Nvidia/BossAI
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: add round robin routing and improve token binding UX

Browse Source
This commit is contained in:
bnvnvnv
2026-03-12 05:17:10 +00:00
parent cb93b28c71
commit 5a7531f635
20 changed files with 1112 additions and 325 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/server/routes/api/settings.ts
+3
View File
@@ -882,7 +882,10 @@ export async function settingsRoutes(app: FastifyInstance) {
totalLatencyMs: 0,
totalCost: 0,
lastUsedAt: null,
lastSelectedAt: null,
lastFailAt: null,
consecutiveFailCount: 0,
cooldownLevel: 0,
cooldownUntil: null,
}).run();
src/server/routes/api/tokens.ts
+42 -34
View File
@@ -1,8 +1,9 @@
import { FastifyInstance } from 'fastify';
import { and, eq, inArray } from 'drizzle-orm';
import { db, schema } from '../../db/index.js';
import { rebuildTokenRoutesFromAvailability, refreshModelsAndRebuildRoutes } from '../../services/modelService.js';
import { invalidateTokenRouterCache, matchesModelPattern, tokenRouter } from '../../services/tokenRouter.js';
import { rebuildTokenRoutesFromAvailability, refreshModelsAndRebuildRoutes } from '../../services/modelService.js';
import { normalizeRouteRoutingStrategy } from '../../services/routeRoutingStrategy.js';
import { invalidateTokenRouterCache, matchesModelPattern, tokenRouter } from '../../services/tokenRouter.js';
import { startBackgroundTask } from '../../services/backgroundTaskService.js';
import {
clearRouteDecisionSnapshot,
@@ -415,14 +416,15 @@ async function fetchChannelsForRoutes(routeIds: number[]): Promise<Map<number, a
export async function tokensRoutes(app: FastifyInstance) {
// List routes with basic info only (lightweight for selectors)
app.get('/api/routes/lite', async () => {
return await db.select({
id: schema.tokenRoutes.id,
modelPattern: schema.tokenRoutes.modelPattern,
displayName: schema.tokenRoutes.displayName,
displayIcon: schema.tokenRoutes.displayIcon,
enabled: schema.tokenRoutes.enabled,
}).from(schema.tokenRoutes).all();
});
return await db.select({
id: schema.tokenRoutes.id,
modelPattern: schema.tokenRoutes.modelPattern,
displayName: schema.tokenRoutes.displayName,
displayIcon: schema.tokenRoutes.displayIcon,
routingStrategy: schema.tokenRoutes.routingStrategy,
enabled: schema.tokenRoutes.enabled,
}).from(schema.tokenRoutes).all();
});
// Route summary (no channel details) for first-screen rendering
app.get('/api/routes/summary', async () => {
@@ -460,10 +462,11 @@ export async function tokensRoutes(app: FastifyInstance) {
id: route.id,
modelPattern: route.modelPattern,
displayName: route.displayName ?? null,
displayIcon: route.displayIcon ?? null,
modelMapping: route.modelMapping ?? null,
enabled: route.enabled,
channelCount: agg?.channelCount ?? 0,
displayIcon: route.displayIcon ?? null,
modelMapping: route.modelMapping ?? null,
routingStrategy: route.routingStrategy ?? 'weighted',
enabled: route.enabled,
channelCount: agg?.channelCount ?? 0,
enabledChannelCount: agg?.enabledChannelCount ?? 0,
siteNames: agg ? Array.from(agg.siteNames) : [],
decisionSnapshot: parseRouteDecisionSnapshot(route.decisionSnapshot),
@@ -676,15 +679,16 @@ export async function tokensRoutes(app: FastifyInstance) {
});
// Create a route
app.post<{ Body: { modelPattern: string; displayName?: string; displayIcon?: string; modelMapping?: string; enabled?: boolean } }>('/api/routes', async (request) => {
const body = request.body;
const insertedRoute = await db.insert(schema.tokenRoutes).values({
modelPattern: body.modelPattern,
displayName: body.displayName,
displayIcon: body.displayIcon,
modelMapping: body.modelMapping,
enabled: body.enabled ?? true,
}).run();
app.post<{ Body: { modelPattern: string; displayName?: string; displayIcon?: string; modelMapping?: string; routingStrategy?: string; enabled?: boolean } }>('/api/routes', async (request) => {
const body = request.body;
const insertedRoute = await db.insert(schema.tokenRoutes).values({
modelPattern: body.modelPattern,
displayName: body.displayName,
displayIcon: body.displayIcon,
modelMapping: body.modelMapping,
routingStrategy: normalizeRouteRoutingStrategy(body.routingStrategy),
enabled: body.enabled ?? true,
}).run();
const routeId = Number(insertedRoute.lastInsertRowid || 0);
if (routeId <= 0) {
return { success: false, message: '创建路由失败' };
@@ -713,17 +717,21 @@ export async function tokensRoutes(app: FastifyInstance) {
if (body.displayName !== undefined) updates.displayName = body.displayName;
if (body.displayIcon !== undefined) updates.displayIcon = body.displayIcon;
if (body.modelPattern !== undefined) {
nextModelPattern = String(body.modelPattern);
updates.modelPattern = nextModelPattern;
}
if (body.modelMapping !== undefined) updates.modelMapping = body.modelMapping;
if (body.enabled !== undefined) updates.enabled = body.enabled;
updates.updatedAt = new Date().toISOString();
await db.update(schema.tokenRoutes).set(updates).where(eq(schema.tokenRoutes.id, id)).run();
const modelPatternChanged = body.modelPattern !== undefined && nextModelPattern !== existingRoute.modelPattern;
const routeBehaviorChanged = modelPatternChanged || body.modelMapping !== undefined || body.enabled !== undefined;
if (body.modelPattern !== undefined) {
nextModelPattern = String(body.modelPattern);
updates.modelPattern = nextModelPattern;
}
if (body.modelMapping !== undefined) updates.modelMapping = body.modelMapping;
if (body.routingStrategy !== undefined) updates.routingStrategy = normalizeRouteRoutingStrategy(body.routingStrategy);
if (body.enabled !== undefined) updates.enabled = body.enabled;
updates.updatedAt = new Date().toISOString();
await db.update(schema.tokenRoutes).set(updates).where(eq(schema.tokenRoutes.id, id)).run();
const modelPatternChanged = body.modelPattern !== undefined && nextModelPattern !== existingRoute.modelPattern;
const routeBehaviorChanged = modelPatternChanged
|| body.modelMapping !== undefined
|| body.routingStrategy !== undefined
|| body.enabled !== undefined;
if (modelPatternChanged) {
await rebuildAutomaticRouteChannelsByModelPattern(id, nextModelPattern);
}