feat: track thread ids and retry stream disconnects

README.md

@@ -9,6 +9,8 @@ tg群:https://t.me/AI_INPUT_IM
 - 保持 Codex 继续使用现有 `auth.json`
 - 只把 `config.toml` 的当前 provider `base_url` 改成本地网关
 - 非流式命中 `reasoning_tokens = 516` 时返回 `502`
+- 上游若返回明确的容量错误（默认匹配错误文案 `Selected model is at capacity. Please try a different model.`，以及 `stream disconnected before completion: Concurrency limit exceeded for account, please retry later`），也会自动重试；重试耗尽后转成本地 `502`
+- 除了 `429/503` JSON 错误响应，也会识别 `200` 但返回体本质是错误、以及流式失败事件里携带同样文案的情况
 - 流式命中时默认先缓存并判断；一旦命中 `516`，统一返回 `502`
 - 默认同时拦截 root 路径和 `/v1` 路径：
   - `/responses`
@@ -139,6 +141,8 @@ profile env 默认放在：
 - `CODEX_RETRY_GATEWAY_LISTEN_PORT`
 - `CODEX_RETRY_GATEWAY_UPSTREAM_BASE_URL`
 - `CODEX_RETRY_GATEWAY_REASONING_EQUALS`
+- `CODEX_RETRY_GATEWAY_RETRYABLE_STATUS_CODES`
+- `CODEX_RETRY_GATEWAY_RETRYABLE_ERROR_MESSAGES`
 - `CODEX_RETRY_GATEWAY_UPSTREAM_AUTH_MODE`
 - `CODEX_RETRY_GATEWAY_UPSTREAM_AUTH_ENV`
 - `CODEX_RETRY_GATEWAY_UPSTREAM_AUTH_FILE`
@@ -159,6 +163,12 @@ profile env 默认放在：
 node ./scripts/run-profile.mjs default
 ```
 
+如果已经由 state 记录了当前活跃 profile，也可以直接不传 profile 参数：
+
+```bash
+node ./scripts/run-profile.mjs --prefer-state-profile
+```
+
 ## 如何恢复
 
 Windows:
@@ -199,7 +209,7 @@ gateway 运行时只负责 API 与静态文件服务，不再把复杂 UI 硬写
 页面里可以直接做这几件事：
 
 - 看当前监听地址、真实上游、当前 provider、当前 Codex base URL
-- 看本次 gateway 启动以来的实时统计
+- 看基于持久请求记录累计的实时统计
   - 代理请求总数
   - 被检查响应总数
   - 累计 input / output / total / reasoning tokens
@@ -207,12 +217,14 @@ gateway 运行时只负责 API 与静态文件服务，不再把复杂 UI 硬写
   - `516` 占比
 - 看最近请求记录
   - 请求时间戳、首字耗时、总耗时、请求体大小、路径、模型、状态码
+  - 相同重发请求会带相同的 `request_id`
   - `usage` 中的 input / output / total / reasoning tokens
 - 管理 profiles
   - 新建 / 编辑 profile env
   - 切换 provider `base_url`
   - 切换 `passthrough` / `manual_bearer` / `fixed_bearer` / `auth_json` 认证模式
 - 改 `reasoning_equals`
+- 改 capacity error 的匹配状态码和错误文案
 - 改 `endpoints`
 - 改 `non_stream_status_code`
 - 开关 `log_match`
@@ -225,11 +237,11 @@ gateway 运行时只负责 API 与静态文件服务，不再把复杂 UI 硬写
 - 页面点“恢复 Codex 原设置并关闭网关”后，当前页面会失联，这是预期行为
 - 日常恢复优先用 UI；`restore-codex-config.ps1` 作为脚本级应急回滚入口保留
 - UI 恢复不会再额外拉起恢复子进程，而是由当前 gateway 直接完成恢复并退出
-- 统计口径默认按“本次 gateway 启动以来”累计
+- 统计口径默认按持久请求记录累计，重启后会从 `requests.sqlite` / `requests.jsonl` 回灌
 - `516` 占比 = `reasoning_tokens = 516` 的响应次数 / 被检查响应总数
 - 请求历史只记录元数据、请求体字节数和 token usage，不保存请求正文或响应正文；默认展示最近 200 条
 - gateway 日志持久化到 `~/.codex-retry-gateway/logs/gateway.log`
-- 请求记录持久化到 `~/.codex-retry-gateway/logs/requests.jsonl`，重启后仍可用于 UI 请求页和 token totals
+- 请求记录持久化到 `~/.codex-retry-gateway/logs/requests.jsonl` 与 `~/.codex-retry-gateway/logs/requests.sqlite`，重启后会用于 UI 请求页以及 overview 的累计 token / reasoning 统计
 - `manual_bearer` 的手动 token/password 只写入系统 secret 文件，API/UI 不读回明文；profile env 只保存 secret 文件路径
 - 其他 profile env 不保存明文 `sk-...`；固定密钥请使用 env/file 引用，或用 `auth_json` 指向 `auth.json` 字段名
 
@@ -246,6 +258,12 @@ macOS / Linux: ~/.codex-retry-gateway/config/config.json
 
 - `reasoning_equals`
   - 例如 `[516]`
+- `retryable_status_codes`
+  - 默认 `[429, 503]`
+- `retryable_error_messages`
+  - 默认包含 `Selected model is at capacity. Please try a different model.`
+  - 也默认包含 `stream disconnected before completion: Concurrency limit exceeded for account, please retry later`
+  - 只要上游 JSON 错误里包含这些文案之一，gateway 就会把上游错误翻成本地 `non_stream_status_code`
 - `endpoints`
   - 默认包含 root 与 `/v1` 两套路径
 - `non_stream_status_code`

build.md

@@ -81,7 +81,7 @@ http://127.0.0.1:4610/__codex_retry_gateway/ui
 页面支持：
 
 - 查看当前接管状态
-- 查看本次启动以来的实时日志
+- 查看跨重启保留的请求累计统计，以及当前进程的实时日志
 - 查看 `516` 命中次数与 `516` 占比
 - 热更新 `reasoning_equals` / `endpoints` / `non_stream_status_code` / `log_match`
 - 一键恢复 Codex 原设置并关闭 gateway

config.example.json

@@ -12,6 +12,11 @@
   "request_history_limit": 200,
   "endpoints": ["/responses", "/chat/completions", "/v1/responses", "/v1/chat/completions"],
   "reasoning_equals": [516],
+  "retryable_status_codes": [429, 503],
+  "retryable_error_messages": [
+    "Selected model is at capacity. Please try a different model.",
+    "stream disconnected before completion: Concurrency limit exceeded for account, please retry later"
+  ],
   "non_stream_status_code": 502,
   "stream_action": "strict_502",
   "log_match": true,

scripts/admin-lib.mjs

@@ -177,16 +177,27 @@ export function normalizeIntArray(values, fallback = [516]) {
   return normalized.length > 0 ? [...new Set(normalized)] : [...fallback];
 }
 
-export function normalizeStringArray(values, fallback = []) {
-  const source = values === undefined || values === null ? fallback : values;
-  const queue = Array.isArray(source) ? source.flat(Infinity) : [source];
-  const normalized = queue
+export function normalizeStringArray(values, fallback = []) {
+  const source = values === undefined || values === null ? fallback : values;
+  const queue = Array.isArray(source) ? source.flat(Infinity) : [source];
+  const normalized = queue
     .flatMap((value) => `${value ?? ""}`.split(/[\s,]+/))
     .map((value) => value.trim())
     .filter(Boolean);
-
-  return normalized.length > 0 ? [...new Set(normalized)] : [...fallback];
-}
+
+  return normalized.length > 0 ? [...new Set(normalized)] : [...fallback];
+}
+
+export function normalizePhraseArray(values, fallback = []) {
+  const source = values === undefined || values === null ? fallback : values;
+  const queue = Array.isArray(source) ? source.flat(Infinity) : [source];
+  const normalized = queue
+    .flatMap((value) => (typeof value === "string" ? value.split(/\r?\n/) : [value]))
+    .map((value) => `${value ?? ""}`.trim())
+    .filter(Boolean);
+
+  return normalized.length > 0 ? [...new Set(normalized)] : [...fallback];
+}
 
 export function isProcessAlive(processId) {
   try {
@@ -404,19 +415,24 @@ export async function installForCurrentProvider({
     }
   }
 
-  const gatewayConfig = {
-    listen_host: listenHost,
-    listen_port: listenPort,
-    upstream_base_url: originalBaseUrl,
+  const gatewayConfig = {
+    listen_host: listenHost,
+    listen_port: listenPort,
+    upstream_base_url: originalBaseUrl,
     request_body_limit_bytes:
       existingGatewayConfig?.request_body_limit_bytes === undefined || existingGatewayConfig?.request_body_limit_bytes === null
         ? 10485760
         : Number.parseInt(`${existingGatewayConfig.request_body_limit_bytes}`, 10),
-    endpoints: mergedEndpoints,
-    reasoning_equals: normalizeIntArray(existingGatewayConfig?.reasoning_equals, [516]),
-    non_stream_status_code:
-      existingGatewayConfig?.non_stream_status_code === undefined || existingGatewayConfig?.non_stream_status_code === null
-        ? 502
+    endpoints: mergedEndpoints,
+    reasoning_equals: normalizeIntArray(existingGatewayConfig?.reasoning_equals, [516]),
+    retryable_status_codes: normalizeIntArray(existingGatewayConfig?.retryable_status_codes, [429, 503]),
+    retryable_error_messages: normalizePhraseArray(existingGatewayConfig?.retryable_error_messages, [
+      "Selected model is at capacity. Please try a different model.",
+      "stream disconnected before completion: Concurrency limit exceeded for account, please retry later",
+    ]),
+    non_stream_status_code:
+      existingGatewayConfig?.non_stream_status_code === undefined || existingGatewayConfig?.non_stream_status_code === null
+        ? 502
         : Number.parseInt(`${existingGatewayConfig.non_stream_status_code}`, 10),
     stream_action: existingGatewayConfig?.stream_action || "strict_502",
     log_match: existingGatewayConfig?.log_match === undefined ? true : Boolean(existingGatewayConfig.log_match),

scripts/run-profile.mjs

@@ -17,6 +17,7 @@ import {
   getGatewayBaseUrl,
   getGatewayStatePaths,
   normalizeIntArray,
+  normalizePhraseArray,
   normalizeStringArray,
   parseOptions,
   readJsonFile,
@@ -54,7 +55,15 @@ function parseEnvFile(content) {
 
 function getProfileName(options) {
   const positional = options._?.[0];
-  return `${options.profile || positional || process.env.CODEX_RETRY_GATEWAY_PROFILE || "default"}`;
+  return `${options.profile || positional || process.env.CODEX_RETRY_GATEWAY_PROFILE || ""}`.trim();
+}
+
+function resolveDefaultRequestedProfileName(existingState) {
+  const stateProfileName = `${existingState?.profile_name || ""}`.trim();
+  if (stateProfileName) {
+    return stateProfileName;
+  }
+  return process.env.CODEX_RETRY_GATEWAY_DEFAULT_PROFILE || "default";
 }
 
 function resolvePreferredProfileName({ requestedProfileName, existingState, preferStateProfile, profilesDir }) {
@@ -185,6 +194,23 @@ function buildProfileConfig({ profileName, profileEnv, existingGatewayConfig, pr
       profileEnv.CODEX_RETRY_GATEWAY_REASONING_EQUALS || existingGatewayConfig?.reasoning_equals,
       [516],
     ),
+    retryable_status_codes: normalizeIntArray(
+      profileEnv.CODEX_RETRY_GATEWAY_RETRYABLE_STATUS_CODES || existingGatewayConfig?.retryable_status_codes,
+      [429, 503],
+    ),
+    retryable_error_messages: normalizePhraseArray(
+      profileEnv.CODEX_RETRY_GATEWAY_RETRYABLE_ERROR_MESSAGES || existingGatewayConfig?.retryable_error_messages,
+      [
+        "Selected model is at capacity. Please try a different model.",
+        "stream disconnected before completion: Concurrency limit exceeded for account, please retry later",
