Merge pull request #314 from sczheng189/feat/5xx-error-circuit-breaker

feat: 改进5xx错误熔断机制和重置状态功能

src/services/claudeAccountService.js

@@ -1769,6 +1769,9 @@ class ClaudeAccountService {
       delete updatedAccountData.rateLimitedAt
       delete updatedAccountData.rateLimitStatus
       delete updatedAccountData.rateLimitEndAt
+      delete updatedAccountData.tempErrorAt
+      delete updatedAccountData.sessionWindowStart
+      delete updatedAccountData.sessionWindowEnd
 
       // 保存更新后的账户数据
       await redis.setClaudeAccount(accountId, updatedAccountData)
@@ -1781,6 +1784,10 @@ class ClaudeAccountService {
       const rateLimitKey = `ratelimit:${accountId}`
       await redis.client.del(rateLimitKey)
 
+      // 清除5xx错误计数
+      const serverErrorKey = `claude_account:${accountId}:5xx_errors`
+      await redis.client.del(serverErrorKey)
+
       logger.info(
         `✅ Successfully reset all error states for account ${accountData.name} (${accountId})`
       )
@@ -1805,7 +1812,7 @@ class ClaudeAccountService {
     try {
       const accounts = await redis.getAllClaudeAccounts()
       let cleanedCount = 0
-      const TEMP_ERROR_RECOVERY_MINUTES = 60 // 临时错误状态恢复时间（分钟）
+      const TEMP_ERROR_RECOVERY_MINUTES = 5 // 临时错误状态恢复时间（分钟）
 
       for (const account of accounts) {
         if (account.status === 'temp_error' && account.tempErrorAt) {

src/services/claudeRelayService.js

@@ -207,7 +207,7 @@ class ClaudeRelayService {
           logger.info(
             `🔥 Account ${accountId} has ${errorCount} consecutive 5xx errors in the last 5 minutes`
           )
-          if (errorCount >= 3) {
+          if (errorCount > 10) {
             logger.error(
               `❌ Account ${accountId} exceeded 5xx error threshold (${errorCount} errors), marking as temp_error`
             )
@@ -939,7 +939,7 @@ class ClaudeRelayService {
               logger.info(
                 `🔥 [Stream] Account ${accountId} has ${errorCount} consecutive 5xx errors in the last 5 minutes`
               )
-              if (errorCount >= 3) {
+              if (errorCount > 10) {
                 logger.error(
                   `❌ [Stream] Account ${accountId} exceeded 5xx error threshold (${errorCount} errors), marking as temp_error`
                 )