Based on customer experiences in stress testing, this document summarizes common performance issues in stress testing, and provides troubleshooting solutions as well as suggestions.
If public network access is not enabled when you purchase CVM, forwarding may fail when a public network CLB is mounted to the CVM instance.
If the real server has a low bandwidth, it cannot return packets to the CLB when the threshold is exceeded. CLB will return a 504 or 502 error to the client.
If the number of clients is too small or the range of client ports is too narrow, client ports become insufficient and connections will fail to be established. In addition, if the keep_alive
value is greater than 0 when a persistent connection is established, the connection will permanently use the port, which reduces the number of available client ports.
After a request reaches a real server through CLB, the load on the real server is normal. However, because applications on real servers also rely on other applications such as database, performance issues in the database may also affect the stress testing performance.
The health status of real servers may be ignored in stress testing. If the real server has a health check failure or unstable health check status (sometimes good and sometimes bad with rapid changes), stress testing may have poor performance.
After session persistence is enabled for CLB, requests may be distributed to fixed real servers. The traffic distribution becomes uneven, affecting the performance of stress testing. We recommend you disable session persistence during stress testing.
The following configurations are only used for CLB stress testing. You do not need to have them in your production environment.
Was this page helpful?