+      ],
+    ),
+    upstream_fetch_retry_attempts: profileEnv.CODEX_RETRY_GATEWAY_UPSTREAM_FETCH_RETRY_ATTEMPTS
+      ? Number.parseInt(`${profileEnv.CODEX_RETRY_GATEWAY_UPSTREAM_FETCH_RETRY_ATTEMPTS}`, 10)
+      : Number.parseInt(`${existingGatewayConfig?.upstream_fetch_retry_attempts || 5}`, 10),
+    upstream_fetch_retry_backoff_ms: profileEnv.CODEX_RETRY_GATEWAY_UPSTREAM_FETCH_RETRY_BACKOFF_MS
+      ? Number.parseInt(`${profileEnv.CODEX_RETRY_GATEWAY_UPSTREAM_FETCH_RETRY_BACKOFF_MS}`, 10)
+      : Number.parseInt(`${existingGatewayConfig?.upstream_fetch_retry_backoff_ms || 350}`, 10),
     non_stream_status_code: profileEnv.CODEX_RETRY_GATEWAY_NON_STREAM_STATUS_CODE
       ? Number.parseInt(`${profileEnv.CODEX_RETRY_GATEWAY_NON_STREAM_STATUS_CODE}`, 10)
       : Number.parseInt(`${existingGatewayConfig?.non_stream_status_code || 502}`, 10),
@@ -231,7 +257,6 @@ async function ensureCodexPointsToGateway({ paths, codexConfigPath, providerCont
 
 async function main() {
   const options = parseOptions(process.argv, { booleanFlags: ["no-codex-config-update", "prefer-state-profile"] });
-  const requestedProfileName = getProfileName(options);
   const profilesDir = options.profilesDir || DEFAULT_PROFILES_DIR;
   const stateRoot = options.stateRoot || process.env.CODEX_RETRY_GATEWAY_STATE_ROOT || DEFAULT_STATE_ROOT;
   const codexConfigPath =
@@ -246,6 +271,7 @@ async function main() {
   await ensureDirectory(paths.backupDir);
 
   const existingState = await readJsonFile(paths.statePath);
+  const requestedProfileName = getProfileName(options) || resolveDefaultRequestedProfileName(existingState);
   const profileName = resolvePreferredProfileName({
     requestedProfileName,
     existingState,

scripts/test-gateway-e2e.mjs

@@ -39,24 +39,24 @@ function createJsonResponse(res, statusCode, body, extraHeaders = {}) {
   res.end(JSON.stringify(body));
 }
 
-function createSseResponse(res, chunks) {
-  res.writeHead(200, {
-    "content-type": "text/event-stream; charset=utf-8",
-    "cache-control": "no-cache",
-    connection: "keep-alive",
+function createSseResponse(res, chunks, intervalMs = 20) {
+  res.writeHead(200, {
+    "content-type": "text/event-stream; charset=utf-8",
+    "cache-control": "no-cache",
+    connection: "keep-alive",
     "x-upstream-test": "sse",
   });
 
   let index = 0;
   const timer = setInterval(() => {
-    if (index >= chunks.length) {
-      clearInterval(timer);
-      res.end();
-      return;
-    }
-    res.write(chunks[index]);
-    index += 1;
-  }, 20);
+    if (index >= chunks.length) {
+      clearInterval(timer);
+      res.end();
+      return;
+    }
+    res.write(chunks[index]);
+    index += 1;
+  }, intervalMs);
 
   res.on("close", () => {
     clearInterval(timer);
@@ -80,8 +80,48 @@ function createTerminatedSseResponse(res, chunks, destroyDelayMs = 20) {
   }, destroyDelayMs);
 }
 
+function buildCapacityStreamPayload(message, shape = "default") {
+  if (shape === "response_failed") {
+    return {
+      type: "response.failed",
+      response: {
+        status: "failed",
+        error: {
+          message,
+          type: "server_error",
+        },
+      },
+    };
+  }
+  return {
+    error: {
+      message,
+      type: "server_error",
+    },
+  };
+}
+
+function createCapacityErrorSseResponse(
+  res,
+  message = "Selected model is at capacity. Please try a different model.",
+  intervalMs = 20,
+  options = {},
+) {
+  const eventName = options.eventName || "error";
+  const payload = options.payload || buildCapacityStreamPayload(message, options.payloadShape || "default");
+  createSseResponse(
+    res,
+    [
+      `event: ${eventName}\n`,
+      `data: ${JSON.stringify(payload)}\n\n`,
+    ],
+    intervalMs,
+  );
+}
+
 function startFakeUpstream(port) {
   const failBeforeResponseCounts = new Map();
+  const capacityBeforeSuccessCounts = new Map();
   const server = http.createServer((req, res) => {
     const responsePaths = new Set(["/responses", "/v1/responses"]);
     const chatCompletionPaths = new Set(["/chat/completions", "/v1/chat/completions"]);
@@ -123,12 +163,79 @@ function startFakeUpstream(port) {
           ]);
           return;
         }
+        if (parsed.test_capacity_error) {
+          const capacityMessage = parsed.test_capacity_message || "Selected model is at capacity. Please try a different model.";
+          createJsonResponse(
+            res,
+            parsed.test_capacity_status ?? 503,
+            {
+              error: {
+                message: capacityMessage,
+                type: "server_error",
+              },
+            },
+            { "x-upstream-test": `capacity-error-${parsed.test_capacity_status ?? 503}` },
+          );
+          return;
+        }
+        if (parsed.test_capacity_before_success_times) {
+          const capacityKey = [
+            req.url,
+            "capacity-before-success",
+            parsed.test_capacity_before_success_times,
+            parsed.stream ? "stream" : "non-stream",
+            parsed.test_capacity_status ?? "default",
+            parsed.test_capacity_stream_event_name || "",
+            parsed.test_capacity_stream_payload_shape || "",
+          ].join(":");
+          const capacityCount = (capacityBeforeSuccessCounts.get(capacityKey) || 0) + 1;
+          capacityBeforeSuccessCounts.set(capacityKey, capacityCount);
+          if (capacityCount <= parsed.test_capacity_before_success_times) {
+            const capacityMessage = parsed.test_capacity_message || "Selected model is at capacity. Please try a different model.";
+            if (parsed.stream) {
+              createCapacityErrorSseResponse(
+                res,
+                capacityMessage,
+                20,
+                {
+                  eventName: parsed.test_capacity_stream_event_name || "error",
+                  payloadShape: parsed.test_capacity_stream_payload_shape || "default",
+                },
+              );
+              return;
+            }
+            createJsonResponse(
+              res,
+              parsed.test_capacity_status ?? 503,
+              {
+                error: {
+                  message: capacityMessage,
+                  type: "server_error",
+                },
+              },
+              { "x-upstream-test": `capacity-error-${parsed.test_capacity_status ?? 503}` },
+            );
+            return;
+          }
+        }
+        if (parsed.stream && parsed.test_capacity_error) {
+          createCapacityErrorSseResponse(
+            res,
+            parsed.test_capacity_message || "Selected model is at capacity. Please try a different model.",
+            20,
+            {
+              eventName: parsed.test_capacity_stream_event_name || "error",
+              payloadShape: parsed.test_capacity_stream_payload_shape || "default",
+            },
+          );
+          return;
+        }
         if (parsed.stream) {
           createSseResponse(res, [
-            'data: {"type":"response.output_text.delta","delta":"hello"}\n\n',
+            `data: ${JSON.stringify({ type: "response.output_text.delta", delta: "hello", response_id: "resp_stream", thread_id: parsed.thread_id || "thread_stream" })}\n\n`,
             `data: {"response":{"usage":{"output_tokens_details":{"reasoning_tokens":${reasoning}}}}}\n\n`,
             "data: [DONE]\n\n",
-          ]);
+          ], parsed.test_stream_chunk_delay_ms ?? 20);
           return;
         }
         createJsonResponse(
@@ -136,6 +243,7 @@ function startFakeUpstream(port) {
           200,
           {
             id: "resp_test",
+            thread_id: parsed.thread_id || "thread_test",
             retry_attempt: parsed.test_fail_before_response_once
               ? failBeforeResponseCounts.get(`${req.url}:fail-before-response-once`) || 0
               : 0,
@@ -269,22 +377,29 @@ async function run() {
   const logPath = path.join(tempRoot, "gateway.log");
 
   const config = {
-    listen_host: "127.0.0.1",
-    listen_port: gatewayPort,
-    upstream_base_url: `http://127.0.0.1:${upstreamPort}`,
-    request_body_limit_bytes: 10 * 1024 * 1024,
-    endpoints: ["/responses", "/chat/completions", "/v1/responses", "/v1/chat/completions"],
-    reasoning_equals: [516],
-    non_stream_status_code: 502,
+    listen_host: "127.0.0.1",
+    listen_port: gatewayPort,
+    upstream_base_url: `http://127.0.0.1:${upstreamPort}`,
+    request_body_limit_bytes: 10 * 1024 * 1024,
+    endpoints: ["/responses", "/chat/completions", "/v1/responses", "/v1/chat/completions"],
+    reasoning_equals: [516],
+    retryable_status_codes: [429, 503],
+    retryable_error_messages: [
+      "Selected model is at capacity. Please try a different model.",
+      "stream disconnected before completion: Concurrency limit exceeded for account, please retry later",
+    ],
+    upstream_fetch_retry_attempts: 5,
+    upstream_fetch_retry_backoff_ms: 25,
+    non_stream_status_code: 502,
     stream_action: "strict_502",
-    log_match: true,
-    health_path: "/__codex_retry_gateway/health",
-  };
+    log_match: true,
+    health_path: "/__codex_retry_gateway/health",
+  };
 
