢弃。",{"type":56,"children":308},[309],{"text":310,"type":29},"收到消息时按已确认的序号作用域检查连续性；出现缺口后，按协议选择缓冲等待、补拉历史或进入降级\u002F重新同步，不规定所有系统都必须立即停止消费。",{"type":56,"children":312},[313],{"text":314,"type":29},"分别记录重复接收、去重后重复产生的业务副作用、最终缺失条数，作为恢复质量指标。",{"type":25,"children":316},[317],{"text":318,"type":29},"序号不能只靠时间戳，同一毫秒内的多条消息会无法排序。服务端生成序号时保证单调，客户端只做校验与去重。",{"type":31,"level":21,"children":320},[321],{"text":322,"type":29},"如何确认重连后业务状态仍然一致",{"type":25,"children":324},[325],{"text":326,"type":29},"连接层和消息层都正常，业务仍可能不一致。核对三项：",{"type":25,"children":328},[329],{"text":330,"type":29},"1. 关键状态字段：账户余额、订单状态、会话归属是否与服务端一致。\n2. 未确认操作：断线前发出但未收到确认的请求，重连后是否重发、是否被服务端去重。\n3. 幂等性：重发的写操作是否只生效一次（去重机制见上一节）。",{"type":25,"children":332},[333],{"text":334,"type":29},"业务核对要读服务端的权威状态，不能只信客户端本地缓存。本地缓存与服务端不一致时，以服务端为准并记录差异。",{"type":31,"level":21,"children":336},[337],{"text":338,"type":29},"如何记录恢复时间并完成验收",{"type":25,"children":340},[341],{"text":342,"type":29},"恢复时间要拆成两个数：断线到新连接可用，断线到业务补齐完成。前者反映重连策略，后者反映恢复机制。只报前者会低估用户实际等待时间。",{"type":52,"format":53,"children":344},[345,349,353],{"type":56,"children":346},[347],{"text":348,"type":29},"断线类型：客户端 close(1000) ｜ 恢复耗时：1.2 s ｜ 丢失条数：0 ｜ 重复条数：0 ｜ 业务状态一致：是 ｜ 证据：序号连续、状态字段一致",{"type":56,"children":350},[351],{"text":352,"type":29},"断线类型：网络中断（1006） ｜ 恢复耗时：3.5 s ｜ 丢失条数：0 ｜ 重复条数：2 ｜ 业务状态一致：是 ｜ 证据：去重窗口命中、补齐完成",{"type":56,"children":354},[355],{"text":356,"type":29},"断线类型：服务端重启 ｜ 恢复耗时：8.0 s ｜ 丢失条数：0 ｜ 重复条数：0 ｜ 业务状态一致：是 ｜ 证据：重新鉴权、游标续传",{"type":25,"children":358},[359],{"text":360,"type":29},"表中数值仅为示例，实际按测试记录填写；网络中断示例中的 1006 仅表示该次客户端本地观察到异常关闭状态，不代表确定的根因。重复接收不必然是缺陷，需进一步核对去重后的业务副作用。",{"type":25,"children":362},[363],{"text":364,"type":29},"验收表包含断线类型、恢复耗时、最终缺失条数、重复接收数、重复业务副作用、状态一致性和证据位置。覆盖次数与阈值按业务 SLA、风险和测量波动事先确定；3 次或 5 次只能作为试跑示例，不能当成通用充分门槛。正常 `close(1000)` 应按产品预定策略验证停止或恢复，不把不自动重连直接判错。任一关键证据缺失，结论应标注“证据不足”或“观察到重连成功”，不能写“恢复完成”。",{"type":25,"children":366},[367,369,374],{"text":368,"type":29},"测试工具的接入参数可对照",{"text":370,"type":29,"marks":371},"代理设置帮助",[372],{"href":373,"type":97},"https:\u002F\u002Fhelp.quanqiudaili.com\u002Fdocs\u002Fproxy-Settings",{"text":236,"type":29},{"type":25,"children":376},[377,379,384],{"text":378,"type":29},"WebSocket 关闭状态和 Close 帧语义见 ",{"text":380,"type":29,"marks":381},"RFC 6455",[382],{"href":383,"type":97},"https:\u002F\u002Fwww.rfc-editor.org\u002Frfc\u002Frfc6455",{"text":236,"type":29},{"type":31,"level":21,"children":386},[387],{"text":141,"type":29},{"type":25,"children":389},[390],{"text":391,"type":29,"marks":392},"Q：重连成功但消息少了一段，问题出在哪？",[393],{"type":148},{"type":25,"children":395},[396],{"text":397,"type":29},"A：先对照客户端、服务端和网络事件确定可观测到的故障位置，再按协议检查恢复位置。恢复状态可能来自客户端凭据，也可能由服务端保存；没有客户端序号不能单独证明服务端无法续传。",{"type":25,"children":399},[400],{"text":401,"type":29,"marks":402},"Q：close code 1006 一定能复现吗？",[403],{"type":148},{"type":25,"children":405},[406],{"text":407,"type":29},"A：不保证。