I'm playing around with tuning various tcp settings for high number of outgoing client connections. I've put following in my /etc/sysctl.conf on Ubuntu 16.04.
net.ipv4.ip_local_port_range = 18000 65535
net.ipv4.tcp_tw_recycle = 1
net.ipv4.tcp_tw_reuse = 1
net.ipv4.tcp_fin_timeout = 3
I've applied them using sysctl -p then i made a new connection to one of my servers using netcat then terminated it:
nc 35.158.221.XXX 22
Then i checked netstat using following command :
netstat -on | grep TIME_WAIT | less
In every situation, the connection was removed correctly within 3 seconds period.
Now the problem:
Shortly after above, i've started a python script with a lot of threads spawning outgoing http connections. In this script however, the connections' TIME_WAIT state is expiring STILL after default number of 60 seconds.
Any reason why is this happening and python's connections are ignoring the kernel setting? How is that even possible? Why the setting is not being applied to this process? Did i forget about flushing some cache or anything?