   await writeFile(configPath, JSON.stringify(config, null, 2), "utf8");
 
-  const upstream = await startFakeUpstream(upstreamPort);
-  const gateway = startGateway(configPath, logPath);
+  const upstream = await startFakeUpstream(upstreamPort);
+  let gateway = startGateway(configPath, logPath);
 
   try {
     try {
@@ -348,6 +463,256 @@ async function run() {
       recoveredEntry?.request_body_bytes === Buffer.byteLength(recoveredPayload),
       `请求体大小记录异常: ${recoveredEntry?.request_body_bytes}`,
     );
+    assert(recoveredEntry?.request_id, "请求记录未生成 request_id");
+
+    const threadTrackedResponse = await fetch(`http://127.0.0.1:${gatewayPort}/responses`, {
+      method: "POST",
+      headers: { "content-type": "application/json" },
+      body: JSON.stringify({ test_reasoning_tokens: 128, thread_id: "thread_nonstream" }),
+    });
+    const threadTrackedBody = await threadTrackedResponse.json();
+    assert(threadTrackedResponse.status === 200, `thread non-stream 请求失败: ${threadTrackedResponse.status}`);
+    assert(threadTrackedBody?.id === "resp_test", "thread non-stream 返回体缺少 response id");
+    assert(threadTrackedBody?.thread_id === "thread_nonstream", "thread non-stream 返回体缺少 thread_id");
+
+    const threadRequestsResponse = await fetch(`http://127.0.0.1:${gatewayPort}/__codex_retry_gateway/api/requests?query=${encodeURIComponent("thread_nonstream")}`);
+    const threadRequestsPayload = await threadRequestsResponse.json();
+    const threadEntry = (threadRequestsPayload?.entries || []).find((entry) => entry.thread_id === "thread_nonstream");
+    assert(threadEntry?.response_id === "resp_test", "non-stream 请求记录未保留 response_id");
+    assert(threadEntry?.thread_id === "thread_nonstream", "non-stream 请求记录未保留 thread_id");
+
+    const sameRequestPayload = JSON.stringify({ test_reasoning_tokens: 128, test_request_id_marker: "same" });
+    const sameRequestFirstResponse = await fetch(`http://127.0.0.1:${gatewayPort}/responses`, {
+      method: "POST",
+      headers: { "content-type": "application/json" },
+      body: sameRequestPayload,
+    });
+    assert(sameRequestFirstResponse.status === 200, `相同请求首次发送失败: ${sameRequestFirstResponse.status}`);
+    await sameRequestFirstResponse.json();
+
+    const sameRequestSecondResponse = await fetch(`http://127.0.0.1:${gatewayPort}/responses`, {
+      method: "POST",
+      headers: { "content-type": "application/json" },
+      body: sameRequestPayload,
+    });
+    assert(sameRequestSecondResponse.status === 200, `相同请求二次发送失败: ${sameRequestSecondResponse.status}`);
+    await sameRequestSecondResponse.json();
+
+    const differentBodyPayload = JSON.stringify({ test_reasoning_tokens: 128, test_request_id_marker: "different" });
+    const differentBodyResponse = await fetch(`http://127.0.0.1:${gatewayPort}/responses`, {
+      method: "POST",
+      headers: { "content-type": "application/json" },
+      body: differentBodyPayload,
+    });
+    assert(differentBodyResponse.status === 200, `不同请求体发送失败: ${differentBodyResponse.status}`);
+    await differentBodyResponse.json();
+
+    const differentPathResponse = await fetch(`http://127.0.0.1:${gatewayPort}/v1/responses`, {
+      method: "POST",
+      headers: { "content-type": "application/json" },
+      body: sameRequestPayload,
+    });
+    assert(differentPathResponse.status === 200, `不同路径发送失败: ${differentPathResponse.status}`);
+    await differentPathResponse.json();
+
+    const requestIdRequestsResponse = await fetch(`http://127.0.0.1:${gatewayPort}/__codex_retry_gateway/api/requests?limit=40`);
+    const requestIdRequestsPayload = await requestIdRequestsResponse.json();
+    const sameRequestEntries = (requestIdRequestsPayload?.entries || []).filter(
+      (entry) => entry.path === "/responses" && entry.request_body_bytes === Buffer.byteLength(sameRequestPayload),
+    );
+    assert(sameRequestEntries.length >= 2, "未找到两条相同请求记录");
+    const [sameRequestEntryA, sameRequestEntryB] = sameRequestEntries;
+    assert(sameRequestEntryA.request_id, "相同请求记录缺少 request_id");
+    assert(sameRequestEntryA.request_id === sameRequestEntryB.request_id, "相同请求未复用 request_id");
+    assert(sameRequestEntryA.seq !== sameRequestEntryB.seq, "相同请求不应复用 seq");
+
+    const differentBodyEntry = (requestIdRequestsPayload?.entries || []).find(
+      (entry) =>
+        entry.path === "/responses" &&
+        entry.request_body_bytes !== Buffer.byteLength(sameRequestPayload) &&
+        entry.request_body_bytes === Buffer.byteLength(differentBodyPayload),
+    );
+    assert(differentBodyEntry?.request_id, "不同请求体记录缺少 request_id");
+    assert(differentBodyEntry.request_id !== sameRequestEntryA.request_id, "不同请求体错误复用 request_id");
+
+    const differentPathEntry = (requestIdRequestsPayload?.entries || []).find(
+      (entry) => entry.path === "/v1/responses" && entry.request_body_bytes === Buffer.byteLength(sameRequestPayload),
+    );
+    assert(differentPathEntry?.request_id, "不同路径记录缺少 request_id");
+    assert(differentPathEntry.request_id !== sameRequestEntryA.request_id, "不同路径错误复用 request_id");
+
+    const requestIdQueryResponse = await fetch(
+      `http://127.0.0.1:${gatewayPort}/__codex_retry_gateway/api/requests?query=${encodeURIComponent(sameRequestEntryA.request_id)}`,
+    );
+    const requestIdQueryPayload = await requestIdQueryResponse.json();
+    assert(requestIdQueryResponse.status === 200, `request_id 搜索失败: ${requestIdQueryResponse.status}`);
+    assert(
+      (requestIdQueryPayload?.entries || []).some((entry) => entry.request_id === sameRequestEntryA.request_id),
+      "request_id 搜索未命中对应记录",
+    );
+
+    const capacityResponse = await fetch(`http://127.0.0.1:${gatewayPort}/responses`, {
+      method: "POST",
+      headers: { "content-type": "application/json" },
+      body: JSON.stringify({ test_capacity_error: true }),
+    });
+    const capacityBody = await capacityResponse.json();
+    assert(capacityResponse.status === 502, `capacity error 未返回 502: ${capacityResponse.status}`);
+    assert(
+      capacityBody?.error?.code === "upstream_error_retry_triggered",
+      "capacity error 返回体未标记 retry trigger",
+    );
+    assert(
+      capacityBody?.error?.upstream_status_code === 503,
+      "capacity error 返回体未保留 upstream status",
+    );
+
+    const capacityStatus200Response = await fetch(`http://127.0.0.1:${gatewayPort}/responses`, {
+      method: "POST",
+      headers: { "content-type": "application/json" },
+      body: JSON.stringify({ test_capacity_error: true, test_capacity_status: 200 }),
+    });
+    const capacityStatus200Body = await capacityStatus200Response.json();
+    assert(capacityStatus200Response.status === 502, `200+capacity error 未返回 502: ${capacityStatus200Response.status}`);
+    assert(
+      capacityStatus200Body?.error?.code === "upstream_error_retry_triggered",
+      "200+capacity error 返回体未标记 retry trigger",
+    );
+    assert(
+      capacityStatus200Body?.error?.upstream_status_code === 200,
+      "200+capacity error 返回体未保留 upstream status",
+    );
+
+    const capacityRecoveredResponse = await fetch(`http://127.0.0.1:${gatewayPort}/responses`, {
+      method: "POST",
+      headers: { "content-type": "application/json" },
+      body: JSON.stringify({ test_capacity_before_success_times: 2, test_reasoning_tokens: 128 }),
+    });
+    const capacityRecoveredBody = await capacityRecoveredResponse.json();
+    assert(capacityRecoveredResponse.status === 200, `capacity 抖动后未自动恢复: ${capacityRecoveredResponse.status}`);
+    assert(
+      capacityRecoveredBody?.usage?.output_tokens_details?.reasoning_tokens === 128,
+      "capacity 抖动恢复后的返回体异常",
+    );
+
+    const requestsAfterCapacityRecoveryResponse = await fetch(`http://127.0.0.1:${gatewayPort}/__codex_retry_gateway/api/requests?limit=20`);
+    const requestsAfterCapacityRecovery = await requestsAfterCapacityRecoveryResponse.json();
+    const capacityRecoveredEntry = requestsAfterCapacityRecovery?.entries?.find(
+      (entry) => entry.path === "/responses" && entry.status_code === 200 && entry.upstream_attempt_count >= 3,
+    );
+    assert(capacityRecoveredEntry, "capacity 抖动恢复后的请求记录未保留重试次数");
+    assert(capacityRecoveredEntry.request_id, "capacity 恢复后的请求记录缺少 request_id");
+
+    const capacityStatus200RecoveredResponse = await fetch(`http://127.0.0.1:${gatewayPort}/responses`, {
+      method: "POST",
+      headers: { "content-type": "application/json" },
+      body: JSON.stringify({ test_capacity_before_success_times: 2, test_capacity_status: 200, test_reasoning_tokens: 128 }),
+    });
+    const capacityStatus200RecoveredBody = await capacityStatus200RecoveredResponse.json();
+    assert(capacityStatus200RecoveredResponse.status === 200, `200+capacity 抖动后未自动恢复: ${capacityStatus200RecoveredResponse.status}`);
+    assert(
+      capacityStatus200RecoveredBody?.usage?.output_tokens_details?.reasoning_tokens === 128,
+      "200+capacity 抖动恢复后的返回体异常",
+    );
+
+    const requestsAfterCapacityStatus200RecoveryResponse = await fetch(`http://127.0.0.1:${gatewayPort}/__codex_retry_gateway/api/requests?limit=30`);
+    const requestsAfterCapacityStatus200Recovery = await requestsAfterCapacityStatus200RecoveryResponse.json();