1006 是本地报告的异常关闭状态，表示没有收到有效 Close 帧；不同客户端实现和故障方式下的观测结果可能不同，需记录具体故障注入和客户端事件。",{"type":25,"children":409},[410],{"text":411,"type":29,"marks":412},"Q：重复消息一定要在客户端去重吗？",[413],{"type":148},{"type":25,"children":415},[416],{"text":417,"type":29},"A：取决于协议契约。服务端游标续传、时间窗重放和客户端去重的责任不同，应分别检查收到重复与重复业务副作用，不能只统计客户端收到的消息数。",{"source":6,"id":419,"documentId":420,"translationKey":420,"locale":9,"slug":421,"path":422,"_path":422,"title":423,"description":424,"excerpt":424,"image":425,"publishedAt":426,"updatedAt":427,"author":17,"tags":428,"featured":19,"sortOrder":20,"readingTime":186,"bodyType":22,"bodyBlocks":429,"bodyHtml":173},3936,"wiumgl5vcc6qclqx4xe26751","websocket-proxy-test-report-metrics","\u002Fblog\u002Fwebsocket-proxy-test-report-metrics","WebSocket 代理IP 测试报告应该记录哪些指标","一份只能写「连接成功、延迟正常」的 WebSocket 测试报告，问题不在结论而在字段。把报告拆成环境、连接、业务三层，逐项给出字段、单位、采集方式与用途，配合握手分段、断连码分布、重连次数与 RTT 分位数，形成可直接落表、可复核的模板。","https:\u002F\u002Fcms.quanqiudaili.com\u002Fuploads\u002Flarge_Web_Socket_cover_bc45bc1e13.webp","2026-11-04T06:50:00.175Z","2026-11-04T06:50:00.160Z",[],[430,434,438,452,456,460,464,482,486,499,503,525,529,533,555,559,563,567,601,605,609,613,631,635,639,643,647,656,659,665,669,675,679,685],{"type":25,"children":431},[432],{"text":433,"type":29},"一份只能写\"连接成功、延迟正常\"的 WebSocket 报告，问题在于缺少可复核字段。本文按环境、连接、业务三层列出报告项，并注明采集工具与可观测边界。浏览器标准 WebSocket 事件通常不能直接提供全部握手分段、原始响应头或每一跳耗时；拿不到的字段应标为 N\u002FA 并说明原因。",{"type":31,"level":21,"children":435},[436],{"text":437,"type":29},"WebSocket 代理测试报告应该分哪三层记录",{"type":52,"format":53,"children":439},[440,444,448],{"type":56,"children":441},[442],{"text":443,"type":29},"环境层：这次测试在什么条件下发生。地区、出口 IP 与 ASN、代理协议、客户端版本、测试时间窗。",{"type":56,"children":445},[446],{"text":447,"type":29},"连接层：连接本身发生了什么。握手各段耗时、会话寿命、关闭码、异常关闭原因。",{"type":56,"children":449},[450],{"text":451,"type":29},"业务层：连接之上传了什么。消息计数、往返时延、丢失与重复、顺序连续性、恢复耗时。",{"type":25,"children":453},[454],{"text":455,"type":29},"三层分开写的好处是定位快：业务层异常先看连接层是否断过，连接层异常再看环境层是否出口漂移。混在一起写，读者只能看到\"整体不好\"。",{"type":31,"level":21,"children":457},[458],{"text":459,"type":29},"握手和建连阶段要记录哪些字段",{"type":25,"children":461},[462],{"text":463,"type":29},"握手要分四段记录，不能只写总和：",{"type":52,"format":53,"children":465},[466,470,474,478],{"type":56,"children":467},[468],{"text":469,"type":29},"DNS 解析耗时（ms）。",{"type":56,"children":471},[472],{"text":473,"type":29},"TCP 建连耗时（ms）。",{"type":56,"children":475},[476],{"text":477,"type":29},"TLS 握手耗时（ms，wss 才有）。",{"type":56,"children":479},[480],{"text":481,"type":29},"Upgrade 往返耗时（ms）：请求发出到收到 `101 Switching Protocols`。",{"type":25,"children":483},[484],{"text":485,"type":29},"同时记录校验结果：`Sec-WebSocket-Accept` 是否与服务端计算值一致、协商的子协议与扩展是否与预期相同、响应头是否被中间设备改写。