[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"blog-strapi-article:zh:proxy-pool-ip-count-sizing":3,"blog-article:\u002Fblog\u002Fproxy-pool-ip-count-sizing":668,"blog-related:\u002Fblog\u002Fproxy-pool-ip-count-sizing":669,"blog-translations:\u002Fblog\u002Fproxy-pool-ip-count-sizing":670},{"available":4,"article":5,"relatedArticles":145,"translations":577},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":136},"strapi",3496,"z40ct6quyyvnlh8ux8zdik0g","zh","proxy-pool-ip-count-sizing","\u002Fblog\u002Fproxy-pool-ip-count-sizing","代理池需要多少 IP：按并发、目标站点和失败率倒推","代理池规模要同时看连接并发、请求速率、目标站限制和账号配额。用实测负载估算容量，再根据出口故障和恢复时间校准冗余。","https:\u002F\u002Fcms.quanqiudaili.com\u002Fuploads\u002Flarge_ip_cover_81e9cdb74b.webp","2026-10-21T08:23:49.934Z","2026-10-21T08:23:49.923Z","Global Proxies Team",[],false,0,2,"blocks",[24,30,35,39,43,47,51,55,59,63,68,72,76,80,84,88,92,96,100,104,108,113,117,121,125,129,133,137,141],{"type":25,"children":26},"paragraph",[27],{"text":28,"type":29},"代理池规模不能只用同时并发除以单 IP 并发来估算，还要看请求速率、响应时间、目标站限制、账号或 API Key 配额，以及是否需要为账号保留专属出口。本文给出拆分估算方法和记录模板；示例数字只用于演示，最终容量要用获准业务的实测负载校准。","text",{"type":31,"level":21,"children":32},"heading",[33],{"text":34,"type":29},"按峰值并发和单 IP 并发估算所需 IP 数",{"type":25,"children":36},[37],{"text":38,"type":29},"核心只有一个式子：",{"type":25,"children":40},[41],{"text":42,"type":29},"**连接容量所需 IP 数 = ceil（峰值并发 ÷ 实测单 IP 并发）**",{"type":25,"children":44},[45],{"text":46,"type":29},"这里的单 IP 并发应按相同目标、请求类型和观察时段实测。它只给出连接容量约束，不能代表目标站允许的请求速率，也不能解决绑定在账号或 API Key 上的配额。代理接入参数见帮助中心的[参数说明](https:\u002F\u002Fhelp.quanqiudaili.com\u002Fdocs\u002Fintro)。",{"type":31,"level":21,"children":48},[49],{"text":50,"type":29},"目标站频控和响应速度如何影响单 IP 并发",{"type":25,"children":52},[53],{"text":54,"type":29},"请求速率约束要单独估算。稳定负载下，可用请求率（QPS）× 平均响应时间估算平均在途请求数；再与实测单 IP 允许的并发和目标站允许的请求速率比较。突发请求另留余量，并以服务端限额和实际响应为边界。连接数上限与 QPS 上限不是同一个指标，不能互相替代。",{"type":25,"children":56},[57],{"text":58,"type":29},"账号或 API Key 配额触顶时，增加 IP 通常不能解决问题；目标站返回 429 时，应按其响应指示退避并降低请求，不以增加 IP 绕过限制。失败率和延迟只是线索，需结合状态码、失败阶段及代理侧容量记录判断原因。延迟和可用率的记录方法见[代理 IP 可用率、延迟和稳定性怎么测试](https:\u002F\u002Fwww.quanqiudaili.com\u002Fblog\u002Fproxy-ip-availability-latency-stability-test)。",{"type":31,"level":21,"children":60},[61],{"text":62,"type":29},"采集、登录和 API 调用各需要多少 IP",{"type":64,"children":65},"code",[66],{"text":67,"type":29},"| 示例场景 | 峰值在途请求 | 实测单 IP 并发假设 | 连接容量估算 | 还需单独确认 |\n|---|---:|---:|---:|---|\n| 公开页面请求 | 50 | 2 | 25 | 允许 QPS、授权范围、请求间隔 |\n| 登录态任务 | 20 | 1 | 20 | 是否需要账号专属且持续保留的出口 |\n| 多账号后台 | 10 | 1 | 10 | 实际同时在线数、平台规则和账号权限 |\n| API 批量调用 | 100 | 5 | 20 | 账号\u002FAPI Key 配额及服务端速率上限 |",{"type":25,"children":69},[70],{"text":71,"type":29},"表内是简化的连接容量算例，没有计入请求速率约束、突发余量和出口故障，因此不是采购数量建议。账号如需持续绑定出口，应按实际需要保留的出口数单独计算，不能直接用同时在线数代替。",{"type":25,"children":73},[74],{"text":75,"type":29},"这张表仅演示连接容量的计算，不是建议采购数量。真实业务应使用获准的请求集测量；代理接入方式见帮助中心的[代理设置](https:\u002F\u002Fhelp.quanqiudaili.com\u002Fdocs\u002Fproxy-Settings)，产品切换、测试和续费见[代理 IP 测试、续费、更换和子账号 FAQ](https:\u002F\u002Fwww.quanqiudaili.com\u002Fblog\u002Fproxy-ip-testing-renewal-replacement-subaccount-faq)。",{"type":31,"level":21,"children":77},[78],{"text":79,"type":29},"失败率和延迟升高时，先扩容还是降速",{"type":25,"children":81},[82],{"text":83,"type":29},"可观察失败率、状态码分布、P95 延迟、请求速率和代理侧连接占用。