+    const capacityStatus200RecoveredEntry = requestsAfterCapacityStatus200Recovery?.entries?.find(
+      (entry) => entry.path === "/responses" && entry.status_code === 200 && entry.upstream_attempt_count >= 3,
+    );
+    assert(capacityStatus200RecoveredEntry, "200+capacity 抖动恢复后的请求记录未保留重试次数");
+
+    const streamCapacityResponse = await fetch(`http://127.0.0.1:${gatewayPort}/responses`, {
+      method: "POST",
+      headers: { "content-type": "application/json" },
+      body: JSON.stringify({ stream: true, test_capacity_error: true }),
+    });
+    const streamCapacityBody = await streamCapacityResponse.json();
+    assert(streamCapacityResponse.status === 502, `stream+capacity error 未返回 502: ${streamCapacityResponse.status}`);
+    assert(
+      streamCapacityBody?.error?.code === "upstream_error_retry_triggered",
+      "stream+capacity error 返回体未标记 retry trigger",
+    );
+
+    const streamCapacityResponseFailed = await fetch(`http://127.0.0.1:${gatewayPort}/responses`, {
+      method: "POST",
+      headers: { "content-type": "application/json" },
+      body: JSON.stringify({
+        stream: true,
+        test_capacity_error: true,
+        test_capacity_stream_event_name: "response.failed",
+        test_capacity_stream_payload_shape: "response_failed",
+      }),
+    });
+    const streamCapacityResponseFailedBody = await streamCapacityResponseFailed.json();
+    assert(streamCapacityResponseFailed.status === 502, `stream+response.failed capacity error 未返回 502: ${streamCapacityResponseFailed.status}`);
+    assert(
+      streamCapacityResponseFailedBody?.error?.code === "upstream_error_retry_triggered",
+      "stream+response.failed capacity error 返回体未标记 retry trigger",
+    );
+
+    const streamCapacityRecoveredResponse = await fetch(`http://127.0.0.1:${gatewayPort}/responses`, {
+      method: "POST",
+      headers: { "content-type": "application/json" },
+      body: JSON.stringify({ stream: true, test_capacity_before_success_times: 2, test_reasoning_tokens: 128 }),
+    });
+    const streamCapacityRecoveredText = await streamCapacityRecoveredResponse.text();
+    assert(streamCapacityRecoveredResponse.status === 200, `stream capacity 抖动后未自动恢复: ${streamCapacityRecoveredResponse.status}`);
+    assert(streamCapacityRecoveredText.includes("hello"), "stream capacity 恢复后未拿到正常 SSE 内容");
+
+    const requestsAfterStreamCapacityRecoveryResponse = await fetch(`http://127.0.0.1:${gatewayPort}/__codex_retry_gateway/api/requests?limit=20`);
+    const requestsAfterStreamCapacityRecovery = await requestsAfterStreamCapacityRecoveryResponse.json();
+    const streamCapacityRecoveredEntry = requestsAfterStreamCapacityRecovery?.entries?.find(
+      (entry) => entry.path === "/responses" && entry.status_code === 200 && entry.response_stream && entry.upstream_attempt_count >= 3,
+    );
+    assert(streamCapacityRecoveredEntry, "stream capacity 抖动恢复后的请求记录未保留重试次数");
+
+    const streamCapacityResponseFailedRecoveredResponse = await fetch(`http://127.0.0.1:${gatewayPort}/responses`, {
+      method: "POST",
+      headers: { "content-type": "application/json" },
+      body: JSON.stringify({
+        stream: true,
+        test_capacity_before_success_times: 2,
+        test_reasoning_tokens: 128,
+        test_capacity_stream_event_name: "response.failed",
+        test_capacity_stream_payload_shape: "response_failed",
+      }),
+    });
+    const streamCapacityResponseFailedRecoveredText = await streamCapacityResponseFailedRecoveredResponse.text();
+    assert(streamCapacityResponseFailedRecoveredResponse.status === 200, `stream response.failed capacity 抖动后未自动恢复: ${streamCapacityResponseFailedRecoveredResponse.status}`);
+    assert(streamCapacityResponseFailedRecoveredText.includes("hello"), "stream response.failed capacity 恢复后未拿到正常 SSE 内容");
+
+    const requestsAfterStreamCapacityResponseFailedRecoveryResponse = await fetch(`http://127.0.0.1:${gatewayPort}/__codex_retry_gateway/api/requests?limit=30`);
+    const requestsAfterStreamCapacityResponseFailedRecovery = await requestsAfterStreamCapacityResponseFailedRecoveryResponse.json();
+    const streamCapacityResponseFailedRecoveredEntry = requestsAfterStreamCapacityResponseFailedRecovery?.entries?.find(
+      (entry) => entry.path === "/responses" && entry.status_code === 200 && entry.response_stream && entry.upstream_attempt_count >= 3,
+    );
+    assert(streamCapacityResponseFailedRecoveredEntry, "stream response.failed capacity 恢复后的请求记录未保留重试次数");
+
+    const streamThreadResponse = await readSseUntilClose(
+      `http://127.0.0.1:${gatewayPort}/responses`,
+      { stream: true, test_reasoning_tokens: 128, thread_id: "thread_stream_ok" },
+    );
+    assert(streamThreadResponse.status === 200, `stream thread 请求失败: ${streamThreadResponse.status}`);
+    const streamThreadRequestsResponse = await fetch(`http://127.0.0.1:${gatewayPort}/__codex_retry_gateway/api/requests?query=${encodeURIComponent("thread_stream_ok")}`);
+    const streamThreadRequestsPayload = await streamThreadRequestsResponse.json();
+    const streamThreadEntry = (streamThreadRequestsPayload?.entries || []).find((entry) => entry.thread_id === "thread_stream_ok");
+    assert(streamThreadEntry?.response_id === "resp_stream", "stream 请求记录未保留 response_id");
+    assert(streamThreadEntry?.thread_id === "thread_stream_ok", "stream 请求记录未保留 thread_id");
+
+    const streamDisconnectedRetryResponse = await fetch(`http://127.0.0.1:${gatewayPort}/responses`, {
+      method: "POST",
+      headers: { "content-type": "application/json" },
+      body: JSON.stringify({
+        stream: true,
+        test_capacity_before_success_times: 2,
+        test_capacity_message: "stream disconnected before completion: Concurrency limit exceeded for account, please retry later",
+        test_reasoning_tokens: 128,
+      }),
+    });
+    const streamDisconnectedRetryText = await streamDisconnectedRetryResponse.text();
+    assert(streamDisconnectedRetryResponse.status === 200, `stream disconnected capacity 抖动后未自动恢复: ${streamDisconnectedRetryResponse.status}`);
+    assert(streamDisconnectedRetryText.includes("hello"), "stream disconnected capacity 恢复后未拿到正常 SSE 内容");
 
     for (const streamPath of [
       "/responses",
@@ -377,12 +742,62 @@ async function run() {
       assert(!okStream.closedByError, `${streamPath} 流式 128 不应异常断开`);
     }
 
+    const streamProgressPromise = fetch(`http://127.0.0.1:${gatewayPort}/responses`, {
+      method: "POST",
+      headers: { "content-type": "application/json" },
+      body: JSON.stringify({ stream: true, test_reasoning_tokens: 128, test_stream_chunk_delay_ms: 180 }),
+    });
+    await new Promise((resolve) => setTimeout(resolve, 260));
+    const midRequestsResponse = await fetch(`http://127.0.0.1:${gatewayPort}/__codex_retry_gateway/api/requests?limit=20`);
+    const midRequestsPayload = await midRequestsResponse.json();
+    const inFlightStreamEntry = midRequestsPayload?.entries?.find(
+      (entry) =>
+        entry.path === "/responses" &&
+        entry.response_stream === true &&
+        (entry.lifecycle_state === "streaming" || entry.lifecycle_state === "receive_first") &&
+        (entry.stream_chunk_count || 0) >= 1,
+    );
+    assert(inFlightStreamEntry, "流式请求过程中未暴露进行中状态");
+    assert(
+      (inFlightStreamEntry?.response_bytes_received || 0) > 0,
+      "流式请求过程中未累计接收字节数",
+    );
+    const streamProgressResponse = await streamProgressPromise;
+    assert(streamProgressResponse.status === 200, `stream progress 响应状态异常: ${streamProgressResponse.status}`);
+    const streamReader = streamProgressResponse.body.getReader();
+    while (true) {
+      const { done } = await streamReader.read();
+      if (done) {
+        break;
+      }
+    }
+
     const terminatedStream = await readSseUntilClose(
       `http://127.0.0.1:${gatewayPort}/responses`,
       { stream: true, test_force_terminate: true },
     );
     assert(terminatedStream.status === 502, `/responses 上游半路断流未返回 502: ${terminatedStream.status}`);
 
+    const metricsBeforeRestartResponse = await fetch(`http://127.0.0.1:${gatewayPort}/__codex_retry_gateway/api/status`);
+    const metricsBeforeRestart = await metricsBeforeRestartResponse.json();
+    assert(metricsBeforeRestartResponse.status === 200, `status API 状态异常: ${metricsBeforeRestartResponse.status}`);
+    assert(metricsBeforeRestart?.metrics?.reasoning_516_count >= 1, "重启前 reasoning_516_count 未累计");
+    assert(metricsBeforeRestart?.metrics?.observed_reasoning_counts?.["128"] >= 1, "重启前 reasoning 128 未累计");
+    assert(metricsBeforeRestart?.metrics?.total_proxy_request_count >= 1, "重启前 total_proxy_request_count 未累计");
+
+    gateway.child.kill();
+    await once(gateway.child, "exit");
+    gateway = startGateway(configPath, logPath);
+    await waitForHealth(`http://127.0.0.1:${gatewayPort}${config.health_path}`);
+
+    const metricsAfterRestartResponse = await fetch(`http://127.0.0.1:${gatewayPort}/__codex_retry_gateway/api/status`);
+    const metricsAfterRestart = await metricsAfterRestartResponse.json();
+    assert(metricsAfterRestartResponse.status === 200, `重启后 status API 状态异常: ${metricsAfterRestartResponse.status}`);
