20

I am stacked with following errors on my site when I test 200 hits per second.

First I received 499 errors

2017-04-09 03:22:45 Error 162.158.79.219 499 GET / HTTP/1.1 0 nginx access

2017-04-09 03:22:45 Error 162.158.79.87 499 GET / HTTP/1.1 0 nginx access

2017-04-09 03:22:45 Error 162.158.78.170 499 GET / HTTP/1.1 0 nginx access

2017-04-09 03:22:45 Error 162.158.78.68 499 GET / HTTP/1.1 0 nginx access

2nd Error start showing 502

2017-04-09 03:22:45 Error 162.158.79.135 502 GET / HTTP/1.1 166 nginx access

2017-04-09 03:22:45 Error 162.158.79.225 502 GET / HTTP/1.1 166 nginx access

2017-04-09 03:22:45 Error 162.158.78.110 502 GET / HTTP/1.1 166 nginx access

2017-04-09 03:22:45 Error 162.158.79.225 502 GET / HTTP/1.1 166 nginx access

and Finally I start receiving php-fpm.sock failed errors

2017-04-09 03:22:45 Error 162.158.79.207 20699#0: *3826365 connect() to unix:///var/www/vhosts/system/playhdpk.top/php-fpm.sock failed (11: Resource temporarily unavailable) while connecting to upstream nginx error

2017-04-09 03:22:45 Error 162.158.79.207 20695#0: *3826367 connect() to unix:///var/www/vhosts/system/playhdpk.top/php-fpm.sock failed (11: Resource temporarily unavailable) while connecting to upstream nginx error

2017-04-09 03:22:45 Error 162.158.79.207 20697#0: *3826369 connect() to unix:///var/www/vhosts/system/playhdpk.top/php-fpm.sock failed (11: Resource temporarily unavailable) while connecting to upstream nginx error

My php-fpm-pool-settings are below, I believe this one is being generating errors may I am wrong

listen.backlog = 65535 ;[php-fpm-pool-settings] pm = dynamic pm.max_children = 5000 pm.start_servers = 50 pm.min_spare_servers = 20 pm.max_spare_servers = 70 pm.max_requests = 2000

My nginx conf are below

user nginx; worker_processes 8; # number of file descriptors used for nginx # the limit for the maximum FDs on the server is usually set by the OS. # if you don't set FD's then OS settings will be used which is by default 2000 worker_rlimit_nofile 99999; error_log /var/log/nginx/error.log crit; include /etc/nginx/modules.conf.d/*.conf; events { worker_connections 16192; use epoll; multi_accept on; } http { include mime.types; default_type application/octet-stream; open_file_cache max=2048 inactive=20s; open_file_cache_valid 30s; open_file_cache_min_uses 5; open_file_cache_errors off; sendfile on; tcp_nopush on; tcp_nodelay on; keepalive_requests 100000; reset_timedout_connection on; client_body_timeout 30; send_timeout 15; client_header_timeout 12; proxy_connect_timeout 600s; proxy_send_timeout 600s; proxy_read_timeout 600s; fastcgi_buffers 8 128k; fastcgi_buffer_size 256k; fastcgi_send_timeout 600s; fastcgi_read_timeout 600s; types_hash_max_size 2048; gzip on; gzip_min_length 1000; gzip_proxied expired no-cache no-store private auth; gzip_types application/x-javascript text/css application/javascript text/javascript text/plain text/xml application/json application/v$ gzip_disable "MSIE [1-6]\.(?!.*SV1)"; server_tokens off; include /etc/nginx/conf.d/*.conf; }

Note: Server specifications are below

OS CentOS 7.3

Processor: Intel Xeon E5-1620v2 - 4c/8t - 3.7 GHz/3.9 GH

Sever Ram: 64GB DDR3

Improve this question
1

3 Answers 3

Reset to default
23

If I could downvote the answer from @Artsiom forever, I would.

pm.max_children = 4000 means up to 4K worker processes. If traffic flows in very fast and together with pm.max_requests = 0, the workers are never recycled, the RAM usage will grow indefinitely over time and the server will be in out of memory condition (down, frozen) sooner or later.

PHP-FPM max_children should be raised carefully and gradually while monitoring swap usage.

You can use a formula liks the following:

pm.max_children = ((total RAM in MB) - (how much MySQL and others take in RAM)) / 80

Where 80 MB is average weight of a PHP-FPM workers process if your PHP framework is light. For heavy things like Magento 2, take at least 128 MB instead.

And the pm.max_requests should be some "limited" value. In higher spec servers you can indeed raise it (10000, for example), while on low end servers this should be set to smallest (e.g. 500, to even 100) to reduce RAM "usage" fluctuations. But it no scenario I would set it to 0 (unlimited) because a value of 0 implies that your code / PHP and all of its extensions are absolutely free from memory leaks. Only then it would be fine to be set to 0!!!

Improve this answer
4
  • Well I will give you an upvote for your great explanation. Any suggestions for settings like pm.max_children on a low traffic setup that is slowly growing in trafic? Commented Feb 12, 2020 at 13:02
  • pm.max_children should follow the same calculations, irrespective of the traffic. For nearly zero traffic, you want to have pm set to ondemand so that if no traffic, there are no worker processes at all. Commented Feb 12, 2020 at 13:04
  • how'd you get the (how much MySQL and others take in RAM) Commented Sep 15, 2022 at 2:12
  • @KimCarlo You can use mysqltuner to see the max memory footprint possible for your MySQL configuration. Commented Sep 15, 2022 at 11:25
16

This happens because the operating system rejects nginx's attempts to connect to a unix socket.

The reason is either the maximum number of socket connections or the maximum number of unhandled socket connections has been exceeded.

Checking the limits:

sysctl net.core

We are interested in the lines:

net.core.somaxconn = 128 net.core.netdev_max_backlog = 200

Because of them, an error occurs, since the maximum number of connections is 128 and the maximum number of unprocessed is 200

Change the limits, write the lines in the / etc / sysctl.conf file

nano /etc/sysctl.conf

add

net.core.somaxconn = 20000 net.core.netdev_max_backlog = 65535

Apply parameters

sysctl -p

Restart php-fpm

service php-fpm restart

Source: https://galaxydata.ru/community/sock-failed-11-resource-temporarily-unavailable-459

Improve this answer
-4

Do not limit requests Give them work free)

pm = ondemand pm.max_children = 4000 pm.start_servers = 10 pm.min_spare_servers = 10 pm.max_requests = 0
Improve this answer
2
  • 1
    This makes no sense. max_requests is jsut to occasionally recycle children. 2k should be plenty to not have issues. Commented Apr 26, 2019 at 6:37
  • 2
    @jgmjgm not only it makes no sense, it's also dangerous and completely irrational. The upvoters here applied Artsiom's suggestion only to gain grief later. See my answer with an explanation. Commented Nov 16, 2019 at 23:24

You must log in to answer this question.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.