本文不设通用阈值；应先用获准的请求集建立基线，并区分目标站限流、账号配额、网络故障和代理资源耗尽。",{"type":25,"children":85},[86],{"text":87,"type":29},"只有证据指向代理资源瓶颈（例如代理侧连接容量耗尽或出口故障率异常）时，扩容才是候选方案。若平台返回 429，应先按响应要求等待并降低请求；若是账号配额不足，应检查账户限额。对照不同请求速率下的状态码、延迟和代理侧占用，再决定是否需要增加代理资源。",{"type":31,"level":21,"children":89},[90],{"text":91,"type":29},"记录并发、成功率和延迟，判断是否需要扩容",{"type":25,"children":93},[94],{"text":95,"type":29},"一周一行，看趋势不看单日：",{"type":64,"children":97},[98],{"text":99,"type":29},"| 日期 | 峰值并发 | 在用 IP 数 | 成功率 | 429 占比 | P95 延迟 |\n|---|---|---|---|---|---|\n| 2026-10-13 | 42 | 30 | 96.2% | 1.1% | 820ms |\n| 2026-10-14 | 55 | 30 | 91.4% | 3.8% | 1.4s |",{"type":25,"children":101},[102],{"text":103,"type":29},"这是假设案例，不是实测或推荐阈值：并发从 42 增至 55，约增加 31%；成功率从 96.2% 降至 91.4%，下降 4.8 个百分点。它只能提示需要进一步排查，不能单凭两天数据断定 IP 不足。实测应固定请求集、地区和时段，并同时记录状态码、请求速率、出口故障和恢复时间；据此校准冗余和容量。动态住宅的计费及时长口径见帮助中心的[动态住宅 FAQ](https:\u002F\u002Fhelp.quanqiudaili.com\u002Fdocs\u002FFaq\u002FDynamic-Residence-Faq)。",{"type":31,"level":21,"children":105},[106],{"text":107,"type":29},"FAQ",{"type":31,"level":109,"children":110},3,[111],{"text":112,"type":29},"IP 买得越多越稳吗",{"type":25,"children":114},[115],{"text":116,"type":29},"不是。IP 超过实际并发需要后，单个 IP 的请求密度下降，成本却线性上升。更常见的问题是调度没做好，新 IP 一上线就被打到同一个目标站，等于没扩容。",{"type":31,"level":109,"children":118},[119],{"text":120,"type":29},"动态住宅和静态住宅怎么配比",{"type":25,"children":122},[123],{"text":124,"type":29},"按业务会话要求评估：短时、多地区任务可核对动态住宅代理 IP 的会话和地区规格；需要长期同一出口时核对静态住宅代理 IP 的固定时长与更换条件。账号专属出口需求另行核算，不把两类资源混用当作通用要求。",{"type":31,"level":109,"children":126},[127],{"text":128,"type":29},"池子多大才需要做调度",{"type":25,"children":130},[131],{"text":132,"type":29},"是否需要调度取决于出口数量、任务类型和维护能力，没有适用于所有业务的固定 IP 数门槛。出现出口表现差异或需要按健康状态分配时，可先建立检查、记录和暂停分配规则，再按实际规模决定自动化程度。",{"type":25,"children":134},[135],{"text":136,"type":29},"",{"type":25,"children":138},[139],{"text":140,"type":29},"--",{"type":25,"children":142},[143],{"text":144,"type":29},"如果你的并发集中在公开页面采集，可以先用[动态住宅代理 IP](https:\u002F\u002Fwww.quanqiudaili.com\u002Fproduct\u002Fdynamic-residential) 跑一轮容量测试，用实测成功率再定最终数量。",[146,290,445],{"source":6,"id":147,"documentId":148,"translationKey":148,"locale":9,"slug":149,"path":150,"_path":150,"title":151,"description":152,"excerpt":152,"image":153,"publishedAt":154,"updatedAt":155,"author":17,"tags":156,"featured":19,"sortOrder":20,"readingTime":157,"bodyType":22,"bodyBlocks":158,"bodyHtml":136},3495,"s72p6uye6i8b0jr2mlb0l97b","caddy-nginx-websocket-proxy-configuration","\u002Fblog\u002Fcaddy-nginx-websocket-proxy-configuration","Caddy 与 Nginx 的 WebSocket 代理配置有什么区别：升级头与超时","对比 Caddy 与 Nginx 的 WebSocket 反向代理配置，说明 HTTP\u002F1.1 Upgrade、空闲超时和版本差异，并给出断连排查方法。","https:\u002F\u002Fcms.quanqiudaili.com\u002Fuploads\u002Flarge_caddy_nginx_websocket_cover_7b9d0db2cc.webp","2026-10-21T08:23:34.532Z","2026-10-21T08:23:34.515Z",[],4,[159,163,167,171,175,179,183,187,191,195,199,203,207,211,215,219,223,227,231,235,239,243,247,251,255,258,262,266,270,274,278,282,286],{"type":25,"children":160},[161],{"text":162,"type":29},"两者都能代理 WebSocket。本文示例采用传统 HTTP\u002F1.1 Upgrade：Nginx 需要正确转发升级头；Caddy 的 `reverse_proxy` 会处理升级。连接建立后是否断开，还要看读空闲超时、连接总时长和中间层策略。下面给出配置示例与版本边界。",{"type":25,"children":164},[165],{"text":166,"type":29},"排查时先看握手是否成功，再看连接建立后的断开时间和各层日志。