握手失败要记录失败阶段与错误码，而不是只记\"连接失败\"。",{"type":25,"children":487},[488,490,493,495,498],{"text":489,"type":29},"握手前的代理参数可对照",{"text":370,"type":29,"marks":491},[492],{"href":373,"type":97},{"text":494,"type":29},"；Upgrade 握手字段定义见 ",{"text":380,"type":29,"marks":496},[497],{"href":383,"type":97},{"text":236,"type":29},{"type":31,"level":21,"children":500},[501],{"text":502,"type":29},"连接持续多久、为什么断开要怎么记录",{"type":52,"format":53,"children":504},[505,509,513,517,521],{"type":56,"children":506},[507],{"text":508,"type":29},"会话观察时长（s）：从收到 101 到关闭或观察窗口结束；窗口结束仍在线单独计数，不当作自然断开。",{"type":56,"children":510},[511],{"text":512,"type":29},"关闭发起方：只有两端事件或网络日志能佐证时才填写；无法确认则记未知。",{"type":56,"children":514},[515],{"text":516,"type":29},"本地关闭事件码与实际 Close 帧分开记录。1006 是本地报告的异常关闭状态，不是线上发送的 Close 帧；正常码也不能单独证明业务正常。",{"type":56,"children":518},[519],{"text":520,"type":29},"异常关闭率：明确分子、分母和观察窗口；握手失败、已关闭连接与窗口结束仍在线的连接分别统计。",{"type":56,"children":522},[523],{"text":524,"type":29},"心跳配置：间隔与超时阈值（s），以及心跳是否被中间设备吞掉。",{"type":25,"children":526},[527],{"text":528,"type":29},"关闭事件码分布可帮助筛选样本，但不能仅凭 code 判定发起方或根因。1006 表示本地未观察到有效 Close 帧；需结合客户端、服务端或网络侧日志继续定位。",{"type":31,"level":21,"children":530},[531],{"text":532,"type":29},"重连过程应该记录哪些指标",{"type":52,"format":53,"children":534},[535,539,543,547,551],{"type":56,"children":536},[537],{"text":538,"type":29},"首次重连延迟（ms\u002Fs）：从检测到断线到发起重连。",{"type":56,"children":540},[541],{"text":542,"type":29},"重连次数与退避序列：是否按指数退避，是否有抖动。",{"type":56,"children":544},[545],{"text":546,"type":29},"重连成功率：成功建立新连接的比例。",{"type":56,"children":548},[549],{"text":550,"type":29},"恢复耗时：从断线到业务可用的时间。",{"type":56,"children":552},[553],{"text":554,"type":29},"是否携带恢复凭据：如 resume token、last sequence。",{"type":25,"children":556},[557],{"text":558,"type":29},"重连次数多不代表质量差，可能是服务端主动轮转；但没有退避、密集重连会把故障放大。报告应同时给次数与间隔分布。",{"type":31,"level":21,"children":560},[561],{"text":562,"type":29},"如何汇总三层字段并补充消息细节",{"type":25,"children":564},[565],{"text":566,"type":29},"下表把三层字段汇总成可直接落表的形式。",{"type":52,"format":53,"children":568},[569,573,577,581,585,589,593,597],{"type":56,"children":570},[571],{"text":572,"type":29},"字段组：握手 ｜ 字段：DNS \u002F TCP \u002F TLS \u002F Upgrade ｜ 单位：ms ｜ 采集方式：受控网络日志、协议工具或服务端\u002F代理日志；浏览器标准事件通常不可提供全部分段 ｜ 用途：定位可观测范围内的建连耗时",{"type":56,"children":574},[575],{"text":576,"type":29},"字段组：连接 ｜ 字段：观察时长、结束状态、仍在线数 ｜ 单位：s \u002F 次 ｜ 采集方式：连接事件与观察窗口 ｜ 用途：区分自然断开与测试截止",{"type":56,"children":578},[579],{"text":580,"type":29},"字段组：连接 ｜ 字段：本地关闭事件码、Close 帧收发、可证实的发起方 ｜ 单位：次数 ｜ 采集方式：客户端事件及可用两端日志；不可观测时记 N\u002FA ｜ 用途：作为定位线索，不单独归因",{"type":56,"children":582},[583],{"text":584,"type":29},"字段组：重连 ｜ 字段：重连次数与间隔 ｜ 单位：次 \u002F s ｜ 采集方式：重连逻辑埋点 ｜ 用途：评估退避是否合理",{"type":56,"children":586},[587],{"text":588,"type":29},"字段组：消息 ｜ 字段：收发计数 ｜ 单位：条 ｜ 采集方式：消息回调 ｜ 用途：核对是否丢消息",{"type":56,"children":590},[591],{"text":592,"type":29},"字段组：消息 ｜ 字段：RTT p50 \u002F p95 ｜ 单位：ms ｜ 采集方式：发送与回执时间戳 ｜ 用途：评估体感延迟",{"type":56,"children":594},[595],{"text":596,"type":29},"字段组：消息 ｜ 字段：序号缺口与重复数 ｜ 单位：条 ｜ 采集方式：序号连续性校验 ｜ 用途：判断顺序与去重",{"type":56,"children":598},[599],{"text":600,"type":29},"字段组：代理IP ｜ 字段：类型、出口和会话分配 ｜ 单位：枚举 \u002F 地址 \u002F ID ｜ 采集方式：供应商控制台、分配记录及受控回显 ｜ 用途：说明本次测试路径，不能单独证明物理路由",{"type":25,"children":602},[603],{"text":604,"type":29},"表外还要补两项：消息大小分布与发送频率，否则 RTT 无法横向比较。