Laravel 5.2 + PHP-FPM + NGINX CPU leak

Workers processes should probably match number of cores, not number of threads

@willvincent this is not a problem at all because this are nginx settings and all the CPU goes to php-fpm processes. My current php pool settings are like so: pm = dynamic pm.max_children = 10 pm.start_servers = 3 pm.min_spare_servers = 2 pm.max_spare_servers = 4 pm.max_requests = 500

The CPU usage distributes between 10 php-fpm processes and the CPU load is between 60 and 100% when a lot of concurrent requests are made. And it's the same with fresh Laravel install, no packages, no database, nothing.

Any ideas how to handle heavy load systems in Laravel?

I have the same problem with my website right now.

This is image with CPU usage: https://farm2.staticflickr.com/1534/24828268030_ea90659ca7_o.jpg

I have search on Google and trying to change some setting on www.conf as pm = dynamic pm.max_children, pm.start_servers ... but it still exist.

It does this on any request regardless of what the code is. Your machine/setup just can't handle the connections. This is where you would need a scalable solution in place.

I'm having the same issue right now: http://imgur.com/SfJ0RQ0

It was working fine earlier then this suddenly happen. You guys? figure it out?

Still not :(. But before to do anything else I'll ask a sys admin guy to do some server kernel tuning. As he explained me there is a chance that the network adapter can be flooded because of the heavy traffic. To be hones he was speaking in Chinese when explaining this and was science-fiction to me :D. He told that for heavy loaded server each network adapter interrupt needs to be handled from a different CPU core. This will make your setup to be able to handle much more traffic. In some web articles I red about that the people were speaking for from 20% to 220% better results when doing this. In his opinion the problem is that you/we are receiving too much requests in same time. The server can handle them, the bandwidth can handle them but network adapter not. In that case you will get a lot requests, but the responses will delayed because of this problem and the php will wait and will not release the resources it's using. He had another experience with this problem before with a system that is used 24/7 from 700 people. The issues there were just the same I have- slow response and timeouts. After this tuning it's working as a charm.

Not sure that this will help but it's another point of view.

@aginev What cache do you have? I usually use the default settings and have no issues. About the max_spare_servers, I know it is used to calculate the pm.start_servers but I wouldn't worry too much and put more than expected. My default value is 35.
About the pm.max_requests = 500. It is usually default to 0 for "endless request processing" as in the doc (I am working with CeNtOS)

Now that said I am not sure I understand chinese either ;-) Hope this help

@mdecooman I tested with a lot of different values in the php config. I tried adding many pools, different types of php pooling, HHVM... but with no reasonable improvement. The app is really heavy loaded 1000 active user in the same time. Nginx CGI cache is enabled (static pages only), all user related data is stored in Redis for 10min, all css and js files are merged and minified to limit the number of requests to the server, all the file downloads are via Nginx X-Accel...

Hi @aginev, @willvincent made a valid point about worker_processes. The nginx doc:

## If you're using an Nginx version below 1.3.8 or 1.2. then uncomment
## the line below and set it to the number of cores of the
## server. Otherwise nginx will determine it automatically.
#worker_processes 4;        

You probably checked this but just in case. Try to test from a very basic environment (with Nginx and Php-fpm before adding the other layers). I would love to have more time to help since I love playing with servers...