客户端出网代理的协议选择见[HTTP、HTTPS、SOCKS5 代理怎么选](https:\u002F\u002Fwww.quanqiudaili.com\u002Fblog\u002Fhttp-https-socks5-proxy-selection)。",{"type":31,"level":21,"children":168},[169],{"text":170,"type":29},"Nginx 如何配置 WebSocket 升级头和读超时",{"type":25,"children":172},[173],{"text":174,"type":29},"```nginx\nlocation \u002Fws\u002F {\nproxy_pass http:\u002F\u002F127.0.0.1:8080;\nproxy_http_version 1.1;\nproxy_set_header Upgrade $http_upgrade;\nproxy_set_header Connection \"upgrade\";\nproxy_read_timeout 3600s;\nproxy_buffering off;\n}\n```",{"type":25,"children":176},[177],{"text":178,"type":29},"`Upgrade` 与 `Connection` 用于传统 HTTP\u002F1.1 握手；`proxy_read_timeout` 控制两次上游读取之间允许的空闲时间，长空闲连接应按应用心跳和实际链路设置。Nginx 官方文档对 HTTP\u002F1.1 默认值有版本说明：1.29.7 之前需显式设置；示例保留显式配置以兼容较旧版本。升级后的连接按隧道工作，不能把普通 HTTP 响应缓冲直接当作握手条件。长连接保活见[WebSocket 代理如何保持长连接](https:\u002F\u002Fwww.quanqiudaili.com\u002Fblog\u002Fwebsocket-proxy-long-lived-connection)。",{"type":31,"level":21,"children":180},[181],{"text":182,"type":29},"Caddy 如何用 reverse_proxy 转发 WebSocket",{"type":25,"children":184},[185],{"text":186,"type":29},"```caddyfile\nexample.com {\nreverse_proxy 127.0.0.1:8080\n}\n```",{"type":25,"children":188},[189],{"text":190,"type":29},"这就够了。Caddy 会在转发时处理升级，不需要手写 `Upgrade` 和 `Connection`。只有业务需要限制整条流连接的最长存活时间时才设置 `stream_timeout`；它不是空闲超时，达到上限时活动连接也会关闭。外层 `example.com { }` 与上面相同：",{"type":25,"children":192},[193],{"text":194,"type":29},"```caddyfile\nreverse_proxy 127.0.0.1:8080 {\nstream_timeout 24h\n}\n```",{"type":25,"children":196},[197],{"text":198,"type":29},"`stream_timeout` 限制流连接总时长，默认不设此上限；不要用它代替心跳或空闲检测。配置重载时的 `stream_close_delay` 是另一项行为。`transport http` 的读写超时也应按对应文档和应用行为核实。配置项见 [Caddy reverse_proxy 文档](https:\u002F\u002Fcaddyserver.com\u002Fdocs\u002Fcaddyfile\u002Fdirectives\u002Freverse_proxy)。",{"type":25,"children":200},[201],{"text":202,"type":29},"代理本身的主机、端口和鉴权写法，见帮助中心的[代理设置](https:\u002F\u002Fhelp.quanqiudaili.com\u002Fdocs\u002Fproxy-Settings)。",{"type":31,"level":21,"children":204},[205],{"text":206,"type":29},"升级头、空闲超时和响应缓冲会怎样影响连接",{"type":64,"children":208},[209],{"text":210,"type":29},"| 对比项 | Nginx | Caddy | 实际影响 |\n|---|---|---|---|\n| HTTP\u002F1.1 Upgrade | 通常需显式转发升级头；HTTP\u002F1.1 默认值受版本影响 | `reverse_proxy` 处理升级 | 配置与版本不匹配可能导致握手失败 |\n| 空闲读超时 | `proxy_read_timeout` 控制上游读取空闲间隔 | 依据具体传输配置 | 按心跳和链路实际空闲策略判断 |\n| 流连接总时长 | 由具体配置决定 | `stream_timeout` 可设总时长上限 | 到时会关闭活动连接 |\n| 响应缓冲 | Upgrade 后是双向隧道；按实际数据路径排查 | 流式转发 | 不作为握手成功的必备条件 |\n| 证书 | 需另配 certbot | 自动申请续期 | 部署步骤数量不同 |\n| HTTP\u002F2 | 取决于客户端与服务器实现 | 取决于实现 | RFC 8441 定义扩展 CONNECT，支持情况需逐端核实 |",{"type":25,"children":212},[213],{"text":214,"type":29},"若使用传统 Upgrade，按 HTTP\u002F1.1 握手流程排查；HTTP\u002F2 扩展 CONNECT 是否可用取决于客户端、代理和上游实现。若只有特定目标失败，比较该目标的响应和代理日志，不要直接归因于出口。",{"type":31,"level":21,"children":216},[217],{"text":218,"type":29},"如何设置 WebSocket 保活间隔并定位断连",{"type":31,"level":109,"children":220},[221],{"text":222,"type":29},"按链路最短空闲超时设置 ping 间隔",{"type":25,"children":224},[225],{"text":226,"type":29},"链路上任一层的空闲策略都可能关闭连接，实际时限需查对应 CDN、负载均衡和应用配置，并用边缘及后端日志测量。不要把未经核实的固定秒数当作通用默认值。