大消息和小消息的往返时间差异明显，混在一起算分位数会失真。",{"type":31,"level":21,"children":606},[607],{"text":608,"type":29},"地区和代理链路还要记录哪些信息",{"type":25,"children":610},[611],{"text":612,"type":29},"代理IP字段记录实际分配的产品类型、出口和会话变化；无法确认的属性应标为未知。",{"type":52,"format":53,"children":614},[615,619,623,627],{"type":56,"children":616},[617],{"text":618,"type":29},"地区：客户端所在地区，与出口地区分开写。",{"type":56,"children":620},[621],{"text":622,"type":29},"出口 IP 与 ASN：每条连接记录一次，用于验证地区标签。",{"type":56,"children":624},[625],{"text":626,"type":29},"代理协议与会话类型：HTTP、SOCKS5，短会话或长会话；品类按实际分配写动态或静态住宅 IP、动态或静态机房 IP，不写笼统的\"住宅或机房\"，并保留出口 IP 与分配时间。",{"type":56,"children":628},[629],{"text":630,"type":29},"链路跳数：若代理有多跳，记录跳数与各跳耗时。",{"type":25,"children":632},[633],{"text":634,"type":29},"出口 IP 是测试记录的重要字段，但需与服务端回显、代理分配记录和 DNS\u002F调度信息结合；它本身不能证明物理路径或目标服务调度。",{"type":31,"level":21,"children":636},[637],{"text":638,"type":29},"报告模板应该怎样组织和统计",{"type":25,"children":640},[641],{"text":642,"type":29},"模板按\"环境 → 连接 → 业务\"顺序排列，每部分先给汇总表，再列异常样本。统计时报告 p50、p95 和样本数，单次最快值不能代表整体表现。\n失败与超时单独计数，不混入成功样本的分位数。\n按配置地区、出口和时间窗分组；跨窗口比较时同时标明环境差异。",{"type":25,"children":644},[645],{"text":646,"type":29},"异常样本保留 request id、连接 ID、出口 IP、关闭事件、时间戳及可用证据。报告结论写明客户端版本、服务端环境、测试时长、采集工具和不可观测字段；详细握手数据需受控网络日志、协议工具或服务端\u002F代理日志支持。",{"type":25,"children":648},[649,651,654],{"text":650,"type":29},"连接异常样本可按",{"text":231,"type":29,"marks":652},[653],{"href":234,"type":97},{"text":655,"type":29},"的字段顺序补充。",{"type":31,"level":21,"children":657},[658],{"text":141,"type":29},{"type":25,"children":660},[661],{"text":662,"type":29,"marks":663},"Q：报告里必须记录出口 IP 吗？",[664],{"type":148},{"type":25,"children":666},[667],{"text":668,"type":29},"A：建议记录，并同时记代理IP类型。它是验证地区标签与判断出口漂移的重要直接证据之一，但要与服务端回显、代理记录一起看，缺失后地区结论只能靠推断。",{"type":25,"children":670},[671],{"text":672,"type":29,"marks":673},"Q：close code 只记 1006 够不够？",[674],{"type":148},{"type":25,"children":676},[677],{"text":678,"type":29},"A：不够。关闭码是协议层线索，不能单独判断发起方或根因。应同时记录本地事件、实际 Close 帧收发和可取得的两端日志，并明确异常关闭率的分子、分母和仍在线连接的处理方式。",{"type":25,"children":680},[681],{"text":682,"type":29,"marks":683},"Q：时延只报平均值可以吗？",[684],{"type":148},{"type":25,"children":686},[687],{"text":688,"type":29},"A：不建议。