Logrotate causes php5-fpm downtime

logrotatephp-fpm

I've noticed that one of our servers starts returning errors just after logrotate runs, i.e. in nginx error log I can see:

[error] 8501#0: *118126869 recv() failed (104: Connection reset by
peer) while reading response header from upstream, client: xxx.yyy.zz.ww, server: www.test.com, request: "GET /index.php HTTP/1.1", upstream: "fastcgi://127.0.0.1:9011", host: "www.test.com"

I have tried adding a postrotate action that would make the php reload gracefully but the error is still happening, our current logrotate is as follows:

/var/log/php5-fpm.log {
        daily
        missingok
        rotate 52
        compress
        delaycompress
        notifempty
        create 644 root root
        postrotate
                [ ! -f /var/run/php5-fpm.pid ] || kill -USR2 `cat /var/run/php5-fpm.pid`
        endscript
}

PHP config is as follows:

[www-9011]

user = www-data
group = www-data
listen = 127.0.0.1:9011
listen.backlog = 65535
pm = ondemand
pm.max_children = 50
pm.process_idle_timeout = 10s;
pm.max_requests = 500
rlimit_files = 16384
chdir = /
catch_workers_output = no
php_admin_value[error_log] = /var/log/fpm-php.www.log
php_admin_flag[log_errors] = on

We're running on ubuntu 12.04 and php 5.3.10

Best Answer

Send USR1 instead

https://github.com/php/php-src/blob/b7a7b1a624c97945c0aaa49d46ae996fc0bdb6bc/sapi/fpm/fpm/fpm_events.c#L94

The source code shows this is specifially for rotating files, I know Ubuntu 14.04 didn't handle fpm reloads (USR2) very well, I assume its the same for older versions too.

So change to 

postrotate
                [ ! -f /var/run/php5-fpm.pid ] || kill -USR1 `cat /var/run/php5-fpm.pid`
endscript

to simply rotate the logs

Related Topic