fix: 修复openai输入token计算问题

scripts/test-total-cost-limit.js

@@ -1,157 +0,0 @@
-#!/usr/bin/env node
-
-const path = require('path')
-const Module = require('module')
-
-const originalResolveFilename = Module._resolveFilename
-Module._resolveFilename = function resolveConfig(request, parent, isMain, options) {
-  if (request.endsWith('/config/config')) {
-    return path.resolve(__dirname, '../config/config.example.js')
-  }
-  return originalResolveFilename.call(this, request, parent, isMain, options)
-}
-
-const redis = require('../src/models/redis')
-const apiKeyService = require('../src/services/apiKeyService')
-const { authenticateApiKey } = require('../src/middleware/auth')
-
-Module._resolveFilename = originalResolveFilename
-
-function createMockReq(apiKey) {
-  return {
-    headers: {
-      'x-api-key': apiKey,
-      'user-agent': 'total-cost-limit-test'
-    },
-    query: {},
-    body: {},
-    ip: '127.0.0.1',
-    connection: {},
-    originalUrl: '/test-total-cost-limit',
-    once: () => {},
-    on: () => {},
-    get(header) {
-      return this.headers[header.toLowerCase()] || ''
-    }
-  }
-}
-
-function createMockRes() {
-  const state = {
-    status: 200,
-    body: null
-  }
-
-  return {
-    once: () => {},
-    on: () => {},
-    status(code) {
-      state.status = code
-      return this
-    },
-    json(payload) {
-      state.body = payload
-      return this
-    },
-    getState() {
-      return state
-    }
-  }
-}
-
-async function runAuth(apiKey) {
-  const req = createMockReq(apiKey)
-  const res = createMockRes()
-  let nextCalled = false
-
-  await authenticateApiKey(req, res, () => {
-    nextCalled = true
-  })
-
-  const result = res.getState()
-  if (nextCalled && result.status === 200) {
-    return { status: 200, body: null }
-  }
-  return result
-}
-
-async function cleanupKey(keyId) {
-  const client = redis.getClient()
-  if (!client) {
-    return
-  }
-
-  try {
-    await redis.deleteApiKey(keyId)
-    const usageKeys = await client.keys(`usage:*:${keyId}*`)
-    if (usageKeys.length > 0) {
-      await client.del(...usageKeys)
-    }
-    const costKeys = await client.keys(`usage:cost:*:${keyId}*`)
-    if (costKeys.length > 0) {
-      await client.del(...costKeys)
-    }
-    await client.del(`usage:${keyId}`)
-    await client.del(`usage:records:${keyId}`)
-    await client.del(`usage:cost:total:${keyId}`)
-  } catch (error) {
-    console.warn(`Failed to cleanup test key ${keyId}:`, error.message)
-  }
-}
-
-async function main() {
-  await redis.connect()
-
-  const testName = `TotalCostLimitTest-${Date.now()}`
-  const totalCostLimit = 1.0
-  const newKey = await apiKeyService.generateApiKey({
-    name: testName,
-    permissions: 'all',
-    totalCostLimit: totalCostLimit
-  })
-
-  const keyId = newKey.id
-  const { apiKey } = newKey
-
-  console.log(`➕ Created test API key ${keyId} with total cost limit $${totalCostLimit}`)
-
-  let authResult = await runAuth(apiKey)
-  if (authResult.status !== 200) {
-    throw new Error(`Expected success before any usage, got status ${authResult.status}`)
-  }
-  console.log('✅ Authentication succeeds before consuming quota')
-
-  // 增加总费用
-  const client = redis.getClient()
-  await client.set(`usage:cost:total:${keyId}`, '0.6')
-
-  authResult = await runAuth(apiKey)
-  if (authResult.status !== 200) {
-    throw new Error(`Expected success under quota, got status ${authResult.status}`)
-  }
-  console.log('✅ Authentication succeeds while still under quota ($0.60)')
-
-  // 继续增加总费用超过限制
-  await client.set(`usage:cost:total:${keyId}`, '1.1')
-
-  authResult = await runAuth(apiKey)
-  if (authResult.status !== 429) {
-    throw new Error(`Expected 429 after exceeding quota, got status ${authResult.status}`)
-  }
-  console.log('✅ Authentication returns 429 after exceeding total cost limit ($1.10)')
-
-  await cleanupKey(keyId)
-  await redis.disconnect()
-
-  console.log('🎉 Total cost limit test completed successfully')
-}
-
-main().catch(async (error) => {
-  console.error('❌ Total cost limit test failed:', error)
-  try {
-    await redis.disconnect()
-  } catch (_) {
-    // Ignore disconnect errors during cleanup
-  }
-  process.exitCode = 1
-})