应用层心跳间隔应依据实测到的最短空闲时限设置，并留出网络抖动余量。",{"type":31,"level":109,"children":228},[229],{"text":230,"type":29},"握手失败或连接中断时按四步排查",{"type":25,"children":232},[233],{"text":234,"type":29},"排障按这个顺序走，不要跳步：",{"type":25,"children":236},[237],{"text":238,"type":29},"1. `curl -i -N -H \"Connection: Upgrade\" -H \"Upgrade: websocket\" -H \"Sec-WebSocket-Version: 13\" -H \"Sec-WebSocket-Key: x3JJHMbDL1EzLkh9GBhXDw==\" http:\u002F\u002F你的域名\u002Fws\u002F`，看是否返回 `101 Switching Protocols`\n2. 没有 101 就查 Nginx 的头和 `proxy_pass` 路径\n3. 有 101 但很快断，记录断开间隔并对照各层实际超时配置与日志\n4. 断得没有规律，结合代理、CDN 和后端日志检查连接限制或异常关闭原因",{"type":31,"level":109,"children":240},[241],{"text":242,"type":29},"根据 400、502、504 和固定时长断连定位故障层",{"type":25,"children":244},[245],{"text":246,"type":29},"出现下面这些现象时，先查对应的一层，不要先改代理：",{"type":64,"children":248},[249],{"text":250,"type":29},"| 现象 | 先查哪一层 | 判断动作 |\n|---|---|---|\n| 握手返回 400 或 426 | Nginx 升级头 | 核对 `Upgrade`、`Connection` 和 HTTP\u002F1.1 |\n| 返回 502 | 后端 | 后端进程、端口、路径是否与 `proxy_pass` 一致 |\n| 返回 504 | Nginx 读超时 | 看后端处理耗时，再调 `proxy_read_timeout` |\n| 101 之后 60 秒整断 | Nginx | `proxy_read_timeout` 仍是默认值 |\n| 101 后固定时长断开 | CDN、负载均衡或应用 | 对照该层配置及日志确认空闲策略 |\n| 消息延迟或成批到达 | 应用写入、代理或中间层 | 检查数据流和缓冲行为，不据此判断握手失败 |",{"type":25,"children":252},[253],{"text":254,"type":29},"超时、连接被拒和 TLS 证书错误是三组不同原因：超时多半卡在链路或上游，连接被拒说明端口那头没在听，证书错误说明中间有人在拦截或替换证书。先按 curl 报出的类型分开，再决定动哪一端。",{"type":31,"level":21,"children":256},[257],{"text":107,"type":29},{"type":31,"level":109,"children":259},[260],{"text":261,"type":29},"Caddy 还要不要显式写 Upgrade 头",{"type":25,"children":263},[264],{"text":265,"type":29},"不用。v2 的 reverse_proxy 会自动处理升级，手写反而可能覆盖掉它自己的逻辑。只有当你需要改写头内容时才用 `header_up`。",{"type":31,"level":109,"children":267},[268],{"text":269,"type":29},"Nginx 报 502 和 504 分别查什么",{"type":25,"children":271},[272],{"text":273,"type":29},"502 是后端拒绝或崩溃，查后端进程和端口；504 是后端在超时内没响应，查 `proxy_read_timeout` 和后端处理耗时。两者都不指向代理 IP 本身。",{"type":31,"level":109,"children":275},[276],{"text":277,"type":29},"HTTP\u002F2 下 WebSocket 还能用吗",{"type":25,"children":279},[280],{"text":281,"type":29},"RFC 8441 定义了 HTTP\u002F2 的扩展 CONNECT；是否可用取决于客户端、代理及服务端实现。本文配置示例采用传统 HTTP\u002F1.1 Upgrade，部署时应核实整条链路支持的方式。",{"type":25,"children":283},[284],{"text":285,"type":29},"---",{"type":25,"children":287},[288],{"text":289,"type":29},"如果长连接主要面向需要稳定出口的业务后台，用[静态住宅代理 IP](https:\u002F\u002Fwww.quanqiudaili.com\u002Fproduct\u002Fstatic-residential-standard) 比频繁更换出口更省事，出口固定之后，超时和缓冲问题也更容易复现。",{"source":6,"id":291,"documentId":292,"translationKey":292,"locale":9,"slug":293,"path":294,"_path":294,"title":295,"description":296,"excerpt":296,"image":297,"publishedAt":298,"updatedAt":299,"author":17,"tags":300,"featured":19,"sortOrder":20,"readingTime":109,"bodyType":22,"bodyBlocks":301,"bodyHtml":136},3494,"wr2sddq3u4r8i6w15lxd9ofv","deepseek-cursor-proxy-setup","\u002Fblog\u002Fdeepseek-cursor-proxy-setup","DeepSeek 在 Cursor 中怎么配置：API 地址、代理和排障","在 Cursor 中配置 DeepSeek 自定义 API Key 和 Base URL 后，分别验证本机网络、Cursor 请求及服务端响应；实际路径和模型名称需按当前版本核对。","https:\u002F\u002Fcms.quanqiudaili.com\u002Fuploads\u002Flarge_deepseek_cursor_cover_6b15e6639b.webp","2026-10-21T08:23:18.321Z","2026-10-21T08:23:18.310Z",[],[302,306,310,314,318,322,326,330,334,338,355,359,363,367,371,375,379,383,387,391,395,399,403,407,411,414,418,422,426,430,434,438,441],{"type":25,"children":303},[304],{"text":305,"type":29},"在 Cursor 使用 DeepSeek 自定义 API Key 时，需要配置服务地址和凭据，并分别检查本机网络、Cursor 的请求路径及模型服务端响应。