RDS Postgres 16.8 遇特定工作负载触发SIGHUP致连接中断求助

环境与问题现象 #

• 实例：AWS RDS t4g.micro
• 数据库版本：Postgres 16.8
• 仅启用扩展：pgvector 0.8.0
• 问题：处理特定工作负载时数据库不稳定，随机失效（有时运行1小时后失效，有时仅5分钟）

数据库日志 #

2025-06-25 14:06:39 UTC::@:[1339]:LOG: checkpoint starting: time
2025-06-25 14:06:41 UTC::@:[1339]:LOG: checkpoint complete: wrote 5 buffers (0.0%); 0 WAL file(s) added, 0 removed, 1 recycled; write=0.899 s, sync=0.013 s, total=2.267 s; sync files=29, longest=0.004 s, average=0.001 s; distance=65975 kB, estimate=65975 kB; lsn=41D/806E348, redo lsn=41D/806E310
2025-06-25 14:11:32 UTC::@:[1338]:LOG: skipping missing configuration file "/rdsdbdata/config/recovery.conf"
2025-06-25 14:11:31 UTC::@:[1337]:LOG: received smart shutdown request
2025-06-25 14:11:31 UTC::@:[1337]:LOG: received SIGHUP, reloading configuration files
2025-06-25 14:11:31 UTC:83.87.175.98(58140):postgres@dbname:[17054]:FATAL: terminating connection due to administrator command
2025-06-25 14:11:31 UTC::@:[1337]:LOG: skipping missing configuration file "/rdsdbdata/config/recovery.conf"
2025-06-25 14:11:31 UTC:83.87.175.98(58136):postgres@dbname:[17049]:FATAL: terminating connection due to administrator command
2025-06-25 14:11:31 UTC:172.31.76.8(39994):postgres@dbname:[17567]:FATAL: terminating connection due to administrator command

已排查信息 #

• 日志分析：收到RDS发送的SIGHUP信号及智能关闭请求前，无任何异常日志
• 资源状态：内存、磁盘资源未耗尽，CPU负载等指标均处于正常范围（监控截图佐证）
• 已尝试的修复操作：
  • 升级RDS实例规格
  • 更新数据库相关软件包
  • 删除现有索引
• 对比测试：在AWS EC2上部署的同版本Postgres+pgvector环境，运行相同工作负载未出现该问题

请求 #

恳请提供针对性的技术解决方案。

内容的提问来源于stack exchange，提问作者Jelmer Wind