From ad672c3c4c255a7810409315e5669bd387095571 Mon Sep 17 00:00:00 2001
From: shaw <shaw-wei@foxmail.com>
Date: Thu, 25 Sep 2025 11:31:30 +0800
Subject: [PATCH] =?UTF-8?q?fix:=20=E4=BF=AE=E5=A4=8Dopenai-respons?=
 =?UTF-8?q?=E8=AE=A1=E8=B4=B9=E9=97=AE=E9=A2=98?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 src/services/openaiResponsesRelayService.js | 46 +++++++++++++++++----
 1 file changed, 38 insertions(+), 8 deletions(-)

diff --git a/src/services/openaiResponsesRelayService.js b/src/services/openaiResponsesRelayService.js
index ca84801b..433242e6 100644
--- a/src/services/openaiResponsesRelayService.js
+++ b/src/services/openaiResponsesRelayService.js
@@ -7,6 +7,32 @@ const unifiedOpenAIScheduler = require('./unifiedOpenAIScheduler')
 const config = require('../../config/config')
 const crypto = require('crypto')
 
+// 抽取缓存写入 token，兼容多种字段命名
+function extractCacheCreationTokens(usageData) {
+  if (!usageData || typeof usageData !== 'object') {
+    return 0
+  }
+
+  const details = usageData.input_tokens_details || usageData.prompt_tokens_details || {}
+  const candidates = [
+    details.cache_creation_input_tokens,
+    details.cache_creation_tokens,
+    usageData.cache_creation_input_tokens,
+    usageData.cache_creation_tokens
+  ]
+
+  for (const value of candidates) {
+    if (value !== undefined && value !== null && value !== '') {
+      const parsed = Number(value)
+      if (!Number.isNaN(parsed)) {
+        return parsed
+      }
+    }
+  }
+
+  return 0
+}
+
 class OpenAIResponsesRelayService {
   constructor() {
     this.defaultTimeout = config.requestTimeout || 600000
@@ -496,24 +522,26 @@ class OpenAIResponsesRelayService {
 
           // 提取缓存相关的 tokens（如果存在）
           const cacheReadTokens = usageData.input_tokens_details?.cached_tokens || 0
+          const cacheCreateTokens = extractCacheCreationTokens(usageData)
           // 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 || totalInputTokens + outputTokens
+          const totalTokens =
+            usageData.total_tokens || totalInputTokens + outputTokens + cacheCreateTokens
           const modelToRecord = actualModel || requestedModel || 'gpt-4'
 
           await apiKeyService.recordUsage(
             apiKeyData.id,
             actualInputTokens, // 传递实际输入（不含缓存）
             outputTokens,
-            0, // OpenAI没有cache_creation_tokens
+            cacheCreateTokens,
             cacheReadTokens,
             modelToRecord,
             account.id
           )
 
           logger.info(
-            `📊 Recorded usage - Input: ${totalInputTokens}(actual:${actualInputTokens}+cached:${cacheReadTokens}), Output: ${outputTokens}, Total: ${totalTokens}, Model: ${modelToRecord}`
+            `📊 Recorded usage - Input: ${totalInputTokens}(actual:${actualInputTokens}+cached:${cacheReadTokens}), CacheCreate: ${cacheCreateTokens}, Output: ${outputTokens}, Total: ${totalTokens}, Model: ${modelToRecord}`
           )
 
           // 更新账户的 token 使用统计
@@ -527,7 +555,7 @@ class OpenAIResponsesRelayService {
               {
                 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_creation_input_tokens: cacheCreateTokens,
                 cache_read_input_tokens: cacheReadTokens
               },
               modelToRecord
@@ -623,23 +651,25 @@ class OpenAIResponsesRelayService {
 
         // 提取缓存相关的 tokens（如果存在）
         const cacheReadTokens = usageData.input_tokens_details?.cached_tokens || 0
+        const cacheCreateTokens = extractCacheCreationTokens(usageData)
         // 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 || totalInputTokens + outputTokens
+        const totalTokens =
+          usageData.total_tokens || totalInputTokens + outputTokens + cacheCreateTokens
 
         await apiKeyService.recordUsage(
           apiKeyData.id,
           actualInputTokens, // 传递实际输入（不含缓存）
           outputTokens,
-          0, // OpenAI没有cache_creation_tokens
+          cacheCreateTokens,
           cacheReadTokens,
           actualModel,
           account.id
         )
 
         logger.info(
-          `📊 Recorded non-stream usage - Input: ${totalInputTokens}(actual:${actualInputTokens}+cached:${cacheReadTokens}), Output: ${outputTokens}, Total: ${totalTokens}, Model: ${actualModel}`
+          `📊 Recorded non-stream usage - Input: ${totalInputTokens}(actual:${actualInputTokens}+cached:${cacheReadTokens}), CacheCreate: ${cacheCreateTokens}, Output: ${outputTokens}, Total: ${totalTokens}, Model: ${actualModel}`
         )
 
         // 更新账户的 token 使用统计
@@ -653,7 +683,7 @@ class OpenAIResponsesRelayService {
             {
               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_creation_input_tokens: cacheCreateTokens,
               cache_read_input_tokens: cacheReadTokens
             },
             actualModel