Cursor BYOK 请求不一定与本机 `curl` 走同一条端到端链路，因此终端测试不能单独证明 Cursor 请求成功。具体界面和路径应按当前 Cursor 与 DeepSeek 文档核实。",{"type":31,"level":21,"children":307},[308],{"text":309,"type":29},"用状态码和终端请求区分模型配置与代理故障",{"type":25,"children":311},[312],{"text":313,"type":29},"模型接入错了，表现是 401、404 或\"model not found\"；网络出口错了，表现是超时、连接被重置、流式输出卡住。前者看 key 和地址，后者看代理和兼容模式，混着查只会来回改配置。顺带说清一个常见混淆：业务场景里代理和 VPN 的差别决定了你要的是固定出口还是临时改道。",{"type":25,"children":315},[316],{"text":317,"type":29},"本机 `curl` 可用于检查本机到 API 端点的连通性及该请求的凭据响应；Cursor 内失败时，还要核对 Cursor 的请求记录和服务端日志，不能据此排除 Cursor 到模型服务之间的问题。",{"type":31,"level":21,"children":319},[320],{"text":321,"type":29},"在 Cursor 中填写 DeepSeek API Key 和 Base URL",{"type":25,"children":323},[324],{"text":325,"type":29},"打开 Cursor Settings > Models，在 OpenAI API Key 一栏填入 DeepSeek 的 key（不是 Cursor 自己的 key），再打开 **Override OpenAI Base URL** 填 `https:\u002F\u002Fapi.deepseek.com\u002Fv1`。",{"type":25,"children":327},[328],{"text":329,"type":29},"基础地址可使用 `https:\u002F\u002Fapi.deepseek.com`；是否在 Base URL 中包含 `\u002Fv1`，取决于当前 Cursor 版本如何拼接最终请求路径。配置前核对最终 URL，避免路径重复或缺失导致 404。模型名称也应以 DeepSeek 当前模型目录及 Cursor 实际请求为准，不要仅凭旧别名判断接入成功。",{"type":31,"level":21,"children":331},[332],{"text":333,"type":29},"让 Cursor 通过系统代理发送模型请求",{"type":25,"children":335},[336],{"text":337,"type":29},"Cursor 没有单独的\"代理地址\"输入框，它跟随系统代理或环境变量（`HTTPS_PROXY`）。在需要经过代理的网络里，还要处理流式传输：",{"type":339,"format":340,"children":341},"list","unordered",[342,347,351],{"type":343,"children":344},"list-item",[345],{"text":346,"type":29},"Cursor 默认用 HTTP\u002F2 双向流拿响应，部分代理会缓冲这种流，表现为回复长时间空白然后一次性出现",{"type":343,"children":348},[349],{"text":350,"type":29},"遇到这种情况，去 Cursor Settings > Network，把 HTTP Compatibility Mode 切到 HTTP\u002F1.1，重启 Cursor",{"type":343,"children":352},[353],{"text":354,"type":29},"官方会在 HTTP\u002F2 不通时自动回退到 HTTP\u002F1.1 的 SSE，但回退不是瞬时的，手动切换更快",{"type":25,"children":356},[357],{"text":358,"type":29},"以上行为以 [Cursor 网络配置文档](https:\u002F\u002Fcursor.com\u002Fdocs\u002Fenterprise\u002Fnetwork-configuration) 为准；界面文案和默认值会随版本变化，以当前设置页为准。",{"type":25,"children":360},[361],{"text":362,"type":29},"代理本身的接入方式（主机、端口、账号密码）见帮助中心的[代理设置](https:\u002F\u002Fhelp.quanqiudaili.com\u002Fdocs\u002Fproxy-Settings) 页面，各客户端写法不同，但出口是同一个。",{"type":31,"level":21,"children":364},[365],{"text":366,"type":29},"从 API 连通性到 Cursor 对话逐步验证",{"type":31,"level":109,"children":368},[369],{"text":370,"type":29},"依次检查 API 凭据、代理出口和 Cursor 对话",{"type":25,"children":372},[373],{"text":374,"type":29},"按顺序做，每一步都能独立判断：",{"type":25,"children":376},[377],{"text":378,"type":29},"1. 在终端请求当前 DeepSeek 文档列出的端点，记录最终 URL、状态码和响应体；这只验证该终端请求。\n2. 如需测试代理，可对同一条终端请求添加代理参数，并记录连接阶段、状态码和响应；这不能代表 Cursor 使用了相同链路。\n3. 在 Cursor 内发起一次请求，记录 Cursor 版本、设置项、时间及界面错误；不要把逐字输出或 UI 模型标签当作服务提供方证明。\n4. 如能查看自定义服务端日志，使用脱敏 request ID、时间戳、实际 model 字段和用量核对是否收到 Cursor 请求。",{"type":31,"level":109,"children":380},[381],{"text":382,"type":29},"对照返回结果确认每一步是否通过",{"type":25,"children":384},[385],{"text":386,"type":29},"每一步的通过标准可以对照下表，避免\"看起来能连\"就往下走：",{"type":64,"children":388},[389],{"text":390,"type":29},"| 步骤 | 要确认的东西 | 通过标准 |\n|---|---|---|\n| 终端验 key | key 是否有效、端点路径是否正确 | 返回 JSON 模型列表，无 401\u002F404 |\n| 终端验代理 | 代理能否到 `api.deepseek.com` | 与直连结果一致，延迟稳定 |\n| Cursor 请求 | Cursor 是否成功提交并收到响应 | 记录实际错误、时间和请求 ID（如可用） |\n| 服务端核对 | 自定义服务是否收到请求 | 对照脱敏 request ID、实际 model 字段及用量 |\n| 网络路径 | 哪一段连接失败 | 分开记录本机网络、Cursor 到服务端和服务端上游日志 |",{"type":25,"children":392},[393],{"text":394,"type":29},"主机、端口和出口 IP 不是一回事，三者的对应关系见[代理主机、端口和出口 IP 有什么区别](https:\u002F\u002Fwww.quanqiudaili.com\u002Fblog\u002Fproxy-host-port-exit-ip-difference)。",{"type":25,"children":396},[397],{"text":398,"type":29},"如怀疑网络变化与连接重置有关，记录对应请求时间、连接阶段及代理日志，再检查 Cursor 与服务端各自的链路；仅看到出口 IP 变化不能证明故障由此造成。