feat: add AI model health ban exclusion

2026-05-27 11:54:17 +08:00 · 2026-05-27 11:54:17 +08:00 · c7966298dc
commit c7966298dc
parent 15a08e6d65
8 changed files with 88 additions and 14 deletions
--- a/app/(admin)/settings/admin-settings-client.tsx
+++ b/app/(admin)/settings/admin-settings-client.tsx
@ -869,6 +869,7 @@ type AiModel = {
  fallbackModelId1?: string;
  fallbackModelId2?: string;
  excludeFromFailover?: boolean;
+  excludeFromHealthBan?: boolean;
  healthStatus?: "healthy" | "degraded" | "auto_disabled";
  healthConsecutiveFailures?: number;
  healthLastEventType?: string | null;
@ -3608,6 +3609,7 @@ export function AdminSettingsClient() {
    fallbackModelId1: "", // 备用模型 ID 1
    fallbackModelId2: "", // 备用模型 ID 2
    excludeFromFailover: false,
+    excludeFromHealthBan: false,
    capabilities: {} as Record<string, unknown>,
    // 🆕 图片生成模型特性配置
    imageGenConfig: createDefaultImageGenConfig(),
@ -6892,6 +6894,7 @@ export function AdminSettingsClient() {
      fallbackModelId1: "",
      fallbackModelId2: "",
      excludeFromFailover: false,
+      excludeFromHealthBan: false,
      capabilities: {},
      imageGenConfig: createDefaultImageGenConfig(),
    });
@ -14978,19 +14981,44 @@ export function AdminSettingsClient() {
                            多模态模型（Agent 识图与分析使用同一模型）
                          </Label>
                        </div>
-                        <div className="flex items-center gap-2">
-                          <Switch
-                            checked={modelForm.excludeFromFailover || false}
-                            onCheckedChange={(checked) =>
-                              setModelForm({
-                                ...modelForm,
-                                excludeFromFailover: checked,
-                              })
-                            }
-                          />
-                          <Label>
-                            禁止参与故障迁移（不作为其他模型的自动备用模型）
-                          </Label>
+                        <div className="space-y-3 rounded-md border p-3">
+                          <div className="text-sm font-medium">
+                            故障保护策略
+                          </div>
+                          <div className="flex items-center gap-2">
+                            <Switch
+                              checked={modelForm.excludeFromFailover || false}
+                              onCheckedChange={(checked) =>
+                                setModelForm({
+                                  ...modelForm,
+                                  excludeFromFailover: checked,
+                                })
+                              }
+                            />
+                            <Label>
+                              不参与故障迁移候选池
+                            </Label>
+                          </div>
+                          <p className="text-xs text-muted-foreground">
+                            开启后，该模型不会作为其他模型请求失败时的自动备用模型。
+                          </p>
+                          <div className="flex items-center gap-2">
+                            <Switch
+                              checked={modelForm.excludeFromHealthBan || false}
+                              onCheckedChange={(checked) =>
+                                setModelForm({
+                                  ...modelForm,
+                                  excludeFromHealthBan: checked,
+                                })
+                              }
+                            />
+                            <Label>
+                              不参与故障封禁池
+                            </Label>
+                          </div>
+                          <p className="text-xs text-muted-foreground">
+                            开启后，健康检查仍会记录故障日志，但不会因为连续失败自动禁用该模型。
+                          </p>