+    assert(metricsAfterRestart?.metrics?.reasoning_516_count >= metricsBeforeRestart?.metrics?.reasoning_516_count, "重启后 reasoning_516_count 未保留");
+    assert(metricsAfterRestart?.metrics?.observed_reasoning_counts?.["128"] >= metricsBeforeRestart?.metrics?.observed_reasoning_counts?.["128"], "重启后 reasoning 128 计数未保留");
+    assert(metricsAfterRestart?.metrics?.total_proxy_request_count >= metricsBeforeRestart?.metrics?.total_proxy_request_count, "重启后 total_proxy_request_count 未保留");
+    assert(metricsAfterRestart?.metrics?.persistent_since, "重启后未返回 persistent_since");
+
     await new Promise((resolve) => setTimeout(resolve, 120));
     const logText = await readFile(logPath, "utf8");
     assert(
@@ -391,9 +806,9 @@ async function run() {
     );
 
     process.stdout.write("PASS codex-retry-gateway e2e\n");
-  } finally {
-    gateway.child.kill();
-    upstream.close();
+  } finally {
+    gateway.child.kill();
+    upstream.close();
     await once(upstream, "close");
     await rm(tempRoot, { recursive: true, force: true });
   }

scripts/test-install-restore.mjs

@@ -221,6 +221,8 @@ async function run() {
       headers: { "content-type": "application/json" },
       body: JSON.stringify({
         reasoning_equals: [1024],
+        retryable_status_codes: [429, 503, 529],
+        retryable_error_messages: ["Selected model is at capacity. Please try a different model."],
         endpoints: ["/responses", "/v1/responses"],
         non_stream_status_code: 503,
         log_match: false,
@@ -237,6 +239,10 @@ async function run() {
       JSON.stringify(updatedGatewayConfig.reasoning_equals) === JSON.stringify([1024]),
       "Saved config file did not persist reasoning_equals",
     );
+    assert(
+      JSON.stringify(updatedGatewayConfig.retryable_status_codes) === JSON.stringify([429, 503, 529]),
+      "Saved config file did not persist retryable_status_codes",
+    );
 
     const incrementalLogsResponse = await fetch(
       `http://127.0.0.1:${gatewayPort}/__codex_retry_gateway/api/logs?since_seq=${logsPayload.latest_seq}`,

ui-src/src/App.tsx

@@ -1,4 +1,4 @@
-import { FormEvent, useEffect, useState } from "react";
+import { FormEvent, startTransition, useEffect, useState } from "react";
 
 type PageKey = "overview" | "requests" | "profiles" | "rules" | "logs";
 type Tone = "" | "success" | "error";
@@ -15,12 +15,17 @@ type GatewayConfig = {
   model_remap?: string;
   endpoints?: string[];
   reasoning_equals?: number[];
+  retryable_status_codes?: number[];
+  retryable_error_messages?: string[];
+  upstream_fetch_retry_attempts?: number;
+  upstream_fetch_retry_backoff_ms?: number;
   non_stream_status_code?: number;
   log_match?: boolean;
 };
 
 type Metrics = {
   started_at?: string;
+  persistent_since?: string | null;
   total_proxy_request_count?: number;
   inspected_response_count?: number;
   matched_response_count?: number;
@@ -34,6 +39,8 @@ type Metrics = {
     cached_tokens?: number;
   };
   observed_reasoning_counts?: Record<string, number>;
+  observed_reasoning_counts_total_keys?: number;
+  observed_reasoning_counts_omitted?: number;
 };
 
 type StatusPayload = {
@@ -63,20 +70,28 @@ type Usage = {
 
 type RequestEntry = {
   seq: number;
+  request_id?: string | null;
+  response_id?: string | null;
+  thread_id?: string | null;
   lifecycle_state?: string | null;
   started_at?: string;
   first_response_at?: string | null;
   first_response_delay_ms?: number | null;
   finished_at?: string | null;
+  last_activity_at?: string | null;
   duration_ms?: number | null;
   profile_name?: string;
   method?: string;
   path?: string;
   request_body_bytes?: number | null;
+  response_bytes_received?: number | null;
   model?: string | null;
   requested_model?: string | null;
   forwarded_model?: string | null;
   response_stream?: boolean;
+  stream_chunk_count?: number | null;
+  usage_last_updated_at?: string | null;
+  upstream_attempt_count?: number | null;
   matched?: boolean;
   status_code?: number | null;
   upstream_status_code?: number | null;
@@ -114,6 +129,10 @@ type ProfileFormModel = {
   request_history_limit?: string;
   model_remap?: string;
   reasoning_equals?: string;
+  retryable_status_codes?: string;
+  retryable_error_messages?: string[];
+  upstream_fetch_retry_attempts?: string;
+  upstream_fetch_retry_backoff_ms?: string;
   endpoints?: string[];
 };
 
@@ -179,11 +198,19 @@ type ProfileFormState = {
   request_history_limit: string;
   model_remap: string;
   reasoning_equals: string;
+  retryable_status_codes: string;
+  retryable_error_messages: string;
+  upstream_fetch_retry_attempts: string;
+  upstream_fetch_retry_backoff_ms: string;
   endpoints: string;
 };
 
 type RuleFormState = {
   reasoning_equals: string;
+  retryable_status_codes: string;
+  retryable_error_messages: string;
+  upstream_fetch_retry_attempts: string;
+  upstream_fetch_retry_backoff_ms: string;
   endpoints: string;
   non_stream_status_code: string;
   log_match: boolean;
@@ -200,6 +227,20 @@ const api = {
   restore: "/__codex_retry_gateway/api/restore",
 };
 
+const REQUEST_PAGE_SIZE = 20;
+const LOG_PAGE_SIZE = 200;
+
+const zhNumberFormatter = new Intl.NumberFormat("zh-CN");
+const zhTimestampFormatter = new Intl.DateTimeFormat("zh-CN", {
+  year: "numeric",
+  month: "2-digit",
+  day: "2-digit",
+  hour: "2-digit",
+  minute: "2-digit",
+  second: "2-digit",
+  hourCycle: "h23",
+});
+
 type ProfileProbePayload = {
   ok: boolean;
   profile?: string;
@@ -222,7 +263,7 @@ type ProfileProbePayload = {
 const pageCopy: Record<PageKey, { title: string; subtitle: string; index: string }> = {
   overview: {
     title: "概览",
-    subtitle: "当前 gateway 运行态和本次启动以来的累计统计。",
+    subtitle: "当前 gateway 运行态，以及基于持久请求记录回灌的累计统计。",
     index: "01",
   },
   requests: {
@@ -263,12 +304,16 @@ const defaultProfileForm: ProfileFormState = {
   request_history_limit: "0",
   model_remap: "",
   reasoning_equals: "516,1034,1552",
+  retryable_status_codes: "429,503",
+  retryable_error_messages: "Selected model is at capacity. Please try a different model.\nstream disconnected before completion: Concurrency limit exceeded for account, please retry later",
+  upstream_fetch_retry_attempts: "5",
+  upstream_fetch_retry_backoff_ms: "350",
   endpoints: "/responses\n/chat/completions\n/v1/responses\n/v1/chat/completions",
 };
 
 function numberFormat(value: unknown) {
   return typeof value === "number" && Number.isFinite(value)
-    ? new Intl.NumberFormat("zh-CN").format(value)
+    ? zhNumberFormatter.format(value)
     : "-";
 }
 
@@ -277,13 +322,25 @@ function timestamp(value?: string | null) {
     return "-";
   }
   const date = new Date(value);
-  return Number.isNaN(date.getTime()) ? value : date.toLocaleString("zh-CN", { hour12: false });
+  return Number.isNaN(date.getTime()) ? value : zhTimestampFormatter.format(date);
 }
 
 function durationSeconds(value?: number | null) {
   return typeof value === "number" && Number.isFinite(value) ? `${(value / 1000).toFixed(2)} s` : "-";
 }
 
+function secondsSince(startedAt: string | null | undefined, updatedAt: string | null | undefined) {
+  if (!startedAt || !updatedAt) {
+    return "-";
+  }
+  const startedMs = new Date(startedAt).getTime();
+  const updatedMs = new Date(updatedAt).getTime();
+  if (Number.isNaN(startedMs) || Number.isNaN(updatedMs)) {
+    return "-";
+  }
+  return `${Math.max(0, (updatedMs - startedMs) / 1000).toFixed(1)} s`;
+}
+
 function bytesFormat(value?: number | null) {
   if (typeof value !== "number" || !Number.isFinite(value)) {
     return "-";
@@ -327,6 +384,10 @@ function cachedRatio(inputTokens?: number | null, cachedTokens?: number | null)
   return Math.max(0, Math.min(1, cached / inputTokens));
 }
 
+function requestPrimaryId(entry: RequestEntry) {
+  return entry.response_id || entry.request_id || "-";
+}
+
 function splitList(value: string) {
   return value
     .split(/[\s,]+/)
@@ -368,14 +429,69 @@ function profileFormFromStatus(status: StatusPayload | null): ProfileFormState {
     reasoning_equals: Array.isArray(config.reasoning_equals)
       ? config.reasoning_equals.join(",")
       : defaultProfileForm.reasoning_equals,
+    retryable_status_codes: Array.isArray(config.retryable_status_codes)
+      ? config.retryable_status_codes.join(",")
+      : defaultProfileForm.retryable_status_codes,
+    retryable_error_messages: Array.isArray(config.retryable_error_messages)
+      ? config.retryable_error_messages.join("\n")
+      : defaultProfileForm.retryable_error_messages,
+    upstream_fetch_retry_attempts: String(
+      config.upstream_fetch_retry_attempts ?? defaultProfileForm.upstream_fetch_retry_attempts,
+    ),
+    upstream_fetch_retry_backoff_ms: String(
+      config.upstream_fetch_retry_backoff_ms ?? defaultProfileForm.upstream_fetch_retry_backoff_ms,
+    ),
     endpoints: Array.isArray(config.endpoints) ? config.endpoints.join("\n") : defaultProfileForm.endpoints,
   };
 }
 
+function profileFormFromProfile(profile: Profile): ProfileFormState {
+  const form = profile.form || {};
+  return {
+    ...defaultProfileForm,
+    name: profile.name,
+    listen_host: form.listen_host || "",
+    listen_port: form.listen_port || "",
+    upstream_base_url: form.upstream_base_url || "",
