Error 502 in nginx + php5-fpm

I have some error with subj. Server doesn't high loaded: ~15% CPU, there are several Gb of memory, HDD is not buisy. But error 502 throws approximately in 3% of cases.

Programs: Debian 6, nginx/0.7.62, php5-fpm (5.3.3-1).

In error.log of nginx is this error:

connect() to unix:/var/run/php5-fpm.sock failed

State of php5-fpm usually like this:

accepted conn:   41680
pool:             www
process manager:  dynamic
idle processes:   258
active processes: 1
total processes:  259

I think, this mean loading is not high.

I have increased backlog params: in sysctl - net.core.somaxconn = 5000, in php-fpm pool - listen.backlog = 5000. No effect.

I quote a configuration:

/etc/nginx/nginx.conf

user www-data;
worker_processes  8;
timer_resolution 100ms;
worker_rlimit_nofile 20240;
worker_priority -5;

error_log  /var/log/nginx/error.log;
pid        /var/run/nginx.pid;

events {
    worker_connections  2048;
    use epoll;
    # multi_accept on;
}

http {
    include       /etc/nginx/mime.types;

    access_log  /var/log/nginx/access.log;

    sendfile        on;
    tcp_nopush     on;

    #keepalive_timeout  0;
    keepalive_timeout  65;
    tcp_nodelay        on;

    gzip  on;
    gzip_min_length 1100;
    gzip_buffers 64 8k;
    gzip_comp_level 3;
    gzip_http_version 1.1;
    gzip_proxied any;
    gzip_types text/plain application/xml application/x-javascript text/css;
    gzip_disable "MSIE [1-6]\.(?!.*SV1)";

    include /etc/nginx/conf.d/*.conf;
    include /etc/nginx/sites-enabled/*;

    client_max_body_size 100M;
    server_tokens off;
}

/etc/nginx/php_location

fastcgi_pass   unix:/var/run/php5-fpm.sock;
fastcgi_index  index.php;
fastcgi_param  SCRIPT_FILENAME  $document_root/$fastcgi_script_name;
fastcgi_buffers 256 128k;
#fastcgi_buffer_size 16k;
#fastcgi_busy_buffers_size 256k;
fastcgi_connect_timeout 300s;
fastcgi_send_timeout 300s;
fastcgi_read_timeout 300s;
include fastcgi_params;

php-fpm pool

[www]
listen = /var/run/php5-fpm.sock
listen.backlog = 5000
listen.owner = www-data
listen.group = www-data
listen.mode = 0666
user = www-data
group = www-data
pm = dynamic
pm.max_children = 1024
pm.start_servers = 64
pm.min_spare_servers = 64
pm.max_spare_servers = 128
pm.max_requests = 32000
pm.status_path = /system/php5-fpm-status
slowlog = /var/www/log/php-fpm.log.slow
chdir = /var/www

What can I do to optimize this system and make this use all server resources?

PS. I'm sorry, my english is bad.

The issue is socket itself, its problems on high-load cases is well-known. Please consider using TCP\IP connection instead of unix socket, for that you need to make these changes:

• in php-fpm pool configuration replace listen = /var/run/php5-fpm.sock with listen = 127.0.0.1:7777
• in /etc/nginx/php_location replace fastcgi_pass unix:/var/run/php5-fpm.sock; with fastcgi_pass 127.0.0.1:7777;

On Centos 7, Plesk 12.5

I had this problem after my harddisc went full and some services failed. Other Domains workt perfectly, but not one of them it only gave me 502 and similar as timeouts. From the Errorlog:

[crit] 3112#0: *65746768 connect() to 
unix:///var/www/vhosts/system/sub.domain.de/php-fpm.sock failed 
(2: No such file or directory) while connecting to upstream

To solve it, I had to (first make space available and then) restart php-fpm and nginx - then this error vanished!