Resolved [alert] 32414#0: socket() failed (24: Too many open files) while connecting to upstream

[Pinocchio]

Hi,

Today, my website reached 4.000 concurrent users and i got lots of errors like:

[alert] 32414#0: .... socket() failed (24: Too many open files) while connecting to upstream.

How do I fix this?

 

[IgorG]

Increase maximum number of open files:

# ulimit -n 8192

To increase these limits permanently:

1. Add the following lines in /etc/security/limits.conf file before # End of file line:
   # tail -n4 /etc/security/limits.conf
   * soft nofile 8192
   * hard nofile 8192
   
   # End of file
   
   
2. Reboot the server to apply these changes.

 

[Pinocchio]

Thanks, I added those lines and saved the file. Shall I reboot using the following command?

# service sw-engine restart && service sw-cp-server restart

 

[IgorG]

Reboot means server reboot:

# reboot

 

[Pinocchio]

Ok, thanks again.

 

[SLOposum]

Ok, thanks again.

Having same problems. Running a multisite and when one of the sites publishes an articles, many of these errors appear.
Did you solve this @Pinocchio?

 

[dj ramc]

same here and @IgorG solution doesnt helped

 

[dj ramc]

I edited the nginx.conf file.

etc/systemd/system/nginx.service

[Service]
LimitNOFILE=16384

and

/etc/nginx/nginx.conf

worker_processes auto; #1

events {
worker_connections 4096; #1024
}

We'll see if it helps.

 

[luis.talora]

I edited the nginx.conf file.

and


We'll see if it helps.

Did it help?

Thanks!

 

[Hangover2]

The issue can come from different sources on your server. For Debian, for example, consider increasing the following values to avoid hitting open file limits (web server / mail / PHP):

/etc/security/limits.conf

*         hard    nofile      1048576
*         soft    nofile      1048576
root      hard    nofile      1048576
root      soft    nofile      1048576

/etc/sysctl.d/10-own-settings.conf

# Kernel 5.11 will raise those limits automatically based on host memory, but 5.10 is still using default low values: 8192 / 128
fs.inotify.max_user_watches = 560144
fs.inotify.max_user_instances = 1024

/etc/systemd/system.conf

DefaultLimitNOFILE=1048576:1048576
DefaultLimitNPROC=1048576:1048576

/etc/systemd/user.conf

DefaultLimitNOFILE=1048576:1048576
DefaultLimitNPROC=1048576:1048576

/opt/plesk/php/*/etc/php-fpm.conf

rlimit_files = 524288

/etc/nginx/nginx.conf

# CPU core count * 1024
worker_connections 32768;

/etc/nginx/ulimit.global_params

worker_rlimit_nofile 524288;

/etc/dovecot/conf.d/10-master.conf

default_client_limit = 4299

/etc/systemd/system/nginx.service.d/override.conf

[Service]
LimitNOFILE=1048576

/etc/systemd/system/sw-cp-server.service.d/override.conf

[Service]
LimitNOFILE=524288

/etc/systemd/system/mariadb.service.d/override.conf

[Service]
LimitNOFILE=524288

/etc/apache2/envvars

APACHE_ULIMIT_MAX_FILES='ulimit -n 524288'

For other OSes, look for the corresponding settings. It also makes sense to monitor the configuration for changes, for example via Ansible, to make it more resilient.