feat: add proxy first-byte timeout and log badges (#383)

* add proxy first-byte timeout and log badges * fix first-byte timeout review feedback
2026-04-02 10:00:11 +08:00
parent bcd758edac
commit 180d17adfc
43 changed files with 4049 additions and 91 deletions
@@ -387,6 +387,7 @@ export type RuntimeSettingsPayload = {
  notifyCooldownSec?: number;
  adminIpAllowlist?: string[] | string;
  routingFallbackUnitCost?: number;
+  proxyFirstByteTimeoutSec?: number;
  tokenRouterFailureCooldownMaxSec?: number;
  routingWeights?: RuntimeRoutingWeightsPayload;
  proxyErrorKeywords?: string[] | string;
@@ -437,6 +438,8 @@ export type ProxyLogListItem = {
  modelActual: string;
  status: string;
  latencyMs: number;
+  isStream?: boolean | null;
+  firstByteLatencyMs?: number | null;
  totalTokens: number | null;
  retryCount: number;
  accountId?: number | null;
@@ -58,6 +58,8 @@ function buildListResponse(overrides?: Partial<{
        modelActual: 'gpt-4o',
        status: 'success',
        latencyMs: 120,
+        firstByteLatencyMs: 35,
+        isStream: true,
        promptTokens: 10,
        completionTokens: 5,
        totalTokens: 15,
@@ -146,6 +148,8 @@ describe('ProxyLogs server-driven page', () => {
      modelActual: 'gpt-4o',
      status: 'success',
      latencyMs: 120,
+      firstByteLatencyMs: 35,
+      isStream: true,
      promptTokens: 10,
      completionTokens: 5,
      totalTokens: 15,
@@ -264,6 +268,8 @@ describe('ProxyLogs server-driven page', () => {
      expect(text).toContain('Codex');
      expect(text).toContain('推测');
      expect(text).toContain('下游 Key: 移动端灰度');
+      expect(text).toContain('流式');
+      expect(text).toContain('首字');
    } finally {
      await act(async () => {
        root?.unmount();
@@ -801,6 +807,8 @@ describe('ProxyLogs server-driven page', () => {
          modelActual: 'gpt-4o',
          status: 'success',
          latencyMs: 120,
+          firstByteLatencyMs: 22,
+          isStream: false,
          promptTokens: 10,
          completionTokens: 5,
          totalTokens: 15,
@@ -265,6 +265,28 @@ function latencyBgColor(ms: number) {
  return 'color-mix(in srgb, var(--color-success) 12%, transparent)';
 }

+function firstByteColor(ms: number) {
+  if (ms >= 3000) return 'var(--color-danger)';
+  if (ms >= 1000) return 'var(--color-warning)';
+  return 'var(--color-primary)';
+}
+
+function firstByteBgColor(ms: number) {
+  if (ms >= 3000) return 'color-mix(in srgb, var(--color-danger) 12%, transparent)';
+  if (ms >= 1000) return 'color-mix(in srgb, var(--color-warning) 12%, transparent)';
+  return 'color-mix(in srgb, var(--color-primary) 12%, transparent)';
+}
+
+function formatStreamModeLabel(isStream: boolean | null | undefined) {
+  if (isStream == null) return null;
+  return isStream ? '流式' : '非流';
+}
+
+function formatFirstByteLabel(ms: number | null | undefined) {
+  if (!Number.isFinite(ms) || typeof ms !== 'number' || ms < 0) return null;
+  return `首字 ${formatLatency(ms)}`;
+}
+
 function formatCompactNumber(value: number, digits = 6) {
  if (!Number.isFinite(value)) return '0';
  const formatted = value.toFixed(digits).replace(/\.?0+$/, '');
@@ -1877,6 +1899,8 @@ export default function ProxyLogs() {
              const downstreamKeySummary = renderDownstreamKeySummary(detailLog);
              const isExpanded = expanded === log.id;
              const clientDisplay = resolveProxyLogClientDisplay(detailLog);
+              const streamModeLabel = formatStreamModeLabel(detailLog.isStream);
+              const firstByteLabel = formatFirstByteLabel(detailLog.firstByteLatencyMs);