平均值会掩盖尾部延迟，应同时报 p50 与 p95，并注明样本量。",{"source":6,"id":690,"documentId":691,"translationKey":691,"locale":9,"slug":692,"path":693,"_path":693,"title":694,"description":695,"excerpt":695,"image":696,"publishedAt":697,"updatedAt":698,"author":17,"tags":699,"featured":19,"sortOrder":20,"readingTime":700,"bodyType":22,"bodyBlocks":701,"bodyHtml":173},3935,"apg9xmkn8f0bozwgyxntuhtj","python-test-proxy-environment-variable-isolation","\u002Fblog\u002Fpython-test-proxy-environment-variable-isolation","Python 自动化测试：代理IP 环境变量污染如何验证","同一份测试在开发机能过、到 CI 就失败，常常是残留的代理变量在捣乱。文章列出这些变量及其读取顺序，用 fixture 固定环境、请求前断言、回显出口 IP 证明链路，再给出清理检查与故意失败自检。","https:\u002F\u002Fcms.quanqiudaili.com\u002Fuploads\u002Flarge_Python_cover_58829f8376.webp","2026-11-04T06:49:42.792Z","2026-11-04T06:49:42.779Z",[],4,[702,706,710,720,724,728,732,736,754,758,762,766,770,774,782,786,790,794,798,802,812,816,820,846,850,854,858,872,876,880,884,898,902,917,921,925,929,932,938,942,948,952,958],{"type":25,"children":703},[704],{"text":705,"type":29},"同一份测试，在开发机上通过、在 CI 上失败，或者反过来——十有八九是环境里的代理变量在起作用。requests、urllib、httpx 等常见实现通常会读取环境变量，具体行为随库和版本而变；CI 镜像又常预置 `HTTP_PROXY`。判断\"有没有被污染\"不能靠肉眼扫代码，要靠在测试内部固定环境并断言出口。这些变量指向的代理IP，往往和用例约定的不是同一条。NO_PROXY 规则本身的行为验证属于另一个话题，本篇讲的是测试框架里的环境隔离。",{"type":31,"level":21,"children":707},[708],{"text":709,"type":29},"代理环境变量污染通常会表现为什么",{"type":52,"format":53,"children":711},[712,716],{"type":56,"children":713},[714],{"text":715,"type":29},"意外走代理：本应直连的请求，被环境变量送进了代理，出口 IP 与预期不符。",{"type":56,"children":717},[718],{"text":719,"type":29},"意外绕过代理：本应走代理的请求，被 `NO_PROXY` 或空值放行，直接连到了目标。",{"type":25,"children":721},[722],{"text":723,"type":29},"两种表现都会让断言\"通过得莫名其妙\"。前者常出现在本地缓存或 mock 场景，后者常出现在 CI 里变量拼写错误或作用域不对的场景。判定污染不看代码写了什么，看实际请求从哪里出去。",{"type":31,"level":21,"children":725},[726],{"text":727,"type":29},"哪些代理环境变量会生效，优先级如何判断",{"type":25,"children":729},[730],{"text":731,"type":29},"变量不是设了就生效，读取顺序、大小写和库的默认值都会改变结果；变量只描述地址和协议，不描述代理IP的身份，身份得另做出口校验。把\"默认\"当成待实测的假设：按库和版本分别验证，并补一个故意设错变量的负向用例。",{"type":31,"level":186,"children":733},[734],{"text":735,"type":29},"为什么要同时检查大小写变量和 ALL_PROXY",{"type":52,"format":53,"children":737},[738,742,746,750],{"type":56,"children":739},[740],{"text":741,"type":29},"变量：`HTTP_PROXY` \u002F `http_proxy` ｜ 作用范围：http 请求 ｜ Python 客户端是否读取：requests、urllib 默认读取 ｜ 测试中如何处理：fixture 内显式设置或清空",{"type":56,"children":743},[744],{"text":745,"type":29},"变量：`HTTPS_PROXY` \u002F `https_proxy` ｜ 作用范围：https 请求 ｜ Python 客户端是否读取：同上 ｜ 测试中如何处理：同上",{"type":56,"children":747},[748],{"text":749,"type":29},"变量：`ALL_PROXY` \u002F `all_proxy` ｜ 作用范围：未单独指定协议时 ｜ Python 客户端是否读取：requests 读取 ｜ 测试中如何处理：容易遗漏，需一并清理",{"type":56,"children":751},[752],{"text":753,"type":29},"变量：`NO_PROXY` \u002F `no_proxy` ｜ 作用范围：排除列表 ｜ Python 客户端是否读取：requests、urllib 读取 ｜ 测试中如何处理：边界用例单独覆盖",{"type":25,"children":755},[756],{"text":757,"type":29},"`ALL_PROXY` 最容易被漏：未单独指定协议时它接管全部请求，清理时不能只清 HTTP_PROXY 与 HTTPS_PROXY。",{"type":31,"level":186,"children":759},[760],{"text":761,"type":29},"环境变量大小写相同时应该相信哪一个",{"type":25,"children":763},[764],{"text":765,"type":29},"同名不同大小写时，取值顺序可能相反。