代理认证方式见[代理白名单和账号密码怎么选](https:\u002F\u002Fwww.quanqiudaili.com\u002Fblog\u002Fproxy-whitelist-vs-username-password)。",{"type":31,"level":21,"children":400},[401],{"text":402,"type":29},"根据 401、404、超时和流式输出异常排查",{"type":64,"children":404},[405],{"text":406,"type":29},"| 现象 | 最可能原因 | 检查动作 |\n|---|---|---|\n| 401 Unauthorized | 凭据无效、权限或请求配置问题 | 检查服务端错误体、凭据来源和权限；勿在日志暴露密钥 |\n| 404 Not Found | 最终路径、模型或端点不匹配 | 记录实际请求 URL，并按客户端文档核对路径拼接 |\n| 402 或额度报错 | 余额、额度或模型计费条件 | 查看账户状态和响应体的具体错误字段 |\n| 请求超时 | 可能发生在本机、Cursor 或上游链路 | 对照请求阶段、Cursor 诊断信息、服务端日志和代理日志 |\n| 回复延迟或中断 | 流式协议、网络、中间层或服务端处理 | 比较兼容模式并查看对应阶段日志，不预设唯一原因 |\n| 偶发连接重置 | 连接路径或服务端\u002F中间层异常 | 记录时间、request ID、连接阶段和相关日志 |",{"type":25,"children":408},[409],{"text":410,"type":29},"超时、连接被拒和证书错误要分开判断：前者看链路，后两者看端口和中间设备。只想确认链路通不通，可以打开帮助中心的[智能检测助手](https:\u002F\u002Fhelp.quanqiudaili.com\u002Fdocs\u002FSmart-check)。",{"type":31,"level":21,"children":412},[413],{"text":107,"type":29},{"type":31,"level":109,"children":415},[416],{"text":417,"type":29},"Base URL 到底要不要带 \u002Fv1",{"type":25,"children":419},[420],{"text":421,"type":29},"不一定。按当前 Cursor 版本确认它会如何拼接路径，并以最终请求 URL 为准；若客户端已自动添加 `\u002Fv1`，Base URL 再包含它可能造成重复路径。",{"type":31,"level":109,"children":423},[424],{"text":425,"type":29},"代理开着，Cursor 还是报错怎么办",{"type":25,"children":427},[428],{"text":429,"type":29},"终端 `curl` 只能验证该终端请求。记录 Cursor 的错误和版本，并核对代理日志、Cursor 诊断信息及自定义服务端是否收到请求；各段结果对照后再定位故障。",{"type":31,"level":109,"children":431},[432],{"text":433,"type":29},"自定义 key 和 Cursor 订阅额度是什么关系",{"type":25,"children":435},[436],{"text":437,"type":29},"两套计费：用自定义 key 时走 DeepSeek 账户余额，Cursor 的请求额度不参与；反过来，Cursor 自带模型的额度也抵扣不了 DeepSeek 的费用。",{"type":25,"children":439},[440],{"text":285,"type":29},{"type":25,"children":442},[443],{"text":444,"type":29},"若日志确认出口切换造成连接中断，可按实际会话要求评估[静态住宅代理 IP](https:\u002F\u002Fwww.quanqiudaili.com\u002Fproduct\u002Fstatic-residential-standard)；这不能解决 Cursor 服务端、API 地址或鉴权问题。",{"source":6,"id":446,"documentId":447,"translationKey":447,"locale":9,"slug":448,"path":449,"_path":449,"title":450,"description":451,"excerpt":451,"image":452,"publishedAt":453,"updatedAt":454,"author":17,"tags":455,"featured":19,"sortOrder":20,"readingTime":109,"bodyType":22,"bodyBlocks":456,"bodyHtml":136},3493,"gwpcodeirv2ezrljfhnx7hq2","google-maps-data-collection-proxy-selection","\u002Fblog\u002Fgoogle-maps-data-collection-proxy-selection","Google Maps 采集代理怎么选：地区、并发与封禁边界","选择 Google Maps 采集代理前，先确认数据授权和采集方式，再按位置、请求类型与业务规则评估代理。验证码或限制出现时应暂停并记录，不以更换出口继续采样。","https:\u002F\u002Fcms.quanqiudaili.com\u002Fuploads\u002Flarge_google_maps_cover_f65bbe4ba0.webp","2026-10-21T08:23:01.794Z","2026-10-21T08:23:01.773Z",[],[457,461,465,469,479,483,487,491,495,499,503,507,511,515,519,523,527,531,535,539,543,546,550,554,558,562,566,570,573],{"type":25,"children":458},[459],{"text":460,"type":29},"Google Maps 的结果会受查询、地图视窗、语言、账号状态和位置等因素影响，单靠城市出口不能保证结果一致。