容器化Azure应用服务通过Hybrid Connection无法关闭MSSQL连接

环境差异现象 #

• 本地环境：使用node-mssql连接池，连接可正常复用，且能在配置的超时时间（如30秒）内自动关闭
• 生产环境（Azure App Service容器 + Azure Hybrid Connection + 本地SQL Server）：连接可复用，但无法自动关闭；超过超时时间后新请求会建立新连接，旧连接持续残留
• 例外场景：本地通过VPN直接连接生产数据库时，连接池功能完全正常

代码实现 #

return sql.connect(config).then((pool) => {
  return pool.query('select * from location')
});

日志信息 #

mssql:base pool(1): created +0ms
mssql:base pool(1): connecting +2ms
mssql:tedi pool(1): connection #1 created +0ms
mssql:tedi connection(1): establishing +0ms
mssql:tedi connection(1): established +42ms
mssql:base pool(1): connected +48ms
mssql:tedi connection(1): destroying +1ms
mssql:tedi connection(1): destroyed +2ms
mssql:base request(1): created +0ms
mssql:tedi pool(1): connection #2 created +7ms
mssql:tedi connection(2): establishing +1ms
mssql:tedi connection(2): established +19ms
mssql:tedi connection(2): borrowed to request #1 +0ms
mssql:tedi request(1): query select * from location +0ms
mssql:base connection(2): released +45ms
mssql:tedi request(1): completed +7ms
mssql:base pool(1): already connected, executing connect callback immediately +7s
mssql:base request(2): created +7s
mssql:tedi connection(2): borrowed to request #2 +7s
mssql:tedi request(2): query select * from location +0ms
mssql:base connection(2): released +13ms
mssql:tedi request(2): completed +7ms
mssql:tedi connection(2): destroying +37s
mssql:tedi connection(2): destroyed +1ms

排查结论 #

• 日志显示node-mssql已在30秒超时后执行socket销毁操作，但本地SQL Server端的连接并未实际关闭
• 同环境下的.NET Core非容器化应用，通过Azure Hybrid Connection访问数据库时无休眠连接残留
• 深入分析node-mssql代码后发现：客户端在查询执行完成后会主动销毁socket，但Azure Hybrid Connection Manager未将关闭指令同步到本地数据库端，导致连接持续残留

内容的提问来源于stack exchange，提问作者Wright Wang