              return (
                <MobileCard
@@ -1911,6 +1935,24 @@ export default function ProxyLogs() {
                        {clientDisplay.secondary}
                      </span>
                    ) : null}
+                    {streamModeLabel ? (
+                      <span className="badge badge-muted" style={{ fontSize: 10 }}>
+                        {streamModeLabel}
+                      </span>
+                    ) : null}
+                    {firstByteLabel ? (
+                      <span
+                        className="badge"
+                        style={{
+                          fontSize: 10,
+                          color: firstByteColor(detailLog.firstByteLatencyMs ?? 0),
+                          background: firstByteBgColor(detailLog.firstByteLatencyMs ?? 0),
+                          borderColor: 'transparent',
+                        }}
+                      >
+                        {firstByteLabel}
+                      </span>
+                    ) : null}
                  </div>
                  <div className="mobile-summary-grid">
                    <div className="mobile-summary-metric">
@@ -1934,6 +1976,8 @@ export default function ProxyLogs() {
                    <div className="mobile-card-extra">
                      <MobileField label="时间" value={formatDateTimeLocal(log.createdAt)} />
                      <MobileField label="站点" value={<SiteBadgeLink siteId={siteIdByName.get(String(log.siteName || '').trim())} siteName={log.siteName} badgeStyle={{ fontSize: 11 }} />} />
+                      {streamModeLabel ? <MobileField label="模式" value={streamModeLabel} /> : null}
+                      {firstByteLabel ? <MobileField label="首字" value={firstByteLabel.replace(/^首字\s*/, '')} /> : null}
                      <MobileField label="重试" value={log.retryCount > 0 ? log.retryCount : 0} />
                      <MobileField label="用量来源" value={formatProxyLogUsageSource(detailLog.usageSource ?? pathMeta.usageSource) || '--'} />
                      {detailState?.loading && <div style={{ color: 'var(--color-text-muted)' }}>加载详情中...</div>}
@@ -1983,6 +2027,8 @@ export default function ProxyLogs() {
                const billingDetailSummary = detail ? formatBillingDetailSummary(detailLog) : null;
                const billingProcessLines = detail ? buildBillingProcessLines(detailLog) : [];
                const downstreamKeySummary = renderDownstreamKeySummary(detailLog);
+                const streamModeLabel = formatStreamModeLabel(detailLog.isStream);
+                const firstByteLabel = formatFirstByteLabel(detailLog.firstByteLatencyMs);

                return (
                  <React.Fragment key={log.id}>
@@ -2015,6 +2061,28 @@ export default function ProxyLogs() {
                              {downstreamKeySummary}
                            </div>
                          ) : null}
+                          {(streamModeLabel || firstByteLabel) ? (
+                            <div style={{ display: 'flex', gap: 6, flexWrap: 'wrap' }}>
+                              {streamModeLabel ? (
+                                <span className="badge badge-muted" style={{ fontSize: 10 }}>
+                                  {streamModeLabel}
+                                </span>
+                              ) : null}
+                              {firstByteLabel ? (
+                                <span
+                                  className="badge"
+                                  style={{
+                                    fontSize: 10,
+                                    color: firstByteColor(detailLog.firstByteLatencyMs ?? 0),
+                                    background: firstByteBgColor(detailLog.firstByteLatencyMs ?? 0),
+                                    borderColor: 'transparent',
+                                  }}
+                                >
+                                  {firstByteLabel}
+                                </span>
+                              ) : null}
+                            </div>
+                          ) : null}
                        </div>
                      </td>
                      <td style={{ fontSize: 12, color: 'var(--color-text-secondary)' }}>
@@ -2084,6 +2152,12 @@ export default function ProxyLogs() {
                                        <>{' -> '}实际模型: <strong style={{ color: 'var(--color-text-primary)' }}>{detailLog.modelActual}</strong></>
                                      )}
                                      ，状态: <strong style={{ color: detailLog.status === 'success' ? 'var(--color-success)' : 'var(--color-danger)' }}>{detailLog.status === 'success' ? '成功' : '失败'}</strong>
+                                      {streamModeLabel && (
+                                        <>，模式: <strong style={{ color: 'var(--color-text-primary)' }}>{streamModeLabel}</strong></>
+                                      )}
+                                      {firstByteLabel && (
+                                        <>，首字: <strong style={{ color: firstByteColor(detailLog.firstByteLatencyMs ?? 0) }}>{formatLatency(detailLog.firstByteLatencyMs ?? 0)}</strong></>
+                                      )}
                                      ，用时: <strong style={{ color: latencyColor(detailLog.latencyMs) }}>{formatLatency(detailLog.latencyMs)}</strong>
                                      {detail && (
                                        <>
@@ -66,6 +66,7 @@ type RuntimeSettings = {
  proxySessionChannelConcurrencyLimit: number;
  proxySessionChannelQueueWaitMs: number;
  routingFallbackUnitCost: number;
+  proxyFirstByteTimeoutSec: number;
  routeFailureCooldownMaxValue: number;
  routeFailureCooldownMaxUnit: RouteCooldownUnit;
  routingWeights: RoutingWeights;
@@ -244,6 +245,7 @@ export default function Settings() {
    proxySessionChannelConcurrencyLimit: 2,
    proxySessionChannelQueueWaitMs: 1500,
    routingFallbackUnitCost: 1,
+    proxyFirstByteTimeoutSec: 0,
    routeFailureCooldownMaxValue: 30,
    routeFailureCooldownMaxUnit: 'day',
    routingWeights: defaultWeights,
@@ -496,6 +498,9 @@ export default function Settings() {
        routingFallbackUnitCost: Number(runtimeInfo.routingFallbackUnitCost) > 0
          ? Number(runtimeInfo.routingFallbackUnitCost)
          : 1,
+        proxyFirstByteTimeoutSec: Number(runtimeInfo.proxyFirstByteTimeoutSec) >= 0
+          ? Math.trunc(Number(runtimeInfo.proxyFirstByteTimeoutSec))
+          : 0,
        routeFailureCooldownMaxValue: routeCooldownInput.value,
        routeFailureCooldownMaxUnit: routeCooldownInput.unit,
        routingWeights: {
@@ -942,6 +947,9 @@ export default function Settings() {
      await api.updateRuntimeSettings({
        routingWeights: runtime.routingWeights,
        routingFallbackUnitCost: runtime.routingFallbackUnitCost,
+        proxyFirstByteTimeoutSec: Number.isFinite(runtime.proxyFirstByteTimeoutSec)
+          ? Math.max(0, Math.trunc(runtime.proxyFirstByteTimeoutSec))
+          : 0,
        tokenRouterFailureCooldownMaxSec: toRouteCooldownSeconds(
          runtime.routeFailureCooldownMaxValue,
          runtime.routeFailureCooldownMaxUnit,
@@ -1739,6 +1747,32 @@ export default function Settings() {
            </span>
          </label>

