Merge remote-tracking branch 'origin/main' into codex/fix-antigravity-quota

2026-10-06 01:17:46 +08:00 · 2026-09-03 10:46:49 +08:00
parent 45c840b8d3 cae9aa4134
commit f822df6cce
57 changed files with 1850 additions and 696 deletions
@@ -155,17 +155,6 @@
              </SelectContent>
            </Select>
          </div>
-          <div class="space-y-1.5">
-            <Label>{{ legacyT('最大重试次数') }}</Label>
-            <Input
-              :model-value="form.max_retries ?? ''"
-              type="number"
-              min="0"
-              max="999"
-              :placeholder="legacyT('默认 2')"
-              @update:model-value="(v) => form.max_retries = parseNumberInput(v)"
-            />
-          </div>
        </div>

        <!-- 超时配置 -->
@@ -12,12 +12,15 @@ import {
  getModelScheduling,
  modelSchedulingRuleId,
  normalizeRoutingGroupConfig,
+  normalizeStickyKeyAttempts,
  parseAllowedModelsInput,
  removePerModelRoutingConfig,
+  resolveModelKeyPriorityOverride,
  routingModelScopeLabel,
  savePerModelRoutingConfig,
  setDefaultPoolPriorityOverrides,
  setDefaultProviderPriorityOverrides,
+  setModelKeyPriorityOverridesForFormat,
  setRoutingSortingScope,
  updateAllowedModelsFromInput,
  upsertModelSchedulingRule,
@@ -75,6 +78,65 @@ describe('routingPolicy', () => {
    expect(policy.key_priority_overrides).toEqual({})
  })

+  it('defaults sticky key attempts to 2 and normalizes invalid values', () => {
+    expect(createEmptyRoutingGroupConfig().default_policy.sticky_key_attempts).toBe(2)
+    expect(normalizeRoutingGroupConfig({}).default_policy.sticky_key_attempts).toBe(2)
+    expect(normalizeRoutingGroupConfig({
+      default_policy: { priority_mode: 'provider', scheduling_mode: 'cache_affinity', keep_priority_on_conversion: false, sticky_key_attempts: 3 },
+    }).default_policy.sticky_key_attempts).toBe(3)
+    expect(normalizeStickyKeyAttempts('5')).toBe(5)
+    expect(normalizeStickyKeyAttempts(-1)).toBe(2)
+    expect(normalizeStickyKeyAttempts('abc')).toBe(2)
+    expect(normalizeStickyKeyAttempts(500)).toBe(99)
+    expect(getModelScheduling(createEmptyRoutingGroupConfig(), 'gpt-5').sticky_key_attempts).toBe(2)
+  })
+
+  it('keeps key priority overrides independent per api format', () => {
+    let config = setModelKeyPriorityOverridesForFormat(
+      createEmptyRoutingGroupConfig(),
+      DEFAULT_ROUTING_POLICY_MODEL,
+      'OpenAI:Chat',
+      { 'key-a': 0, 'key-b': 1 },
+    )
+    config = setModelKeyPriorityOverridesForFormat(
+      config,
+      DEFAULT_ROUTING_POLICY_MODEL,
+      'claude:messages',
+      { 'key-a': 3 },
+    )
+
+    const policy = getDefaultModelPolicy(config)
+    expect(policy.key_priority_overrides).toEqual({})
+    expect(policy.key_priority_overrides_by_format).toEqual({
+      'openai:chat': { 'key-a': 0, 'key-b': 1 },
+      'claude:messages': { 'key-a': 3 },
+    })
+    expect(resolveModelKeyPriorityOverride(config, DEFAULT_ROUTING_POLICY_MODEL, 'openai:chat', 'key-a')).toBe(0)
+    expect(resolveModelKeyPriorityOverride(config, DEFAULT_ROUTING_POLICY_MODEL, 'claude:messages', 'key-a')).toBe(3)
+    expect(resolveModelKeyPriorityOverride(config, DEFAULT_ROUTING_POLICY_MODEL, 'claude:messages', 'key-b')).toBeUndefined()
+
+    const cleared = setModelKeyPriorityOverridesForFormat(config, DEFAULT_ROUTING_POLICY_MODEL, 'claude:messages', {})
+    expect(getDefaultModelPolicy(cleared).key_priority_overrides_by_format).toEqual({
+      'openai:chat': { 'key-a': 0, 'key-b': 1 },
+    })
+  })
+
+  it('falls back to format-agnostic key overrides and normalizes legacy configs', () => {
+    const config = normalizeRoutingGroupConfig({
+      model_policies: [{
+        ...createEmptyModelPolicy('gpt-5'),
+        key_priority_overrides: { 'key-a': 7 },
+        key_priority_overrides_by_format: { ' OpenAI:Chat ': { 'key-a': 1 } },
+      }],
+    })
+
+    expect(config.model_policies[0].key_priority_overrides_by_format).toEqual({
+      'openai:chat': { 'key-a': 1 },
+    })
+    expect(resolveModelKeyPriorityOverride(config, 'gpt-5', 'openai:chat', 'key-a')).toBe(1)
+    expect(resolveModelKeyPriorityOverride(config, 'gpt-5', 'gemini:generate_content', 'key-a')).toBe(7)
+  })
+
  it('stores per-model scheduling as generated routing rules', () => {
    const next = upsertModelSchedulingRule(createEmptyRoutingGroupConfig(), 'gpt-5', {
      priority_mode: 'global_key',
@@ -315,7 +315,8 @@ import {
  getDefaultModelPolicy,
  getModelPolicy,
  normalizeRoutingGroupConfig,
-  setModelKeyPriorityOverrides,
+  normalizeRoutingApiFormatKey,
+  setModelKeyPriorityOverridesForFormat,
  setModelPoolPriorityOverrides,
  setModelProviderPriorityOverrides,
  type RoutingDefaultPolicy,
@@ -452,9 +453,18 @@ const providerRows = computed<ProviderPriorityRow[]>(() => {
    .sort(comparePriorityRows)
 })

