[nginx-ingress-controller] improve nginx performance #1379
Conversation
|
@bprashanth ping |
| args: | ||
| - -c | ||
| - sysctl -w net.core.somaxconn=32768 | ||
| && sysctl -w net.ipv4.ip_local_port_range='1024 65535' |
There was a problem hiding this comment.
just make this a while true loop and you get free reconciliation?
There was a problem hiding this comment.
eg:
command:
- /bin/sh
- -c
- |
while true; do sysctl && sysctl; sleep 10; done
| - /bin/sh | ||
| - -c | ||
| - | | ||
| while true; do sysctl -w net.core.somaxconn=32768 && sysctl -w net.ipv4.ip_local_port_range='1024 65535'; sleep 10; done |
There was a problem hiding this comment.
you can actually split this into lines now that you have the "|", for readability.
while true; do
sysctl stuff
done
Also aren't there a couple of other nice tcp level tunables? maybe we should consider a custom entrypoint script that people can extend?
There was a problem hiding this comment.
It would also be good to note down what are per ns sysctls vs shared between host and ns. I don't think any of the tcp tunables are per netns actually. I do think the ones in /sys/net/core are per net ns.
| - | | ||
| while true; do | ||
| sysctl -w net.core.somaxconn=32768 | ||
| sysctl -w net.ipv4.ip_local_port_range='1024 65535' |
There was a problem hiding this comment.
what about backlog queue size? is that not per ns?
There was a problem hiding this comment.
what about backlog queue size? is that not per ns?
No
/ # sysctl -A | grep "net.core.netdev_max_backlog"
sysctl: error reading key 'net.ipv6.conf.all.stable_secret': I/O error
sysctl: error reading key 'net.ipv6.conf.default.stable_secret': I/O error
sysctl: error reading key 'net.ipv6.conf.eth0.stable_secret': I/O error
sysctl: error reading key 'net.ipv6.conf.lo.stable_secret': I/O error
/ #
|
LGTM |
| # IPC Namespace: kernel.msgmax, kernel.msgmnb, kernel.msgmni, kernel.sem, kernel.shmall, | ||
| # kernel.shmmax, kernel.shmmni, kernel.shm_rmid_forced and Sysctls | ||
| # beginning with fs.mqueue.* | ||
| # Network Namespace: Sysctls beginning with net.* |
There was a problem hiding this comment.
actually this is not true right? dont think fin timeout is per ns? or is it
Changing container
/procvalues with a privileged sidecar.NGINX reads
net.core.somaxconnto increase the size of backlog queue of pending connections