php-fpm not generating core dump files despite being configured to

Question

I'm running Centos 7 with php-fpm 5.4. I'm encountering a (signal 11) SIGSEGV error in my php-fpm error logs and want to debug it.

I've followed the advice of php5 fpm: how do you generate core dump from segfault? but php-fpm is still not even attempting to generate core dumps. From the php-fpm error log:

WARNING: [pool website.com] child 26953 exited on signal 11 (SIGSEGV) after 0.931337 seconds from start

If a core dump had been generated, it would read (SEGSEGV - core dumped) but it doesn't.

Answer 1

I also had trouble with this. It seems that on centos the master php-fpm process is launched with root privileges, which then launches the individual pools with reduced privileges (user php-fpm on centos by default).

The stumbling block for me, and I suspect for you too, is that if a process has changed it's own execution privileges, it is not permitted to produce a core dump file unless /proc/sys/fs/suid_dumpable is set to allow this. See information about this kernel-level control here.

Answer 2

I was able to solve this problem with the following:

1. Added to /etc/systemd/system.conf file:

DefaultLimitCORE=infinity

2. Changed the kernel.core_pattern:

sysctl kernel.core_pattern='| /usr/lib/systemd/systemd-coredump %p %u %g %s %t %c %e'

3. Reloaded systemd configuration:

systemctl daemon-reload

As an extra step for php-fpm, the following directive needs to be set in the individual php-fpm pool files you want to monitor:

rlimit_core = unlimited

The default directory for the systemd-coredump program is /var/lib/systemd/coredump so you should see dump files generated here.

Answer 3

PHP-FPM switcher user before starting the pools. Hence I had to change fs.suid_dumpable to 1 as mentioned on

https://itnext.io/linux-processes-core-dumps-systemd-coredump-and-debian-cf2ddb11ef9e

sysctl -w fs.suid_dumpable=1