+    auth_mode: (form.auth_mode as ProfileFormState["auth_mode"]) || "passthrough",
+    auth_env: form.auth_env || "",
+    auth_file: form.auth_file || "",
+    manual_secret: "",
+    manual_secret_file: form.manual_secret_file || "",
+    manual_secret_configured: Boolean(form.manual_secret_configured),
+    auth_json_path: form.auth_json_path || "",
+    auth_json_key: form.auth_json_key || "OPENAI_API_KEY",
+    request_history_limit: form.request_history_limit || defaultProfileForm.request_history_limit,
+    model_remap: form.model_remap || "",
+    reasoning_equals: form.reasoning_equals || "",
+    retryable_status_codes: form.retryable_status_codes || defaultProfileForm.retryable_status_codes,
+    retryable_error_messages: Array.isArray(form.retryable_error_messages)
+      ? form.retryable_error_messages.join("\n")
+      : defaultProfileForm.retryable_error_messages,
+    upstream_fetch_retry_attempts:
+      form.upstream_fetch_retry_attempts || defaultProfileForm.upstream_fetch_retry_attempts,
+    upstream_fetch_retry_backoff_ms:
+      form.upstream_fetch_retry_backoff_ms || defaultProfileForm.upstream_fetch_retry_backoff_ms,
+    endpoints: Array.isArray(form.endpoints) ? form.endpoints.join("\n") : "",
+  };
+}
+
 function ruleFormFromStatus(status: StatusPayload | null): RuleFormState {
   const config = status?.config || {};
   return {
     reasoning_equals: Array.isArray(config.reasoning_equals) ? config.reasoning_equals.join(", ") : "",
+    retryable_status_codes: Array.isArray(config.retryable_status_codes)
+      ? config.retryable_status_codes.join(", ")
+      : defaultProfileForm.retryable_status_codes,
+    retryable_error_messages: Array.isArray(config.retryable_error_messages)
+      ? config.retryable_error_messages.join("\n")
+      : defaultProfileForm.retryable_error_messages,
+    upstream_fetch_retry_attempts: String(
+      config.upstream_fetch_retry_attempts ?? defaultProfileForm.upstream_fetch_retry_attempts,
+    ),
+    upstream_fetch_retry_backoff_ms: String(
+      config.upstream_fetch_retry_backoff_ms ?? defaultProfileForm.upstream_fetch_retry_backoff_ms,
+    ),
     endpoints: Array.isArray(config.endpoints) ? config.endpoints.join("\n") : "",
     non_stream_status_code: String(config.non_stream_status_code || 502),
     log_match: Boolean(config.log_match),
@@ -389,6 +505,7 @@ export default function App() {
   });
   const [status, setStatus] = useState<StatusPayload | null>(null);
   const [requests, setRequests] = useState<RequestEntry[]>([]);
+  const [requestsTotal, setRequestsTotal] = useState(0);
   const [requestsMeta, setRequestsMeta] = useState("正在读取请求记录...");
   const [profiles, setProfiles] = useState<Profile[]>([]);
   const [profilesMeta, setProfilesMeta] = useState("正在读取 profiles...");
@@ -397,6 +514,7 @@ export default function App() {
   const [latestLogSeq, setLatestLogSeq] = useState(0);
   const [requestQuery, setRequestQuery] = useState("");
   const [requestFilter, setRequestFilter] = useState("all");
+  const [requestLimit, setRequestLimit] = useState(REQUEST_PAGE_SIZE);
   const [ruleForm, setRuleForm] = useState<RuleFormState>(ruleFormFromStatus(null));
   const [profileForm, setProfileForm] = useState<ProfileFormState>(defaultProfileForm);
   const [ruleMessage, setRuleMessage] = useState<MessageState>({ text: "", tone: "" });
@@ -406,6 +524,7 @@ export default function App() {
   const [deletingProfile, setDeletingProfile] = useState("");
   const [switchingTo, setSwitchingTo] = useState("");
   const [restoreRequested, setRestoreRequested] = useState(false);
+  const [shouldSyncProfileFormToActive, setShouldSyncProfileFormToActive] = useState(true);
 
   const metrics = status?.metrics || {};
   const tokens = metrics.token_totals || {};
@@ -419,16 +538,18 @@ export default function App() {
 
   async function loadStatus(refreshRuleForm = false) {
     const payload = await fetchJson<StatusPayload>(api.status);
-    setStatus(payload);
-    if (refreshRuleForm) {
-      setRuleForm(ruleFormFromStatus(payload));
-      setProfileForm(profileFormFromStatus(payload));
-    }
+    startTransition(() => {
+      setStatus(payload);
+      if (refreshRuleForm) {
+        setRuleForm(ruleFormFromStatus(payload));
+        setProfileForm(profileFormFromStatus(payload));
+      }
+    });
   }
 
-  async function loadRequests() {
+  async function loadRequests(limitOverride = requestLimit) {
     const url = new URL(api.requests, window.location.origin);
-    url.searchParams.set("limit", "100");
+    url.searchParams.set("limit", String(limitOverride));
     if (requestQuery.trim()) {
       url.searchParams.set("query", requestQuery.trim());
     }
@@ -437,49 +558,84 @@ export default function App() {
     }
     const payload = await fetchJson<RequestsPayload>(url.toString());
     const entries = payload.entries || [];
-    setRequests(entries);
-    setRequestsMeta(`查询命中 ${payload.total_entries ?? entries.length} 条，最新序号 ${payload.latest_seq ?? 0}。`);
+    startTransition(() => {
+      setRequests(entries);
+      setRequestsTotal(payload.total_entries ?? entries.length);
+      setRequestsMeta(
+        `查询命中 ${payload.total_entries ?? entries.length} 条，当前展示最近 ${entries.length} 条，最新序号 ${payload.latest_seq ?? 0}。`,
+      );
+    });
   }
 
   async function loadProfiles() {
     const payload = await fetchJson<ProfilesPayload>(api.profiles);
     const items = payload.profiles || [];
-    setProfiles(items);
-    setProfilesMeta(`目录：${payload.profiles_dir || "-"}；当前运行：${payload.active_profile || "-"}。`);
+    startTransition(() => {
+      setProfiles(items);
+      setProfilesMeta(`目录：${payload.profiles_dir || "-"}；当前运行：${payload.active_profile || "-"}。`);
+    });
   }
 
   async function loadLogs(incremental = false) {
     const url = new URL(api.logs, window.location.origin);
     if (incremental && latestLogSeq > 0) {
       url.searchParams.set("since_seq", String(latestLogSeq));
+    } else {
+      url.searchParams.set("limit", String(LOG_PAGE_SIZE));
     }
     const payload = await fetchJson<LogsPayload>(url.toString());
     const rendered = (payload.entries || [])
       .map((entry) => `${entry.at || "-"} ${entry.message || ""}`)
       .join("\n");
-    setLogs((current) => {
-      if (!incremental || latestLogSeq === 0) {
-        return rendered || "当前还没有日志。";
+    startTransition(() => {
+      setLogs((current) => {
+        if (!incremental || latestLogSeq === 0) {
+          return rendered || "当前还没有日志。";
+        }
+        return rendered ? `${current.trim()}\n${rendered}` : current;
+      });
+      setLogsMeta(`已载入 ${payload.total_entries ?? payload.entries?.length ?? 0} 条日志，最新序号 ${payload.latest_seq ?? latestLogSeq}。`);
+      if (typeof payload.latest_seq === "number") {
+        setLatestLogSeq(payload.latest_seq);
       }
-      return rendered ? `${current.trim()}\n${rendered}` : current;
     });
-    setLogsMeta(`已载入 ${payload.total_entries ?? payload.entries?.length ?? 0} 条日志，最新序号 ${payload.latest_seq ?? latestLogSeq}。`);
-    if (typeof payload.latest_seq === "number") {
-      setLatestLogSeq(payload.latest_seq);
+  }
+
+  async function loadPageData(targetPage: PageKey, { incrementalLogs = true } = {}) {
+    if (targetPage === "requests") {
+      await loadRequests();
+      return;
+    }
+    if (targetPage === "profiles") {
+      await loadProfiles();
+      return;
+    }
+    if (targetPage === "logs") {
+      await loadLogs(incrementalLogs && latestLogSeq > 0);
     }
   }
 
-  async function refreshLiveData() {
+  async function refreshLiveData(targetPage: PageKey = page) {
     if (restoreRequested || switchingTo) {
       return;
     }
-    await Promise.all([loadStatus(false), loadRequests(), loadProfiles(), loadLogs(true)]);
+    const tasks: Array<Promise<unknown>> = [loadStatus(false)];
+    if (targetPage !== "overview" && targetPage !== "rules") {
+      tasks.push(loadPageData(targetPage, { incrementalLogs: true }));
+    }
+    await Promise.all(tasks);
   }
 
   useEffect(() => {
     window.location.hash = page;
   }, [page]);
 
+  useEffect(() => {
+    if (page === "profiles") {
+      setShouldSyncProfileFormToActive(true);
+    }
+  }, [page]);
+
   useEffect(() => {
     const onHashChange = () => {
       const next = window.location.hash.replace("#", "") as PageKey;
@@ -492,19 +648,53 @@ export default function App() {
   }, []);
 
   useEffect(() => {
-    Promise.all([loadStatus(true), loadRequests(), loadProfiles(), loadLogs(false)]).catch((error) => {
+    loadStatus(true).catch((error) => {
       setRuleMessage({ text: error?.message || String(error), tone: "error" });
     });
   }, []);
 
   useEffect(() => {
-      const timer = window.setInterval(() => {
-        refreshLiveData().catch((error) => {
-          setRuleMessage({ text: error?.message || String(error), tone: "error" });
-        });
-      }, 2500);
+    const timer = window.setInterval(() => {
+      loadStatus(false).catch((error) => {
+        setRuleMessage({ text: error?.message || String(error), tone: "error" });
+      });
+    }, 10000);
     return () => window.clearInterval(timer);
-  });
+  }, [restoreRequested, switchingTo]);