src/routes/openaiRoutes.js

@@ -390,23 +390,24 @@ const handleResponses = async (req, res) => {
 
         // 记录使用统计
         if (usageData) {
-          const inputTokens = usageData.input_tokens || usageData.prompt_tokens || 0
+          const totalInputTokens = usageData.input_tokens || usageData.prompt_tokens || 0
           const outputTokens = usageData.output_tokens || usageData.completion_tokens || 0
-          const cacheCreateTokens = usageData.input_tokens_details?.cache_creation_tokens || 0
           const cacheReadTokens = usageData.input_tokens_details?.cached_tokens || 0
+          // 计算实际输入token（总输入减去缓存部分）
+          const actualInputTokens = Math.max(0, totalInputTokens - cacheReadTokens)
 
           await apiKeyService.recordUsage(
             apiKeyData.id,
-            inputTokens,
+            actualInputTokens, // 传递实际输入（不含缓存）
             outputTokens,
-            cacheCreateTokens,
+            0, // OpenAI没有cache_creation_tokens
             cacheReadTokens,
             actualModel,
             accountId
           )
 
           logger.info(
-            `📊 Recorded OpenAI non-stream usage - Input: ${inputTokens}, Output: ${outputTokens}, Total: ${usageData.total_tokens || inputTokens + outputTokens}, Model: ${actualModel}`
+            `📊 Recorded OpenAI non-stream usage - Input: ${totalInputTokens}(actual:${actualInputTokens}+cached:${cacheReadTokens}), Output: ${outputTokens}, Total: ${usageData.total_tokens || totalInputTokens + outputTokens}, Model: ${actualModel}`
           )
         }
 
@@ -506,26 +507,27 @@ const handleResponses = async (req, res) => {
       // 记录使用统计
       if (!usageReported && usageData) {
         try {
-          const inputTokens = usageData.input_tokens || 0
+          const totalInputTokens = usageData.input_tokens || 0
           const outputTokens = usageData.output_tokens || 0
-          const cacheCreateTokens = usageData.input_tokens_details?.cache_creation_tokens || 0
           const cacheReadTokens = usageData.input_tokens_details?.cached_tokens || 0
+          // 计算实际输入token（总输入减去缓存部分）
+          const actualInputTokens = Math.max(0, totalInputTokens - cacheReadTokens)
 
           // 使用响应中的真实 model，如果没有则使用请求中的 model，最后回退到默认值
           const modelToRecord = actualModel || requestedModel || 'gpt-4'
 
           await apiKeyService.recordUsage(
             apiKeyData.id,
-            inputTokens,
+            actualInputTokens, // 传递实际输入（不含缓存）
             outputTokens,
-            cacheCreateTokens,
+            0, // OpenAI没有cache_creation_tokens
             cacheReadTokens,
             modelToRecord,
             accountId
           )
 
           logger.info(
-            `📊 Recorded OpenAI usage - Input: ${inputTokens}, Output: ${outputTokens}, Total: ${usageData.total_tokens || inputTokens + outputTokens}, Model: ${modelToRecord} (actual: ${actualModel}, requested: ${requestedModel})`
+            `📊 Recorded OpenAI usage - Input: ${totalInputTokens}(actual:${actualInputTokens}+cached:${cacheReadTokens}), Output: ${outputTokens}, Total: ${usageData.total_tokens || totalInputTokens + outputTokens}, Model: ${modelToRecord} (actual: ${actualModel}, requested: ${requestedModel})`
           )
           usageReported = true
         } catch (error) {

src/services/openaiResponsesRelayService.js

@@ -385,28 +385,29 @@ class OpenAIResponsesRelayService {
       if (usageData) {
         try {
           // OpenAI-Responses 使用 input_tokens/output_tokens，标准 OpenAI 使用 prompt_tokens/completion_tokens
-          const inputTokens = usageData.input_tokens || usageData.prompt_tokens || 0
+          const totalInputTokens = usageData.input_tokens || usageData.prompt_tokens || 0
           const outputTokens = usageData.output_tokens || usageData.completion_tokens || 0
 
           // 提取缓存相关的 tokens（如果存在）
-          const cacheCreateTokens = usageData.input_tokens_details?.cache_creation_tokens || 0
           const cacheReadTokens = usageData.input_tokens_details?.cached_tokens || 0
+          // 计算实际输入token（总输入减去缓存部分）
+          const actualInputTokens = Math.max(0, totalInputTokens - cacheReadTokens)
 
-          const totalTokens = usageData.total_tokens || inputTokens + outputTokens
+          const totalTokens = usageData.total_tokens || totalInputTokens + outputTokens
           const modelToRecord = actualModel || requestedModel || 'gpt-4'
 
           await apiKeyService.recordUsage(
             apiKeyData.id,
-            inputTokens,
+            actualInputTokens, // 传递实际输入（不含缓存）
             outputTokens,
-            cacheCreateTokens,
+            0, // OpenAI没有cache_creation_tokens
             cacheReadTokens,
             modelToRecord,
             account.id
           )
 
           logger.info(
-            `📊 Recorded usage - Input: ${inputTokens}, Output: ${outputTokens}, CacheRead: ${cacheReadTokens}, CacheCreate: ${cacheCreateTokens}, Total: ${totalTokens}, Model: ${modelToRecord}`
+            `📊 Recorded usage - Input: ${totalInputTokens}(actual:${actualInputTokens}+cached:${cacheReadTokens}), Output: ${outputTokens}, Total: ${totalTokens}, Model: ${modelToRecord}`
           )
 