开始采集前，先确认数据来源和用途符合适用条款；API 未提供某字段不代表网页采集自动获得授权。本文按请求类型说明如何做合规的小样本校准，并区分官方 API 与网页结果。",{"type":31,"level":21,"children":462},[463],{"text":464,"type":29},"按搜索结果、商家详情和评论拆分采集任务",{"type":25,"children":466},[467],{"text":468,"type":29},"Maps 上有两类完全不同的请求：",{"type":339,"format":340,"children":470},[471,475],{"type":343,"children":472},[473],{"text":474,"type":29},"搜索结果页：记录查询词、地图视窗或中心坐标、语言和返回字段。",{"type":343,"children":476},[477],{"text":478,"type":29},"Place Details：记录商家标识、请求字段及所使用的数据接口或授权来源。",{"type":25,"children":480},[481],{"text":482,"type":29},"不同请求应分开记录响应和失败阶段，先确认所用接口及用途获准。若获准使用自动化客户端，再按客户端文档配置网络；不要把页面抓取当作 API 的默认替代方案。浏览器代理配置见 [AdsPower 设置代理](https:\u002F\u002Fhelp.quanqiudaili.com\u002Fdocs\u002Fproxy-Settings\u002FFingerprint-Browser-Setting\u002FAdspower)。",{"type":31,"level":21,"children":484},[485],{"text":486,"type":29},"出口地区不匹配会怎样改变 Maps 搜索结果",{"type":25,"children":488},[489],{"text":490,"type":29},"出口位置可能影响本地化结果，但不是充分条件。对照时还要固定查询词、地图视窗或经纬度、语言、账号状态和搜索历史；否则不能把结果变化归因于 IP。可记录以下待验证现象：",{"type":25,"children":492},[493],{"text":494,"type":29},"1. 结果按出口地区重新排序，本地热门商家被挤出前列\n2. 界面语言和地名跟着出口地区走，抓到的字段与预期不符\n3. 地图中心偏移，返回的商家集合本身就不一样",{"type":25,"children":496},[497],{"text":498,"type":29},"如业务确需测试地区影响，应一次只改变出口位置，并在获准的数据来源和相同查询条件下重复记录。结果仍可能受到页面个性化或数据更新影响。主机、端口和出口 IP 的关系见[代理主机、端口和出口 IP 有什么区别](https:\u002F\u002Fwww.quanqiudaili.com\u002Fblog\u002Fproxy-host-port-exit-ip-difference)。",{"type":31,"level":21,"children":500},[501],{"text":502,"type":29},"按采集任务选择住宅代理 IP 或机房代理 IP",{"type":64,"children":504},[505],{"text":506,"type":29},"| 采集对象 | 建议 IP 类型 | 起始并发 | 观察指标 |\n|---|---|---|---|\n| 业务条件 | 可评估的代理属性 | 需要另行确认 |\n|---|---|---|\n| 短时、多地区测试且来源和用途获准 | 可评估动态住宅代理 IP | 会话时长、地区选择方式、允许请求量 |\n| 业务要求同一出口持续不变 | 优先评估静态住宅代理 IP | 实际固定时长、替换条件、套餐规格 |\n| 对住宅属性没有要求、只需一般网络访问 | 可评估机房代理 IP | 目标服务规则、速率限制和访问授权 |",{"type":25,"children":508},[509],{"text":510,"type":29},"产品类别与资源属性应分开理解：ISP 通常指住宅网络来源，IDC 指机房来源；动态\u002F静态描述出口分配方式，共享\u002F独享描述资源使用方式。它们不是四种并列产品。静态住宅代理 IP 规格见[产品说明](https:\u002F\u002Fwww.quanqiudaili.com\u002Fproduct\u002Fstatic-residential-standard)。",{"type":25,"children":512},[513],{"text":514,"type":29},"任何类型的代理都不能保证访问或结果稳定；是否适用应以服务规则、授权范围和实际规格为准。机房代理 IP 的规格见[产品说明](https:\u002F\u002Fwww.quanqiudaili.com\u002Fproduct\u002Fdatacenter-proxy)。",{"type":31,"level":21,"children":516},[517],{"text":518,"type":29},"从低并发小样本开始测试，再逐步调整请求频率",{"type":25,"children":520},[521],{"text":522,"type":29},"若已确认采集方式获准，可用小样本验证流程本身。表格中的数量只作空白记录模板，不代表推荐请求量：",{"type":64,"children":524},[525],{"text":526,"type":29},"| 批次 | 关键词数 | 城市数 | 请求数 | 成功率 | 验证码率 |\n|---|---|---|---|---|---|\n| 校准轮 | 待定 | 待定 | 待记录 | 待填 | 待填 |\n| 复核轮 | 待定 | 待定 | 待记录 | 待填 | 待填 |",{"type":25,"children":528},[529],{"text":530,"type":29},"“待填”不是实测结果，不设通用成功率或验证码率阈值。若出现 403、429 或验证码，应暂停该轮采样，记录时间、请求类型和响应，使用平台正常验证、授权 API 或支持渠道；不要更换出口后继续尝试。代理地区字段说明见帮助中心[获取代理](https:\u002F\u002Fhelp.quanqiudaili.com\u002Fdocs\u002FGet-Proxy\u002F) 页面。",{"type":31,"level":21,"children":532},[533],{"text":534,"type":29},"什么时候该用官方 API",{"type":25,"children":536},[537],{"text":538,"type":29},"Google 提供 [Places API](https:\u002F\u002Fdevelopers.google.com\u002Fmaps\u002Fdocumentation\u002Fplaces\u002Fweb-service\u002Foverview)，按请求计费并有配额限制，返回结构化字段，稳定性远高于页面解析。