requests 通过 `urllib.request.getproxies()` 取值，常见实现小写优先，但版本和平台不同结果会变，也有库只认大写。测试里不要依赖，统一由 fixture 显式设定，并在请求前断言生效值，再用实际出口校验。",{"type":31,"level":186,"children":767},[768],{"text":769,"type":29},"为什么没写代理的 requests 仍可能走代理",{"type":25,"children":771},[772],{"text":773,"type":29},"Python 的 `requests` 常见版本默认 `trust_env=True`，会读取环境配置；该开关还会影响 `.netrc` 等其他来源。需要隔离环境代理时，优先为测试创建专用 `Session` 并设置 `trust_env=False`，再显式配置测试所需的认证与 CA。这样也会关闭该 Session 的其他环境来源，需按实际需求补配。不要把只对 `http`、`https` 设置 `None` 当成覆盖所有代理环境变量的通用方法；应按 requests 版本、目标地址和实际出口验证。",{"type":25,"children":775},[776,778,781],{"text":777,"type":29},"代理环境变量与请求库的关系可对照",{"text":370,"type":29,"marks":779},[780],{"href":373,"type":97},{"text":236,"type":29},{"type":31,"level":21,"children":783},[784],{"text":785,"type":29},"如何用 fixture 固定环境并提前断言",{"type":25,"children":787},[788],{"text":789,"type":29},"在测试层做三件事：",{"type":25,"children":791},[792],{"text":793,"type":29},"1. autouse fixture 在用例开始前清空所有代理变量（大小写各一份），用例结束后恢复。\n2. 需要代理的用例通过参数显式注入，不依赖外部环境。\n3. 请求发出前断言：`os.environ` 中的值、`session.proxies`、`session.trust_env` 与预期一致。",{"type":25,"children":795},[796],{"text":797,"type":29},"```python\nimport pytest",{"type":25,"children":799},[800],{"text":801,"type":29},"@pytest.fixture(autouse=True)\ndef clean_proxy_env(monkeypatch):\nfor key in (\"HTTP_PROXY\", \"HTTPS_PROXY\", \"ALL_PROXY\",\n\"NO_PROXY\", \"http_proxy\", \"https_proxy\", \"all_proxy\",\n\"no_proxy\"):\nmonkeypatch.delenv(key, raising=False)\n```",{"type":25,"children":803},[804,806,811],{"text":805,"type":29},"需要测试 `NO_PROXY` 时，在对应参数化用例中单独设置目标值。`monkeypatch` 会在用例结束时恢复原环境；可加一个检查，确认所有大小写变量恢复到用例前状态。详情见 ",{"text":807,"type":29,"marks":808},"pytest monkeypatch 文档",[809],{"href":810,"type":97},"https:\u002F\u002Fdocs.pytest.org\u002Fen\u002Fstable\u002Fhow-to\u002Fmonkeypatch.html",{"text":236,"type":29},{"type":31,"level":21,"children":813},[814],{"text":815,"type":29},"NO_PROXY 哪些边界情况必须单独测试",{"type":25,"children":817},[818],{"text":819,"type":29},"`NO_PROXY` 的匹配规则容易想当然，至少要覆盖：",{"type":52,"format":53,"children":821},[822,826,830,834,838,842],{"type":56,"children":823},[824],{"text":825,"type":29},"精确域名与子域名：`example.com` 是否覆盖 `api.example.com`。",{"type":56,"children":827},[828],{"text":829,"type":29},"端口写法：`example.com:443` 与不带端口的差异。",{"type":56,"children":831},[832],{"text":833,"type":29},"IP 与 CIDR：单 IP、网段写法是否被支持。",{"type":56,"children":835},[836],{"text":837,"type":29},"本机地址：`localhost`、`127.0.0.1`、`::1`。",{"type":56,"children":839},[840],{"text":841,"type":29},"通配：`*` 是否让所有请求直连。",{"type":56,"children":843},[844],{"text":845,"type":29},"空值：变量存在但为空字符串时的行为。",{"type":25,"children":847},[848],{"text":849,"type":29},"每个边界写成一个用例，断言\"是否走了代理\"，而不是\"变量是否被读取\"。