Silex: ExpressJS for PHP | Hacker News Reader

Silex: ExpressJS for PHP

bergie.iki.fi

19 pointsbergie11 comments

Since I follow most of the mailing lists for Symfony 2, I have come across Silex a little while ago. However, something new I took from this article was the appserver-in-php project https://github.com/indeyets/appserver-in-php

The blurb on the git project has "The main idea is, that your app, if built for this protocol, will be able to preload resources, preconnect to databases and response to requests really fast.", which seems like it's taking a persistent approach to running PHP apps. How does this deal with the memory leaks etc?

Can anybody knowledgeable weigh in on this with a little detail about what it is, how stable it is, who's using it, how it stacks up to using PHP-FPM in a spawner pool w/ nginx etc... ?

I'm the author of the original post. We use AiP quite a bit in production.

Indeed, memory leaks are a potential issue with persistent PHP. Garbage collection in 5.3 helps a bit, but in general AiP exposes such issues more readily than regular mod_php setups. This doesn't mean memory leaks in your code wouldn't be an issue anyway, it just means you'll notice them a lot more quickly.

AiP workers die when PHP's memory_limit is exceeeded, and AiP will then restart them.

Thank you for your reply. I will have a play with it if I find some free time. It's not the memory leaks in _my_ code I'm so worried about, it's the ones in other people's code :-) e.g. Doctrine used to leak like a sieve and no amount of free()s and unset()s cleared the ORM objects up under some circumstances. Long-running Symfony tasks were a great way to illustrate that. It seems like a sensible approach to auto-restart the processes once memory_limit is exceeded.

During one minute of siege to the "hello world" example from my blog post, the memory usage of my PHP processes didn't change.

  13958 bergie    20   0  125m  10m 2136 S    1  0.6   0:00.28 php        
  13959 bergie    20   0  125m  10m 2136 S    1  0.6   0:00.31 php

And the same thing with Node:

  14051 bergie    20   0  615m  12m 4988 S    0  0.7   0:00.34 node

Node and ExpressJS did 3.79 transactions per second, AiP and Silex did 3.80

While I'm relieved to see that, I wouldn't expect anything less from a trivial example such as this. I wonder if the same would be true if you hit a database, hydrated a one-to-many related series of objects from multiple tables using an ORM, modified some of them and save them etc.

However, I don't mean to harp on about the memory leaks; It's just that it seems to have been an assumed wisdom in my long association with PHP that you're going to need to restart PHP processes eventually. Perhaps the GC in 5.3 alleviates some of these concerns (including detection of circular references), as does the maturing of the code base and the commonly used libraries over the years.

The fact that you are using AiP happily in production already means that there is both a niche and a successful way to implement this so I'm glad to see new approaches like this emerging.

The setup we run with AiP is the full Midgard Create stack: Midgard2 extension, Midgard MVC framework, a lot of CMS components.

http://bergie.iki.fi/blog/introducing_the_midgard_create_use...

Hi. I am the author or AiP.

There were a lot of leaks, when I tried this idea originally, back in php 5.2 days. See http://blog.milkfarmsoft.com/2007/06/php-as-appserver-vs-mem... for example.

But 5.3 solved all of the issues. The only leaks happening these days come from "unofficial" extensions, which introduce errors on C-level. PHP code _never_ results in memory leaks these days, as far as I know.

Any reason your tests used HTTP/1.0? It is likely that most of the time is spent in connection establishment and you may not have tested the server.

how is this anything like express?

See klein (https://github.com/chriso/klein.php) for a real PHP express port..

All of them are inspired by Sinatra. But does Klein do templating, like Silex and ExpressJS both can?

Reply on news.ycombinator.com