           // 更新账户的 token 使用统计
@@ -414,9 +415,18 @@ class OpenAIResponsesRelayService {
 
           // 更新账户使用额度（如果设置了额度限制）
           if (parseFloat(account.dailyQuota) > 0) {
-            // 估算费用（根据模型和token数量）
-            const estimatedCost = this._estimateCost(modelToRecord, inputTokens, outputTokens)
-            await openaiResponsesAccountService.updateUsageQuota(account.id, estimatedCost)
+            // 使用CostCalculator正确计算费用（考虑缓存token的不同价格）
+            const CostCalculator = require('../utils/costCalculator')
+            const costInfo = CostCalculator.calculateCost(
+              {
+                input_tokens: actualInputTokens, // 实际输入（不含缓存）
+                output_tokens: outputTokens,
+                cache_creation_input_tokens: 0, // OpenAI没有cache_creation
+                cache_read_input_tokens: cacheReadTokens
+              },
+              modelToRecord
+            )
+            await openaiResponsesAccountService.updateUsageQuota(account.id, costInfo.costs.total)
           }
         } catch (error) {
           logger.error('Failed to record usage:', error)
@@ -502,27 +512,28 @@ class OpenAIResponsesRelayService {
     if (usageData) {
       try {
         // OpenAI-Responses 使用 input_tokens/output_tokens，标准 OpenAI 使用 prompt_tokens/completion_tokens
-        const inputTokens = usageData.input_tokens || usageData.prompt_tokens || 0
+        const totalInputTokens = usageData.input_tokens || usageData.prompt_tokens || 0
         const outputTokens = usageData.output_tokens || usageData.completion_tokens || 0
 
         // 提取缓存相关的 tokens（如果存在）
-        const cacheCreateTokens = usageData.input_tokens_details?.cache_creation_tokens || 0
         const cacheReadTokens = usageData.input_tokens_details?.cached_tokens || 0
+        // 计算实际输入token（总输入减去缓存部分）
+        const actualInputTokens = Math.max(0, totalInputTokens - cacheReadTokens)
 
-        const totalTokens = usageData.total_tokens || inputTokens + outputTokens
+        const totalTokens = usageData.total_tokens || totalInputTokens + outputTokens
 
         await apiKeyService.recordUsage(
           apiKeyData.id,
-          inputTokens,
+          actualInputTokens, // 传递实际输入（不含缓存）
           outputTokens,
-          cacheCreateTokens,
+          0, // OpenAI没有cache_creation_tokens
           cacheReadTokens,
           actualModel,
           account.id
         )
 
         logger.info(
-          `📊 Recorded non-stream usage - Input: ${inputTokens}, Output: ${outputTokens}, CacheRead: ${cacheReadTokens}, CacheCreate: ${cacheCreateTokens}, Total: ${totalTokens}, Model: ${actualModel}`
+          `📊 Recorded non-stream usage - Input: ${totalInputTokens}(actual:${actualInputTokens}+cached:${cacheReadTokens}), Output: ${outputTokens}, Total: ${totalTokens}, Model: ${actualModel}`
         )
 
         // 更新账户的 token 使用统计
@@ -530,9 +541,18 @@ class OpenAIResponsesRelayService {
 
         // 更新账户使用额度（如果设置了额度限制）
         if (parseFloat(account.dailyQuota) > 0) {
-          // 估算费用（根据模型和token数量）
-          const estimatedCost = this._estimateCost(actualModel, inputTokens, outputTokens)
-          await openaiResponsesAccountService.updateUsageQuota(account.id, estimatedCost)
+          // 使用CostCalculator正确计算费用（考虑缓存token的不同价格）
+          const CostCalculator = require('../utils/costCalculator')
+          const costInfo = CostCalculator.calculateCost(
+            {
+              input_tokens: actualInputTokens, // 实际输入（不含缓存）
+              output_tokens: outputTokens,
+              cache_creation_input_tokens: 0, // OpenAI没有cache_creation
+              cache_read_input_tokens: cacheReadTokens
+            },
+            actualModel
+          )
+          await openaiResponsesAccountService.updateUsageQuota(account.id, costInfo.costs.total)
         }
       } catch (error) {
         logger.error('Failed to record usage:', error)

web/admin-spa/src/components/apistats/LimitConfig.vue

@@ -215,7 +215,7 @@
                 class="text-orange-600"
               >
                 <i class="fas fa-exclamation-triangle mr-1 text-xs md:text-sm" />
-                限制 {{ statsData.restrictions.allowedClients.length }} 个客户端
+                限 {{ statsData.restrictions.allowedClients.length }} 种客户端使用
               </span>
               <span v-else class="text-green-600">
                 <i class="fas fa-check-circle mr-1 text-xs md:text-sm" />