变量被读取但匹配失败，同样属于污染。",{"type":31,"level":21,"children":851},[852],{"text":853,"type":29},"如何证明请求真的走了预期出口",{"type":25,"children":855},[856],{"text":857,"type":29},"环境变量只是配置，出口才是结果。回显请求必须由同一个被测客户端、协议和目标匹配规则发出；另用受控目标的访问日志或代理到达记录核对真正的业务请求。不同目标可能命中不同的 `NO_PROXY` 规则，单独请求另一个回显域名不能证明被测请求走了同一路径。直连出口应现场测量，并注明 NAT 或 CI 共享出口的影响：",{"type":52,"format":53,"children":859},[860,864,868],{"type":56,"children":861},[862],{"text":863,"type":29},"期望走代理：出口应属于代理侧地址段。",{"type":56,"children":865},[866],{"text":867,"type":29},"期望直连：出口应属于本机公网地址段。",{"type":56,"children":869},[870],{"text":871,"type":29},"期望绕过：目标应命中 `NO_PROXY`，出口为直连地址。",{"type":25,"children":873},[874],{"text":875,"type":29},"回显服务用自建或受控端点。分别覆盖预期走代理、命中绕过规则、故意设置错误配置和恢复后的请求，并记录目标、时间、客户端版本与代理到达证据。代理 URL 只记录脱敏后的协议、主机、端口和配置来源；删除 userinfo、令牌及敏感查询参数，避免 CI 日志泄露凭据。",{"type":31,"level":21,"children":877},[878],{"text":879,"type":29},"如何隔离开发机和 CI 继承的代理环境",{"type":25,"children":881},[882],{"text":883,"type":29},"本地 shell 里的代理变量会通过 `pytest` 进程继承进来，CI 的容器镜像也可能预置。隔离手段：",{"type":52,"format":53,"children":885},[886,890,894],{"type":56,"children":887},[888],{"text":889,"type":29},"在 `pytest.ini` 或 `conftest.py` 里统一清理，不依赖开发者本地配置。",{"type":56,"children":891},[892],{"text":893,"type":29},"CI 任务显式声明环境，不继承 runner 的全局变量。",{"type":56,"children":895},[896],{"text":897,"type":29},"需要真实代理的集成测试单独标记，默认不跑。",{"type":25,"children":899},[900],{"text":901,"type":29},"区分\"单元测试\"与\"集成测试\"的边界很重要：前者不应依赖任何外部出口，后者才需要真实代理。",{"type":25,"children":903},[904,906,911,913,916],{"text":905,"type":29},"requests 的环境设置行为见 ",{"text":907,"type":29,"marks":908},"Requests 高级用法文档",[909],{"href":910,"type":97},"https:\u002F\u002Frequests.readthedocs.io\u002Fen\u002Fstable\u002Fuser\u002Fadvanced\u002F",{"text":912,"type":29},"；测试环境变量恢复可参考 ",{"text":807,"type":29,"marks":914},[915],{"href":810,"type":97},{"text":236,"type":29},{"type":31,"level":21,"children":918},[919],{"text":920,"type":29},"如何通过清理检查和故意失败验证测试有效",{"type":25,"children":922},[923],{"text":924,"type":29},"最后一步是验证测试本身有效：把预期出口故意写错，测试必须失败。如果仍然通过，说明断言没有真正落到出口上，前面所有隔离都只是形式。",{"type":25,"children":926},[927],{"text":928,"type":29},"清理检查同样要做：用例结束后确认 `os.environ` 已恢复，没有把变量泄漏给后续用例。多个用例串行时，泄漏会表现为\"只有第一个用例正常\"。",{"type":31,"level":21,"children":930},[931],{"text":141,"type":29},{"type":25,"children":933},[934],{"text":935,"type":29,"marks":936},"Q：代码里没写代理，还会被环境变量影响吗？",[937],{"type":148},{"type":25,"children":939},[940],{"text":941,"type":29},"A：会。requests 等常见实现默认读取环境变量，`trust_env` 为 True 时自动生效；具体行为以实际版本为准，测试里要显式清理或关闭，并回显出口验证。",{"type":25,"children":943},[944],{"text":945,"type":29,"marks":946},"Q：`NO_PROXY` 写了域名，为什么子域名还是走了代理？",[947],{"type":148},{"type":25,"children":949},[950],{"text":951,"type":29},"A：不同实现对子域名的匹配规则不同，需要按库的实际行为写用例验证，不能假设后缀匹配一定生效。",{"type":25,"children":953},[954],{"text":955,"type":29,"marks":956},"Q：CI 上通过、本地失败，先查什么？",[957],{"type":148},{"type":25,"children":959},[960],{"text":961,"type":29},"A：先比对变量是否存在、脱敏后的协议\u002F主机\u002F端口和配置来源，再用同一被测请求的目标日志或代理到达记录确认路径。