                        </div>

                        {/* 🆕 图片生成模型特性配置 */}
@ -15911,7 +15939,12 @@ export function AdminSettingsClient() {
                            )}
                            {(model as any).excludeFromFailover && (
                              <span className="text-xs bg-rose-100 text-rose-700 px-2 py-0.5 rounded">
-                                不参与故障迁移
+                                不参与迁移池
+                              </span>
+                            )}
+                            {(model as any).excludeFromHealthBan && (
+                              <span className="text-xs bg-stone-100 text-stone-700 px-2 py-0.5 rounded">
+                                不参与封禁池
                              </span>
                            )}
                            {healthStatus === "auto_disabled" && (
@ -16033,6 +16066,8 @@ export function AdminSettingsClient() {
                                (model as any).fallbackModelId2 || "",
                              excludeFromFailover:
                                (model as any).excludeFromFailover || false,
+                              excludeFromHealthBan:
+                                (model as any).excludeFromHealthBan || false,
                              capabilities:
                                (model as any).capabilities &&
                                typeof (model as any).capabilities === "object" &&
--- a/app/api/admin/ai-models/route.ts
+++ b/app/api/admin/ai-models/route.ts
@ -82,6 +82,7 @@ const createModelSchema = z.object({
  fallbackModelId1: z.string().max(128).optional(),
  fallbackModelId2: z.string().max(128).optional(),
  excludeFromFailover: z.boolean().optional(),
+  excludeFromHealthBan: z.boolean().optional(),
  isAgentModel: z.boolean().optional(),
  agentCodingModelId: z.string().max(128).optional(),
  isMultimodal: z.boolean().optional(),
@ -106,6 +107,7 @@ const updateModelSchema = z.object({
  fallbackModelId1: z.string().max(128).optional(),
  fallbackModelId2: z.string().max(128).optional(),
  excludeFromFailover: z.boolean().optional(),
+  excludeFromHealthBan: z.boolean().optional(),
  isAgentModel: z.boolean().optional(),
  agentCodingModelId: z.string().max(128).optional(),
  isMultimodal: z.boolean().optional(),
--- a/lib/ai/runtime/model-health.ts
+++ b/lib/ai/runtime/model-health.ts
@ -211,6 +211,7 @@ export async function reviveAiModelsFromCooldown(params?: { modelId?: string })
  const conditions = [
    eq(aiModel.isEnabled, false),
    eq(aiModel.excludeFromFailover, false),
+    eq(aiModel.excludeFromHealthBan, false),
  ];

  if (params?.modelId) {
@ -392,6 +393,7 @@ export async function recordAiModelHealthFailure(params: {
      displayName: aiModel.displayName,
      isEnabled: aiModel.isEnabled,
      excludeFromFailover: aiModel.excludeFromFailover,
+      excludeFromHealthBan: aiModel.excludeFromHealthBan,
      healthStatus: aiModel.healthStatus,
      healthConsecutiveFailures: aiModel.healthConsecutiveFailures,
      healthRecoverySuccessStreak: aiModel.healthRecoverySuccessStreak,
@ -414,6 +416,7 @@ export async function recordAiModelHealthFailure(params: {
  const shouldAutoDisable =
    classification.counted &&
    !existing.excludeFromFailover &&
+    !existing.excludeFromHealthBan &&
    ((isObservation && existing.isEnabled) ||
      (existing.healthStatus !== "auto_disabled" &&
        nextConsecutiveFailures >= AUTO_DISABLE_FAILURE_THRESHOLD));
--- a/lib/db/migrate.ts
+++ b/lib/db/migrate.ts
@ -106,6 +106,10 @@ const runMigrate = async () => {
    ALTER TABLE "AiModel"
    ADD COLUMN IF NOT EXISTS "excludeFromFailover" boolean NOT NULL DEFAULT false;
  `);
+  await db.execute(sql`
+    ALTER TABLE "AiModel"
+    ADD COLUMN IF NOT EXISTS "excludeFromHealthBan" boolean NOT NULL DEFAULT false;
+  `);
  await db.execute(sql`