+          <div style={{ marginBottom: 12 }}>
+            <div style={{ fontSize: 12, color: 'var(--color-text-muted)', marginBottom: 6 }}>
+              首字超时（无首包 / 首 token）
+            </div>
+            <input
+              type="number"
+              min={0}
+              step={1}
+              aria-label="首字超时秒数"
+              value={runtime.proxyFirstByteTimeoutSec}
+              onChange={(e) => {
+                const nextValue = Number(e.target.value);
+                setRuntime((prev) => ({
+                  ...prev,
+                  proxyFirstByteTimeoutSec: Number.isFinite(nextValue) && nextValue >= 0
+                    ? Math.trunc(nextValue)
+                    : prev.proxyFirstByteTimeoutSec,
+                }));
+              }}
+              style={inputStyle}
+            />
+            <div style={{ fontSize: 12, color: 'var(--color-text-muted)', lineHeight: 1.7, marginTop: 6 }}>
+              `0` 表示关闭。只有在指定时间内完全没有任何首包 / 首 token 返回时才切换，已经开始输出的请求不会被这项超时打断。
+            </div>
+          </div>
+
          <div className={`anim-collapse ${showAdvancedRouting ? 'is-open' : ''}`.trim()}>
            <div className="anim-collapse-inner" style={{ paddingTop: 2 }}>
              <div style={{ display: 'grid', gridTemplateColumns: isMobile ? '1fr' : '1fr 1fr', gap: 12 }}>
@@ -57,6 +57,7 @@ describe('Settings route cooldown cap', () => {
      logCleanupProgramLogsEnabled: true,
      logCleanupRetentionDays: 14,
      routingFallbackUnitCost: 1,
+      proxyFirstByteTimeoutSec: 0,
      routingWeights: {},
      tokenRouterFailureCooldownMaxSec: 30 * 24 * 60 * 60,
      adminIpAllowlist: [],
@@ -127,6 +128,7 @@ describe('Settings route cooldown cap', () => {
          usageWeight: 0.3,
        },
        routingFallbackUnitCost: 1,
+        proxyFirstByteTimeoutSec: 0,
        tokenRouterFailureCooldownMaxSec: 10,
        disableCrossProtocolFallback: false,
      });
@@ -146,6 +148,7 @@ describe('Settings route cooldown cap', () => {
      logCleanupProgramLogsEnabled: true,
      logCleanupRetentionDays: 14,
      routingFallbackUnitCost: 1,
+      proxyFirstByteTimeoutSec: 0,
      routingWeights: {},
      tokenRouterFailureCooldownMaxSec: 10,
      adminIpAllowlist: [],
@@ -181,4 +184,57 @@ describe('Settings route cooldown cap', () => {
      root?.unmount();
    }
  });
+
+  it('saves the first-byte timeout seconds alongside other routing runtime settings', async () => {
+    let root!: ReactTestRenderer;
+    try {
+      await act(async () => {
+        root = create(
+          <MemoryRouter>
+            <ToastProvider>
+              <Settings />
+            </ToastProvider>
+          </MemoryRouter>,
+        );
+      });
+      await flushMicrotasks();
+
+      const firstByteInput = root.root.find((node) => (
+        node.type === 'input'
+        && node.props.type === 'number'
+        && node.props['aria-label'] === '首字超时秒数'
+      ));
+
+      await act(async () => {
+        firstByteInput.props.onChange({ target: { value: '7' } });
+      });
+
+      const saveButton = root.root.find((node) => (
+        node.type === 'button'
+        && typeof node.props.onClick === 'function'
+        && collectText(node).trim() === '保存路由策略'
+      ));
+
+      await act(async () => {
+        saveButton.props.onClick();
+      });
+      await flushMicrotasks();
+
+      expect(apiMock.updateRuntimeSettings).toHaveBeenCalledWith({
+        routingWeights: {
+          baseWeightFactor: 0.5,
+          valueScoreFactor: 0.5,
+          costWeight: 0.4,
+          balanceWeight: 0.3,
+          usageWeight: 0.3,
+        },
+        routingFallbackUnitCost: 1,
+        proxyFirstByteTimeoutSec: 7,
+        tokenRouterFailureCooldownMaxSec: 30 * 24 * 60 * 60,
+        disableCrossProtocolFallback: false,
+      });
+    } finally {
+      root?.unmount();
+    }
+  });
 });