+const selectedFormatKey = computed(() => normalizeRoutingApiFormatKey(selectedApiFormat.value))
+const selectedFormatKeyOverrides = computed<Record<string, number>>(() => (
+  targetModelPolicy.value.key_priority_overrides_by_format[selectedFormatKey.value] ?? {}
+))
+
 const keyRows = computed<KeyPriorityRow[]>(() => {
  const format = selectedApiFormat.value
-  const keyOverrides = targetModelPolicy.value.key_priority_overrides
+  // 按格式覆盖优先；旧的不分格式覆盖仅作为兜底展示
+  const keyOverrides: Record<string, number> = {
+    ...targetModelPolicy.value.key_priority_overrides,
+    ...selectedFormatKeyOverrides.value,
+  }
  const poolOverrides = targetModelPolicy.value.pool_priority_overrides
  const normalRows: KeyPriorityRow[] = []
  const poolGroups = new Map<string, GlobalKeySource[]>()
@@ -686,7 +696,7 @@ function setKeyPriority(keyId: string, event: Event): void {
    })
  } else {
    updateKeyOverrides({
-      ...targetModelPolicy.value.key_priority_overrides,
+      ...selectedFormatKeyOverrides.value,
      [row.target_id]: priority,
    })
  }
@@ -697,8 +707,14 @@ function moveKey(keyId: string, direction: -1 | 1): void {
  updateVisibleKeyAndPoolOverrides(rows)
 }

+// Key 覆盖始终写入当前选中的 API 格式，不同格式互不影响
 function updateKeyOverrides(overrides: Record<string, number>): void {
-  updateConfig(setModelKeyPriorityOverrides(config.value, targetModel.value, overrides))
+  updateConfig(setModelKeyPriorityOverridesForFormat(
+    config.value,
+    targetModel.value,
+    selectedApiFormat.value,
+    overrides,
+  ))
 }

 function updatePoolOverrides(overrides: Record<string, number>): void {
@@ -710,7 +726,12 @@ function updateKeyAndPoolOverrides(
  poolOverrides: Record<string, number>,
 ): void {
  const next = setModelPoolPriorityOverrides(
-    setModelKeyPriorityOverrides(config.value, targetModel.value, keyOverrides),
+    setModelKeyPriorityOverridesForFormat(
+      config.value,
+      targetModel.value,
+      selectedApiFormat.value,
+      keyOverrides,
+    ),
    targetModel.value,
    poolOverrides,
  )
@@ -718,7 +739,7 @@ function updateKeyAndPoolOverrides(
 }

