一般现在我们配置的PHP的web环境，如LNMP（linux+Nginx+Mysql+PHP）, 这里linux可能是centos, ubuntu..., 数据库可能是mysql, postgresql, sql server等。。

在服务器上安装PHP-FPM， nginx后， 我们要配置Nginx的http模块， 让 .php的文件由nginx 转发给PHP-FPM处理，然后在将php-fpm的处理结果通过http响应传给浏览器，就完成了一次http的请求。。

在配置 Nginx 的http模块的时候， 通常是这样：

server ~ \.php$ {include snippets/fastcgi-php.conf; fastcgi_pass 127.0.0.1:9000; }

也可以这样，

server ~ \.php$ {include snippets/fastcgi-php.conf; fastcgi_pass unix:/var/run/php5-fpm.sock; }

那么这两种方式有什么区别呢？？

这就是我这篇博文所要解释的问题。下面，我带大家来分析一下其中的原理，一下是我的一些理解，不对的地方还请大家不吝赐教，我将很感激~~

PHP-FPM can listen on multiple sockets. I also listen on Unix sockets, or TCP sockets. See how this works and how to ensure Nginx is properly sending requests to PHP-FPM.

Command Rundown

Default Configuration

Edit PHP-FPM configuration

# Configure PHP-FPM default resource poolsudo vim /etc/php5/fpm/pool.d/www.conf

PHP-FPM Listen configuration:

# Stuff omittedlisten = /var/run/php5-fpm.socklisten.owner = www-datalisten.group = www-data

Also edit Nginx and see where it's sending request to PHP-FPM:

# Files: /etc/nginx/sites-available/default# ... stuff omittedserver ~ \.php$ {include snippets/fastcgi-php.conf; fastcgi_pass unix:/var/run/php5-fpm.sock; }

We can see above that Nginx is sending requests to PHP-FPM via a unix socket (faux file) at/var/run/php5-fpm.sock. This is also where thewww.conffile is setting PHP-FPM to listen for connections.

Unix Sockets

These are secure in that they are file-based and can't be read by remote servers. We can further use linux permission to set who can read and write to this socket file.

Nginx is run as user/groupwww-data. PHP-FPM's unix socket therefore needs to be readable/writable by this user.

If we change the Unix socket owner to user/groupubuntu, Nginx will then return a bad gateway error, as it can no longer communicate to the socket file. We would have to change Nginx to run as user "ubuntu" as well, or set the socket file to allow "other" (non user nor group) to be read/written to, which is insecure.

# Stuff omittedlisten = /var/run/php5-fpm.socklisten.owner = ubuntulisten.group = ubuntu

So, file permissions are the security mechanism for PHP-FPM when using a unix socket. The faux-file's user/group and it's user/group/other permissions determines what local users and processes and read and write to the PHP-FPM socket.

TCP Sockets

Setting the Listen directive to a TCP socket (ip address and port) makes PHP-FPM listen over the network rather than as a unix socket. This makes PHP-FPM able to be listened to by remote servers (or still locally over the localhost network).

Change Listen toListen 127.0.0.1:9000to make PHP-FPM listen on the localhost network. For security, we can use thelisten.allowed_clientsrather than set the owner/group of the socket.

PHP-FPM:

# Listen on localhost port 9000Listen 127.0.0.1:9000# Ensure only localhost can connect to PHP-FPMlisten.allowed_clients = 127.0.0.1

Nginx:

# Files: /etc/nginx/sites-available/default# ... stuff omittedserver ~ \.php$ {include snippets/fastcgi-php.conf; fastcgi_pass 127.0.0.1:9000; }

/pipermail/freebsd-performance/-February/001143.html

unix domain sockets vs. internet sockets

Robert Watsonrwatson at

Fri Feb 25 02:29:14 PST

Previous message:unix domain sockets vs. internet socketsNext message:unix domain sockets vs. internet socketsMessages sorted by:[ date ][ thread ][ subject ][ author ]

On Fri, 25 Feb , Baris Simsek wrote:>I am coding a daemon program. I am not sure about which type of sockets>i should use. Could you compare ip sockets and unix domain sockets? My>main criterions are performance and protocol load. What are the>differences between impelementations of them at kernel level?There are a few differences that might be of interest, in addition to thealready pointed out difference that if you start out using IP sockets, youdon't have to migrate to them later when you want inter-machineconnectivity: - UNIX domain sockets use the file system as the address name space. Thismeans you can use UNIX file permissions to control access to communicatewith them. I.e., you can limit what other processes can connect to thedaemon -- maybe one user can, but the web server can't, or the like.With IP sockets, the ability to connect to your daemon is exposed offthe current system, so additional steps may have to be taken forsecurity. On the other hand, you get network transparency. With UNIXdomain sockets, you can actually retrieve the credential of the processthat created the remote socket, and use that for access control also,which can be quite convenient on multi-user systems.- IP sockets over localhost are basically looped back network on-the-wireIP. There is intentionally "no special knowledge" of the fact that theconnection is to the same system, so no effort is made to bypass thenormal IP stack mechanisms for performance reasons. For example,transmission over TCP will always involve two context switches to get tothe remote socket, as you have to switch through the netisr, whichoccurs following the "loopback" of the packet through the syntheticloopback interface. Likewise, you get all the overhead of ACKs, TCPflow control, encapsulation/decapsulation, etc. Routing will beperformed in order to decide if the packets go to the localhost.Large sends will have to be broken down into MTU-size datagrams, whichalso adds overhead for large writes. It's really TCP, it just goes overa loopback interface by virtue of a special address, or discovering thatthe address requested is served locally rather than over an ethernet(etc). - UNIX domain sockets have explicit knowledge that they're executing onthe same system. They avoid the extra context switch through thenetisr, and a sending thread will write the stream or datagrams directlyinto the receiving socket buffer. No checksums are calculated, noheaders are inserted, no routing is performed, etc. Because they haveaccess to the remote socket buffer, they can also directly providefeedback to the sender when it is filling, or more importantly,emptying, rather than having the added overhead of explicitacknowledgement and window changes. The one piece of functionality thatUNIX domain sockets don't provide that TCP does is out-of-band data. Inpractice, this is an issue for almost noone.In general, the argument for implementing over TCP is that it gives youlocation independence and immediate portability -- you can move the clientor the daemon, update an address, and it will "just work". The socketslayer provides a reasonable abstraction of communications services, soit's not hard to write an application so that the connection/bindingportion knows about TCP and UNIX domain sockets, and all the rest justuses the socket it's given. So if you're looking for performance locally,I think UNIX domain sockets probably best meet your need. Many peoplewill code to TCP anyway because performance is often less critical, andthe network portability benefit is substantial.Right now, the UNIX domain socket code is covered by a subsystem lock; Ihave a version that used more fine-grain locking, but have not yetevaluated the performance impact of those changes. I've you're running inan SMP environment with four processors, it could be that those changesmight positively impact performance, so if you'd like the patches, let meknow. Right now they're on my schedule to start testing, but not on thepath for inclusion in FreeBSD 5.4. The primary benefit of greatergranularity would be if you had many pairs of threads/processescommunicating across processors using UNIX domain sockets, and as a resultthere was substantial contention on the UNIX domain socket subsystem lock. The patches don't increase the cost of normal send/receive operations, butdue add extra mutex operations in the listen/accept/connect/bind paths.Robert N M Watson