+
+  useEffect(() => {
+    if (page !== "requests" && page !== "logs") {
+      return;
+    }
+    const timer = window.setInterval(() => {
+      loadPageData(page, { incrementalLogs: true }).catch((error) => {
+        setRuleMessage({ text: error?.message || String(error), tone: "error" });
+      });
+    }, 2500);
+    return () => window.clearInterval(timer);
+  }, [page, requestQuery, requestFilter, latestLogSeq, requestLimit]);
+
+  useEffect(() => {
+    if (page === "overview" || page === "rules" || page === "requests") {
+      return;
+    }
+    loadPageData(page, { incrementalLogs: false }).catch((error) => {
+      setRuleMessage({ text: error?.message || String(error), tone: "error" });
+    });
+  }, [page]);
+
+  useEffect(() => {
+    if (page !== "profiles" || !shouldSyncProfileFormToActive) {
+      return;
+    }
+    const activeProfile = profiles.find((profile) => profile.active);
+    if (!activeProfile) {
+      return;
+    }
+    setProfileForm(profileFormFromProfile(activeProfile));
+    setProfileMessage({ text: "", tone: "" });
+    setShouldSyncProfileFormToActive(false);
+  }, [page, profiles, shouldSyncProfileFormToActive]);
 
   async function saveRules(event: FormEvent) {
     event.preventDefault();
@@ -517,6 +707,12 @@ export default function App() {
           reasoning_equals: splitList(ruleForm.reasoning_equals)
             .map((value) => Number.parseInt(value, 10))
             .filter((value) => Number.isInteger(value)),
+          retryable_status_codes: splitList(ruleForm.retryable_status_codes)
+            .map((value) => Number.parseInt(value, 10))
+            .filter((value) => Number.isInteger(value)),
+          retryable_error_messages: splitLines(ruleForm.retryable_error_messages),
+          upstream_fetch_retry_attempts: Number.parseInt(ruleForm.upstream_fetch_retry_attempts, 10),
+          upstream_fetch_retry_backoff_ms: Number.parseInt(ruleForm.upstream_fetch_retry_backoff_ms, 10),
           endpoints: splitLines(ruleForm.endpoints),
           non_stream_status_code: Number.parseInt(ruleForm.non_stream_status_code, 10),
           log_match: ruleForm.log_match,
@@ -524,7 +720,6 @@ export default function App() {
       });
       setStatus(payload);
       setRuleForm(ruleFormFromStatus(payload));
-      await Promise.all([loadRequests(), loadLogs(false)]);
       setRuleMessage({ text: "配置已保存，并已对当前 gateway 立即生效。", tone: "success" });
     } catch (error) {
       setRuleMessage({ text: error instanceof Error ? error.message : String(error), tone: "error" });
@@ -532,13 +727,20 @@ export default function App() {
   }
 
   useEffect(() => {
+    setRequestLimit(REQUEST_PAGE_SIZE);
+  }, [requestQuery, requestFilter]);
+
+  useEffect(() => {
+    if (page !== "requests") {
+      return;
+    }
     const timer = window.setTimeout(() => {
-      loadRequests().catch((error) => {
+      loadRequests(requestLimit).catch((error) => {
         setRuleMessage({ text: error?.message || String(error), tone: "error" });
       });
     }, 180);
     return () => window.clearTimeout(timer);
-  }, [requestQuery, requestFilter]);
+  }, [page, requestQuery, requestFilter, requestLimit]);
 
   async function saveProfile(event: FormEvent) {
     event.preventDefault();
@@ -563,6 +765,10 @@ export default function App() {
           request_history_limit: Number.parseInt(profileForm.request_history_limit, 10),
           model_remap: profileForm.model_remap,
           reasoning_equals: splitList(profileForm.reasoning_equals),
+          retryable_status_codes: splitList(profileForm.retryable_status_codes),
+          retryable_error_messages: splitLines(profileForm.retryable_error_messages),
+          upstream_fetch_retry_attempts: Number.parseInt(profileForm.upstream_fetch_retry_attempts, 10),
+          upstream_fetch_retry_backoff_ms: Number.parseInt(profileForm.upstream_fetch_retry_backoff_ms, 10),
           endpoints: splitLines(profileForm.endpoints),
         }),
       });
@@ -579,26 +785,8 @@ export default function App() {
   }
 
   function editProfile(profile: Profile) {
-    const form = profile.form || {};
-    setProfileForm({
-      ...defaultProfileForm,
-      name: profile.name,
-      listen_host: form.listen_host || "",
-      listen_port: form.listen_port || "",
-      upstream_base_url: form.upstream_base_url || "",
-      auth_mode: (form.auth_mode as ProfileFormState["auth_mode"]) || "passthrough",
-      auth_env: form.auth_env || "",
-      auth_file: form.auth_file || "",
-      manual_secret: "",
-      manual_secret_file: form.manual_secret_file || "",
-      manual_secret_configured: Boolean(form.manual_secret_configured),
-      auth_json_path: form.auth_json_path || "",
-      auth_json_key: form.auth_json_key || "OPENAI_API_KEY",
-      request_history_limit: form.request_history_limit || defaultProfileForm.request_history_limit,
-      model_remap: form.model_remap || "",
-      reasoning_equals: form.reasoning_equals || "",
-      endpoints: Array.isArray(form.endpoints) ? form.endpoints.join("\n") : "",
-    });
+    setShouldSyncProfileFormToActive(false);
+    setProfileForm(profileFormFromProfile(profile));
     setProfileMessage({ text: "编辑后保存 profile env；如果保存的是当前运行 profile，后端会直接后台热应用。", tone: "" });
   }
 
@@ -686,7 +874,8 @@ export default function App() {
         const payload = await fetchJson<StatusPayload>(api.status);
         if (payload.config?.profile_name === profileName) {
           setSwitchingTo("");
-          Promise.all([loadStatus(true), loadRequests(), loadProfiles(), loadLogs(false)]).catch(() => {});
+          setShouldSyncProfileFormToActive(true);
+          Promise.all([loadStatus(true), loadProfiles()]).catch(() => {});
           return;
         }
       } catch {
@@ -776,7 +965,7 @@ export default function App() {
                   </div>
                 </Card>
 
-                <Card title="Token Totals" hint="按本次 gateway 启动以来累计；后续可以接持久化统计。">
+                <Card title="Token Totals" hint="按持久请求记录累计；重启后会从 requests history 回灌。">
                   <div className="token-strip">
                     <TokenCard label="Input" value={numberFormat(effectiveInputTotal ?? 0)} />
                     <TokenCard label="Output" value={numberFormat(tokens.output_tokens || 0)} />
@@ -807,7 +996,8 @@ export default function App() {
                       rows={[
                         ["516 命中", numberFormat(metrics.reasoning_516_count || 0)],
                         ["516 占比", percent(metrics.reasoning_516_ratio || 0)],
-                        ["启动时间", timestamp(metrics.started_at)],
+                        ["累计起点", timestamp(metrics.persistent_since || metrics.started_at)],
+                        ["本次启动", timestamp(metrics.started_at)],
                         ["Config", status?.paths?.config_path || "-"],
                         ["备份", status?.state?.latest_backup_path || "-"],
                       ]}
@@ -816,11 +1006,18 @@ export default function App() {
                       {reasoningChips.length === 0 ? (
                         <span className="chip">还没有 reasoning 观测</span>
                       ) : (
-                        reasoningChips.map(([reasoning, count]) => (
-                          <span className="chip" key={reasoning}>
-                            reasoning {reasoning}: {count}
-                          </span>
-                        ))
+                        <>
+                          {reasoningChips.map(([reasoning, count]) => (
+                            <span className="chip" key={reasoning}>
+                              reasoning {reasoning}: {count}
+                            </span>
+                          ))}
+                          {(metrics.observed_reasoning_counts_omitted || 0) > 0 ? (
+                            <span className="chip">
+                              其余 {metrics.observed_reasoning_counts_omitted} 项未展开
+                            </span>
+                          ) : null}
+                        </>
                       )}
                     </div>
                   </Card>
@@ -851,6 +1048,14 @@ export default function App() {
                     <button className="secondary" type="button" onClick={() => loadRequests()}>
                       刷新请求
                     </button>
+                    <button
+                      className="ghost"
+                      type="button"
+                      disabled={requests.length >= requestsTotal && requestsTotal > 0}
+                      onClick={() => setRequestLimit((current) => current + REQUEST_PAGE_SIZE)}
+                    >
+                      更多
+                    </button>
                   </div>
                 }
               >
@@ -884,10 +1089,20 @@ export default function App() {
                                   <span className="meta-key">总</span>
                                   {durationSeconds(entry.duration_ms)}
                                 </span>
+                                <span className="meta-pill meta-progress">
+                                  <span className="meta-key">更</span>
+                                  {timestamp(entry.last_activity_at || entry.usage_last_updated_at)}