 function updateVisibleKeyAndPoolOverrides(rows: KeyPriorityRow[]): void {
-  const keyOverrides = { ...targetModelPolicy.value.key_priority_overrides }
+  const keyOverrides = { ...selectedFormatKeyOverrides.value }
  const poolOverrides = { ...targetModelPolicy.value.pool_priority_overrides }

  for (const row of keyRows.value) {
@@ -3,10 +3,15 @@ export type RoutingSchedulingMode = 'fixed_order' | 'cache_affinity' | 'load_bal
 export type RoutingRulePhase = 'client_request' | 'provider_request'
 export type RoutingSortingScope = 'unified' | 'per_model'

+/** 首个候选（粘性 Key）的总尝试次数默认值：失败后同 Key 重试 1 次 */
+export const DEFAULT_STICKY_KEY_ATTEMPTS = 2
+
 export interface RoutingDefaultPolicy {
  priority_mode: RoutingPriorityMode
  scheduling_mode: RoutingSchedulingMode
  keep_priority_on_conversion: boolean
+  /** 首个候选的总尝试次数；后续候选始终只尝试 1 次。0 或 1 表示不重试 */
+  sticky_key_attempts: number
 }

 export interface RoutingPoolSchedulingPreset {
@@ -25,6 +30,8 @@ export interface RoutingModelPolicy {
  allowed_keys: string[]
  provider_priority_overrides: Record<string, number>
  key_priority_overrides: Record<string, number>
+  /** api_format -> key_id -> priority；同一 Key 在不同 API 格式下可独立排序 */
+  key_priority_overrides_by_format: Record<string, Record<string, number>>
  pool_priority_overrides: Record<string, number>
  pool_policy_overrides: Record<string, RoutingPoolPolicyOverride>
 }
@@ -49,6 +56,7 @@ export interface RoutingSetSchedulingAction {
  type: 'set_scheduling'
  priority_mode: RoutingPriorityMode
  scheduling_mode: RoutingSchedulingMode
+  sticky_key_attempts?: number
 }

 export interface RoutingGroupConfig {
@@ -68,12 +76,19 @@ export function createEmptyRoutingGroupConfig(): RoutingGroupConfig {
      priority_mode: 'provider',
      scheduling_mode: 'cache_affinity',
      keep_priority_on_conversion: false,
+      sticky_key_attempts: DEFAULT_STICKY_KEY_ATTEMPTS,
    },
    model_policies: [],
    rules: [],
  }
 }

+export function normalizeStickyKeyAttempts(value: unknown): number {
+  const parsed = Math.trunc(Number(value))
+  if (!Number.isFinite(parsed) || parsed < 0) return DEFAULT_STICKY_KEY_ATTEMPTS
+  return Math.min(parsed, 99)
+}
+
 export function createEmptyModelPolicy(model = ''): RoutingModelPolicy {
  return {
    model,
@@ -81,6 +96,7 @@ export function createEmptyModelPolicy(model = ''): RoutingModelPolicy {
    allowed_keys: [],
    provider_priority_overrides: {},
    key_priority_overrides: {},
+    key_priority_overrides_by_format: {},
    pool_priority_overrides: {},
    pool_policy_overrides: {},
  }
@@ -94,6 +110,9 @@ export function normalizeRoutingGroupConfig(value: Partial<RoutingGroupConfig> |
    default_policy: {
      ...base.default_policy,
      ...(value?.default_policy ?? {}),
+      sticky_key_attempts: normalizeStickyKeyAttempts(
+        value?.default_policy?.sticky_key_attempts ?? DEFAULT_STICKY_KEY_ATTEMPTS,
+      ),
    },
    model_policies: Array.isArray(value?.model_policies)
      ? value.model_policies.map(policy => ({
@@ -103,6 +122,9 @@ export function normalizeRoutingGroupConfig(value: Partial<RoutingGroupConfig> |
          allowed_keys: Array.isArray(policy.allowed_keys) ? [...policy.allowed_keys] : [],
          provider_priority_overrides: { ...(policy.provider_priority_overrides ?? {}) },
          key_priority_overrides: { ...(policy.key_priority_overrides ?? {}) },
+          key_priority_overrides_by_format: normalizeKeyPriorityOverridesByFormat(
+            policy.key_priority_overrides_by_format,
+          ),
          pool_priority_overrides: { ...(policy.pool_priority_overrides ?? {}) },
          pool_policy_overrides: { ...(policy.pool_policy_overrides ?? {}) },
        }))
@@ -296,6 +318,64 @@ export function setModelKeyPriorityOverrides(
  })
 }

