In the morning a RABBITMQ and Java server, the client feedback Super card, see IO and CPU is not high. After finding sixty thousand or seventy thousand messages squeezed, the temporary problem solved, looking at/var/log/messages, found many tcp:time wait bucket table overflow, as follows:
Nov 10:36:08 iz237hn51s7z kernel:TCP:time wait bucket table overflow
Nov 10:36:08 iz237hn51s7z kernel:TCP:time wait bucket table overflow
Nov 10:37:07 iz237hn51s7z kernel: __ratelimit:47 callbacks suppressed
Nov 10:37:07 iz237hn51s7z kernel:TCP:time wait bucket table overflow
Nov 10:37:07 iz237hn51s7z kernel:TCP:time wait bucket table overflow
Nov 10:37:07 iz237hn51s7z kernel:TCP:time wait bucket table overflow
Nov 10:37:07 iz237hn51s7z kernel:TCP:time wait bucket table overflow
Nov 10:37:07 iz237hn51s7z kernel:TCP:time wait bucket table overflow
Nov 10:37:07 iz237hn51s7z kernel:TCP:time wait bucket table overflow
Nov 10:37:07 iz237hn51s7z kernel:TCP:time wait bucket table overflow
After investigation, found net.ipv4.tcp_max_tw_buckets = the #已经修改
Netstat-ano | grep Wait | Wc-l found more than 4,400, further query, found a large number of native to MySQL connection status in this state, further found that there is an application of the maxstatement is set to 0.
Linux tcp:time wait bucket table overflow