I run this blog on a small Amazon Lightsail instance (1 GB RAM, 2 vCPU). Cheap!
Most of the time, it’s fine, but it has a bad habit of dying sometimes on a large image upload. The Jetpack app reports the upload as failing (offering to try again) but my site is completely unresponsive. I can resolve this in one of two ways: either I can still SSH into the box, and restarting PHP with sudo systemctl restart php-fpm recovers and I can try the upload again; or the instance isn’t responsive to SSH and I have to reboot it via the Lightsail console. The CPU utilization is elevated during this period, eating into the burstable zone.
I have seen generally better performance since I updated /etc/php-fpm.d/www.conf to override pm = static and pm.max_children = 2, but something was obviously still wrong. I installed atop since I often couldn’t get on the host to see what was happening right when it was stuck.
Last week, while working on a new post, my instance got wedged in the same way while trying to upload a short video. From the JetPack activity log I could see it became unresponsive at 11:42 PM Seattle time; the instance and atop use UTC so I’d need to look at the minutes leading up to to 0642 in the weekly atop log. That looks like this:
Well there’s my problem! php-fpm spikes in CPU, and then kswapd0 pegs the CPU while the instance’s local disk goes wild with reads trying to keep up. One odd thing is while the memory is low, it doesn’t seem to have changed much for the two php-fpm children. I guess I either need to tune the swappiness on my host or try to put a better clamp on PHP processing images? Need to investigate further.
I’m sure the real solution is to not host my own LAMP stack, but where’s the fun in that?
https://blog.ultranurd.net/2024/07/21/investigating-failed-wordpress-uploads/
#amazonLinux2023 #apache #aws #debugging #httpd #lightsail #linux #phpFpm #wordpress