+export function normalizeRoutingApiFormatKey(apiFormat: string): string {
+  return apiFormat.trim().toLowerCase()
+}
+
+export function getModelKeyPriorityOverridesForFormat(
+  config: RoutingGroupConfig,
+  model: string,
+  apiFormat: string,
+): Record<string, number> {
+  const policy = getModelPolicy(config, model)
+  const format = normalizeRoutingApiFormatKey(apiFormat)
+  return { ...(policy.key_priority_overrides_by_format[format] ?? {}) }
+}
+
+/**
+ * 某个 Key 在指定 API 格式下的生效覆盖值：按格式覆盖优先，其次是不分格式的 Key 覆盖。
+ */
+export function resolveModelKeyPriorityOverride(
+  config: RoutingGroupConfig,
+  model: string,
+  apiFormat: string,
+  keyId: string,
+): number | undefined {
+  const policy = getModelPolicy(config, model)
+  const format = normalizeRoutingApiFormatKey(apiFormat)
+  return policy.key_priority_overrides_by_format[format]?.[keyId]
+    ?? policy.key_priority_overrides[keyId]
+}
+
+export function setModelKeyPriorityOverridesForFormat(
+  config: RoutingGroupConfig,
+  model: string,
+  apiFormat: string,
+  overrides: Record<string, number>,
+): RoutingGroupConfig {
+  const normalizedModel = model.trim() || DEFAULT_ROUTING_POLICY_MODEL
+  const format = normalizeRoutingApiFormatKey(apiFormat)
+  if (!format) return normalizeRoutingGroupConfig(config)
+
+  const current = getModelPolicy(config, normalizedModel)
+  const byFormat = { ...current.key_priority_overrides_by_format }
+  const normalized = normalizePriorityOverrides(overrides)
+  if (Object.keys(normalized).length > 0) {
+    byFormat[format] = normalized
+  } else {
+    delete byFormat[format]
+  }
+
+  if (normalizedModel === DEFAULT_ROUTING_POLICY_MODEL) {
+    return upsertDefaultModelPolicy(config, { key_priority_overrides_by_format: byFormat })
+  }
+  return upsertModelPolicy(config, {
+    ...current,
+    model: normalizedModel,
+    key_priority_overrides_by_format: byFormat,
+  })
+}
+
 export function setModelPoolPriorityOverrides(
  config: RoutingGroupConfig,
  model: string,
@@ -347,6 +427,7 @@ export function getModelScheduling(
    priority_mode: action?.priority_mode ?? normalized.default_policy.priority_mode,
    scheduling_mode: action?.scheduling_mode ?? normalized.default_policy.scheduling_mode,
    keep_priority_on_conversion: normalized.default_policy.keep_priority_on_conversion,
+    sticky_key_attempts: action?.sticky_key_attempts ?? normalized.default_policy.sticky_key_attempts,
  }
 }

@@ -464,6 +545,25 @@ export function normalizePriorityOverrides(overrides: Record<string, number>): R
  return normalized
 }

+function normalizeKeyPriorityOverridesByFormat(
+  value: Record<string, Record<string, number>> | null | undefined,
+): Record<string, Record<string, number>> {
+  const normalized: Record<string, Record<string, number>> = {}
+  if (!value || typeof value !== 'object') return normalized
+  for (const [rawFormat, overrides] of Object.entries(value)) {
+    const format = normalizeRoutingApiFormatKey(rawFormat)
+    if (!format || !overrides || typeof overrides !== 'object') continue
+    const merged = normalizePriorityOverrides({
+      ...(normalized[format] ?? {}),
+      ...overrides,
+    })
+    if (Object.keys(merged).length > 0) {
+      normalized[format] = merged
+    }
+  }
+  return normalized
+}
+
 function isSetSchedulingAction(action: unknown): action is RoutingSetSchedulingAction {
  if (!action || typeof action !== 'object') return false
  const candidate = action as Partial<RoutingSetSchedulingAction>