    ALTER TABLE "AiModel"
    ADD COLUMN IF NOT EXISTS "isDefaultTextEditModel" boolean NOT NULL DEFAULT false;
--- a/lib/db/migrations/0083_add_ai_model_health_ban_exclusion.sql
+++ b/lib/db/migrations/0083_add_ai_model_health_ban_exclusion.sql
@ -0,0 +1,2 @@
+ALTER TABLE "AiModel"
+  ADD COLUMN IF NOT EXISTS "excludeFromHealthBan" boolean NOT NULL DEFAULT false;
--- a/lib/db/migrations/meta/_journal.json
+++ b/lib/db/migrations/meta/_journal.json
@ -484,6 +484,13 @@
      "when": 1779267679691,
      "tag": "0082_add_client_performance_events",
      "breakpoints": true
+    },
+    {
+      "idx": 69,
+      "version": "7",
+      "when": 1779849000000,
+      "tag": "0083_add_ai_model_health_ban_exclusion",
+      "breakpoints": true
    }
  ]
 }
--- a/lib/db/queries.ts
+++ b/lib/db/queries.ts
@ -4068,6 +4068,7 @@ export async function createAiModel({
  fallbackModelId1,
  fallbackModelId2,
  excludeFromFailover,
+  excludeFromHealthBan,
  imageGenConfig,
  capabilities,
  isAgentModel,
@ -4089,6 +4090,7 @@ export async function createAiModel({
  fallbackModelId1?: string;
  fallbackModelId2?: string;
  excludeFromFailover?: boolean;
+  excludeFromHealthBan?: boolean;
  imageGenConfig?: ImageGenConfig;
  capabilities?: Record<string, unknown> | null;
  isAgentModel?: boolean;
@ -4139,6 +4141,7 @@ export async function createAiModel({
        fallbackModelId1: fallbackModelId1 || null,
        fallbackModelId2: fallbackModelId2 || null,
        excludeFromFailover: excludeFromFailover ?? false,
+        excludeFromHealthBan: excludeFromHealthBan ?? false,
        imageGenConfig: imageGenConfig || null,
        defaultParams,
        capabilities:
@ -4172,6 +4175,7 @@ export async function updateAiModel({
  fallbackModelId1,
  fallbackModelId2,
  excludeFromFailover,
+  excludeFromHealthBan,
  imageGenConfig,
  capabilities,
  isAgentModel,
@ -4194,6 +4198,7 @@ export async function updateAiModel({
  fallbackModelId1?: string;
  fallbackModelId2?: string;
  excludeFromFailover?: boolean;
+  excludeFromHealthBan?: boolean;
  imageGenConfig?: ImageGenConfig;
  capabilities?: Record<string, unknown> | null;
  isAgentModel?: boolean;
@ -4261,6 +4266,21 @@ export async function updateAiModel({
    if (fallbackModelId1 !== undefined) updateData.fallbackModelId1 = fallbackModelId1 || null;
    if (fallbackModelId2 !== undefined) updateData.fallbackModelId2 = fallbackModelId2 || null;
    if (excludeFromFailover !== undefined) updateData.excludeFromFailover = excludeFromFailover;
+    if (excludeFromHealthBan !== undefined) {
+      updateData.excludeFromHealthBan = excludeFromHealthBan;
+      if (excludeFromHealthBan) {
+        updateData.healthStatus = "healthy";
+        updateData.healthConsecutiveFailures = 0;
+        updateData.healthLastEventType = null;
+        updateData.healthLastError = null;
+        updateData.healthLastFailureAt = null;
+        updateData.healthAutoDisabledAt = null;
+        updateData.healthAutoDisabledReason = null;
+        updateData.healthRecoverySuccessStreak = 0;
+        updateData.healthObservationStartedAt = null;
+        updateData.healthUpdatedAt = new Date();
+      }
+    }
    // 🆕 图片生成配置
    if (imageGenConfig !== undefined) updateData.imageGenConfig = imageGenConfig || null;
    if (
--- a/lib/db/schema.ts
+++ b/lib/db/schema.ts
@ -335,6 +335,7 @@ export const aiModel = pgTable("AiModel", {
  fallbackModelId1: varchar("fallbackModelId1", { length: 128 }), // 备用模型 ID 1
  fallbackModelId2: varchar("fallbackModelId2", { length: 128 }), // 备用模型 ID 2
  excludeFromFailover: boolean("excludeFromFailover").notNull().default(false), // 不参与故障迁移
+  excludeFromHealthBan: boolean("excludeFromHealthBan").notNull().default(false), // 不参与故障自动封禁池
  healthStatus: varchar("healthStatus", {
    enum: ["healthy", "degraded", "auto_disabled"],
  })