Peak cluster environment within the pod, TCP time_wait state can reach 4 w +, a problem of application of TCP time_wait state peak can reach 11 k +, after reaching a certain number program can not receive data, check the netstat and accumulate a lot of packages, receive buffer connection is short of HTTP connections, test cases are about 200 connections per second, the use of evhttp libevent, currently tried to revise the sysctl. Conf failed, because the file corresponding data in the pod is read-only, for years, a move, the whole, is to modify the libevent somewhere or evhttp low-level calls or make a modified sysctl. Conf read-only access (or mirror packaging can be written to the file?) Or any other operation

CodePudding user response:

https://www.baidu.com/s? B6 wd=8 a E7 timewait % % % % E6 E5 E8 % BF % 87% % 80% % 81% A4 A3 A7 E8 9 a % % % % % E5 B3 % E6%96% % 86% % B9 % E6 A1%88