+                                </span>
                                 <span className="meta-pill meta-body">
                                   <span className="meta-key">体</span>
                                   {bytesFormat(entry.request_body_bytes)}
                                 </span>
+                                <span className="meta-pill meta-stream">
+                                  <span className="meta-key">流</span>
+                                  {entry.response_stream
+                                    ? `${numberFormat(entry.stream_chunk_count)} chunk / ${bytesFormat(entry.response_bytes_received)} / ${secondsSince(entry.started_at, entry.last_activity_at || entry.usage_last_updated_at || entry.finished_at)}`
+                                    : "-"}
+                                </span>
                                 <span className="meta-pill meta-received">
                                   <span className="meta-key">收</span>
                                   {timestamp(entry.finished_at)}
@@ -896,6 +1111,8 @@ export default function App() {
                               <div className="request-subtitle">
                                 <code>{`${entry.method || "-"} ${entry.path || "-"}`}</code>
                                 <span className="hint">{entry.response_stream ? "stream" : "non-stream"}</span>
+                                <span className="hint">id {requestPrimaryId(entry)}</span>
+                                <span className="hint">thread {entry.thread_id || "-"}</span>
                               </div>
                             </div>
                             <div className="request-badges">
@@ -922,18 +1139,29 @@ export default function App() {
                                 {upstream.auth_mode || "-"} / {upstream.auth_source || "-"}
                                 {upstream.authorization_configured === false ? " / no auth" : ""}
                               </span>
+                              <span className="hint">
+                                重试 {numberFormat(entry.upstream_attempt_count)} 次
+                                {entry.upstream_status_code ? ` / upstream ${entry.upstream_status_code}` : ""}
+                              </span>
                             </div>
 
-                            <div className="request-block">
-                              <label>Token</label>
-                              <div className="request-token-row">
-                                <span className="token-pill token-in">in {numberFormat(effectiveInput)}</span>
-                                <span className="token-pill token-out">out {numberFormat(usage.output_tokens)}</span>
-                                <span className="token-pill token-cached">cached {numberFormat(usage.cached_tokens)}{cachedHitRatio !== null ? ` (${percent(cachedHitRatio)})` : ""}</span>
-                                <span className="token-pill token-reasoning">reasoning {numberFormat(entry.reasoning_tokens ?? usage.reasoning_tokens)}</span>
-                                <span className="token-pill token-total">total {numberFormat(usage.total_tokens)}</span>
+                              <div className="request-block">
+                                <label>Token</label>
+                                <div className="request-token-row">
+                                  <span className="token-pill token-in">in {numberFormat(effectiveInput)}</span>
+                                  <span className="token-pill token-out">out {numberFormat(usage.output_tokens)}</span>
+                                  <span className="token-pill token-cached">cached {numberFormat(usage.cached_tokens)}{cachedHitRatio !== null ? ` (${percent(cachedHitRatio)})` : ""}</span>
+                                  <span className="token-pill token-reasoning">reasoning {numberFormat(entry.reasoning_tokens ?? usage.reasoning_tokens)}</span>
+                                  <span className="token-pill token-total">total {numberFormat(usage.total_tokens)}</span>
+                                </div>
+                              </div>
+
+                              <div className="request-block">
+                                <label>IDs</label>
+                                <code>{`response ${entry.response_id || "-"}`}</code>
+                                <span className="hint">{`request ${entry.request_id || "-"}`}</span>
+                                <span className="hint">{`thread ${entry.thread_id || "-"}`}</span>
                               </div>
-                            </div>
                           </div>
                         </article>
                       );
@@ -955,6 +1183,7 @@ export default function App() {
                       className="secondary"
                       type="button"
                       onClick={() => {
+                        setShouldSyncProfileFormToActive(false);
                         setProfileForm(profileFormFromStatus(status));
                         setProfileMessage({ text: "", tone: "" });
                       }}
@@ -1077,6 +1306,20 @@ export default function App() {
                     <Field label="reasoning_equals">
                       <input value={profileForm.reasoning_equals} placeholder="516,1034,1552" onChange={(event) => setProfileForm({ ...profileForm, reasoning_equals: event.target.value })} />
                     </Field>
+                    <Field label="retryable_status_codes">
+                      <input value={profileForm.retryable_status_codes} placeholder="429,503" onChange={(event) => setProfileForm({ ...profileForm, retryable_status_codes: event.target.value })} />
+                    </Field>
+                    <Field label="retryable_error_messages" hint="每行一条错误文案；上游 JSON 错误包含任一行时会转成本地重试状态码。">
+                      <textarea value={profileForm.retryable_error_messages} onChange={(event) => setProfileForm({ ...profileForm, retryable_error_messages: event.target.value })} />
+                    </Field>
+                    <div className="field-row">
+                      <Field label="upstream_fetch_retry_attempts">
+                        <input type="number" min={1} value={profileForm.upstream_fetch_retry_attempts} onChange={(event) => setProfileForm({ ...profileForm, upstream_fetch_retry_attempts: event.target.value })} />
+                      </Field>
+                      <Field label="upstream_fetch_retry_backoff_ms">
+                        <input type="number" min={0} step={50} value={profileForm.upstream_fetch_retry_backoff_ms} onChange={(event) => setProfileForm({ ...profileForm, upstream_fetch_retry_backoff_ms: event.target.value })} />
+                      </Field>
+                    </div>
                     <Field label="endpoints">
                       <textarea value={profileForm.endpoints} onChange={(event) => setProfileForm({ ...profileForm, endpoints: event.target.value })} />
                     </Field>
@@ -1136,6 +1379,20 @@ export default function App() {
                   <Field label="reasoning_equals">
                     <input value={ruleForm.reasoning_equals} placeholder="例如：516, 1034, 1552" onChange={(event) => setRuleForm({ ...ruleForm, reasoning_equals: event.target.value })} />
                   </Field>
+                  <Field label="retryable_status_codes">
+                    <input value={ruleForm.retryable_status_codes} placeholder="例如：429, 503" onChange={(event) => setRuleForm({ ...ruleForm, retryable_status_codes: event.target.value })} />
+                  </Field>
+                  <Field label="retryable_error_messages" hint="每行一条错误文案；命中后会转成本地 non_stream_status_code。">
+                    <textarea value={ruleForm.retryable_error_messages} onChange={(event) => setRuleForm({ ...ruleForm, retryable_error_messages: event.target.value })} />
+                  </Field>
+                  <div className="field-row">
+                    <Field label="upstream_fetch_retry_attempts">
+                      <input type="number" min={1} value={ruleForm.upstream_fetch_retry_attempts} onChange={(event) => setRuleForm({ ...ruleForm, upstream_fetch_retry_attempts: event.target.value })} />
+                    </Field>
+                    <Field label="upstream_fetch_retry_backoff_ms">
+                      <input type="number" min={0} step={50} value={ruleForm.upstream_fetch_retry_backoff_ms} onChange={(event) => setRuleForm({ ...ruleForm, upstream_fetch_retry_backoff_ms: event.target.value })} />
+                    </Field>
+                  </div>
                   <Field label="endpoints">
                     <textarea value={ruleForm.endpoints} onChange={(event) => setRuleForm({ ...ruleForm, endpoints: event.target.value })} />
                   </Field>

ui-src/src/styles.css

@@ -602,11 +602,21 @@ a {
   background: #f6e3bd;
 }
 
+.meta-progress {
+  color: #4a4a86;
+  background: #e4e1fb;
+}
+
 .meta-body {
   color: #773f33;
   background: #f8ddd5;
 }
 
+.meta-stream {
+  color: #5d4f1f;
+  background: #f2ecbf;
+}
+
 .meta-received {
   color: #425268;
   background: #e5eaef;