不要打印完整代理 URL，其中可能含账号、密码或令牌。",[963],{"source":6,"id":7,"documentId":8,"translationKey":8,"locale":9,"slug":10,"path":11,"_path":11,"title":12,"description":13,"excerpt":13,"image":14,"publishedAt":15,"updatedAt":16,"author":17,"tags":964,"featured":19,"sortOrder":20,"readingTime":21,"bodyType":22,"bodyBlocks":965,"bodyHtml":173},[],[966,969,972,975,978,981,984,992,995,998,1001,1004,1007,1010,1017,1020,1023,1031,1034,1037,1040,1043,1046,1049,1054,1057,1062,1065,1070],{"type":25,"children":967},[968],{"text":28,"type":29},{"type":31,"level":21,"children":970},[971],{"text":34,"type":29},{"type":25,"children":973},[974],{"text":38,"type":29},{"type":25,"children":976},[977],{"text":42,"type":29},{"type":31,"level":21,"children":979},[980],{"text":46,"type":29},{"type":25,"children":982},[983],{"text":50,"type":29},{"type":52,"format":53,"children":985},[986,989],{"type":56,"children":987},[988],{"text":59,"type":29},{"type":56,"children":990},[991],{"text":63,"type":29},{"type":25,"children":993},[994],{"text":67,"type":29},{"type":31,"level":21,"children":996},[997],{"text":71,"type":29},{"type":25,"children":999},[1000],{"text":75,"type":29},{"type":25,"children":1002},[1003],{"text":79,"type":29},{"type":31,"level":21,"children":1005},[1006],{"text":83,"type":29},{"type":25,"children":1008},[1009],{"text":87,"type":29},{"type":25,"children":1011},[1012,1013,1016],{"text":91,"type":29},{"text":93,"type":29,"marks":1014},[1015],{"href":96,"type":97},{"text":99,"type":29},{"type":31,"level":21,"children":1018},[1019],{"text":103,"type":29},{"type":25,"children":1021},[1022],{"text":107,"type":29},{"type":52,"format":53,"children":1024},[1025,1028],{"type":56,"children":1026},[1027],{"text":113,"type":29},{"type":56,"children":1029},[1030],{"text":117,"type":29},{"type":25,"children":1032},[1033],{"text":121,"type":29},{"type":31,"level":21,"children":1035},[1036],{"text":125,"type":29},{"type":25,"children":1038},[1039],{"text":129,"type":29},{"type":25,"children":1041},[1042],{"text":133,"type":29},{"type":25,"children":1044},[1045],{"text":137,"type":29},{"type":31,"level":21,"children":1047},[1048],{"text":141,"type":29},{"type":25,"children":1050},[1051],{"text":145,"type":29,"marks":1052},[1053],{"type":148},{"type":25,"children":1055},[1056],{"text":152,"type":29},{"type":25,"children":1058},[1059],{"text":156,"type":29,"marks":1060},[1061],{"type":148},{"type":25,"children":1063},[1064],{"text":162,"type":29},{"type":25,"children":1066},[1067],{"text":166,"type":29,"marks":1068},[1069],{"type":148},{"type":25,"children":1071},[1072],{"text":172,"type":29},null,[],[]]