判断标准很简单：字段能从 API 拿到、量级在预算内，就用 API；只有 API 不提供的字段（比如特定排序下的展示位置）才考虑页面采集。",{"type":25,"children":540},[541],{"text":542,"type":29},"使用任何页面或接口数据前，应先确认适用的服务条款、许可和数据使用边界；不得绕过登录、验证码或访问限制。代理只改变网络出口，不提供额外授权。",{"type":31,"level":21,"children":544},[545],{"text":107,"type":29},{"type":31,"level":109,"children":547},[548],{"text":549,"type":29},"机房 IP 采 Maps 行不行",{"type":25,"children":551},[552],{"text":553,"type":29},"能跑通一部分，但通过率和稳定性通常低于住宅 IP，也更容易触发验证。只做小批量、非敏感字段时可以试，量一大往往就得降速。",{"type":31,"level":109,"children":555},[556],{"text":557,"type":29},"城市级定位怎么保证",{"type":25,"children":559},[560],{"text":561,"type":29},"提取代理时指定国家与城市，并在请求里带上对应的语言和地区参数；每次切换城市后，先用一个已知关键词核对返回的第一页是否与人工结果一致。",{"type":31,"level":109,"children":563},[564],{"text":565,"type":29},"被要求验证后多久恢复",{"type":25,"children":567},[568],{"text":569,"type":29},"没有适用于所有情况的恢复时长。遇到验证或访问限制时暂停采样，记录提示并按平台提供的验证、申诉或授权数据渠道处理；不要通过更换出口继续请求。",{"type":25,"children":571},[572],{"text":285,"type":29},{"type":25,"children":574},[575],{"text":576,"type":29},"如果获准的业务确实要求长期使用同一出口，可评估[静态住宅代理 IP](https:\u002F\u002Fwww.quanqiudaili.com\u002Fproduct\u002Fstatic-residential-standard)，并在购买前确认可固定时长、地区范围和更换条件。动态住宅代理 IP 的会话粘性或地区选择不等于长期固定出口。",[578],{"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":579,"featured":19,"sortOrder":20,"readingTime":21,"bodyType":22,"bodyBlocks":580,"bodyHtml":136},[],[581,584,587,590,593,596,599,602,605,608,611,614,617,620,623,626,629,632,635,638,641,644,647,650,653,656,659,662,665],{"type":25,"children":582},[583],{"text":28,"type":29},{"type":31,"level":21,"children":585},[586],{"text":34,"type":29},{"type":25,"children":588},[589],{"text":38,"type":29},{"type":25,"children":591},[592],{"text":42,"type":29},{"type":25,"children":594},[595],{"text":46,"type":29},{"type":31,"level":21,"children":597},[598],{"text":50,"type":29},{"type":25,"children":600},[601],{"text":54,"type":29},{"type":25,"children":603},[604],{"text":58,"type":29},{"type":31,"level":21,"children":606},[607],{"text":62,"type":29},{"type":64,"children":609},[610],{"text":67,"type":29},{"type":25,"children":612},[613],{"text":71,"type":29},{"type":25,"children":615},[616],{"text":75,"type":29},{"type":31,"level":21,"children":618},[619],{"text":79,"type":29},{"type":25,"children":621},[622],{"text":83,"type":29},{"type":25,"children":624},[625],{"text":87,"type":29},{"type":31,"level":21,"children":627},[628],{"text":91,"type":29},{"type":25,"children":630},[631],{"text":95,"type":29},{"type":64,"children":633},[634],{"text":99,"type":29},{"type":25,"children":636},[637],{"text":103,"type":29},{"type":31,"level":21,"children":639},[640],{"text":107,"type":29},{"type":31,"level":109,"children":642},[643],{"text":112,"type":29},{"type":25,"children":645},[646],{"text":116,"type":29},{"type":31,"level":109,"children":648},[649],{"text":120,"type":29},{"type":25,"children":651},[652],{"text":124,"type":29},{"type":31,"level":109,"children":654},[655],{"text":128,"type":29},{"type":25,"children":657},[658],{"text":132,"type":29},{"type":25,"children":660},[661],{"text":136,"type":29},{"type":25,"children":663},[664],{"text":140,"type":29},{"type":25,"children":666},[667],{"text":144,"type":29},null,[],[]]