Version 3 (modified by 12 years ago) (diff) | ,
---|
Trac with FastCGI
FastCGI interface allows Trac to remain resident much like with mod_python or mod_wsgi. It is faster than external CGI interfaces which must start a new process for each request. Additionally, it is supported by much wider variety of web servers.
Note that unlike mod_python, FastCGI supports Apache SuEXEC, i.e. run with different permissions than web server running with (mod_wsgi
supports the WSGIDaemonProcess
with user / group parameters to achieve the same effect).
Note for Windows: Trac's FastCGI does not run under Windows, as Windows does not implement Socket.fromfd
, which is used by _fcgi.py
. If you want to connect to IIS, you may want to try AJP/ISAPI.
Overview
- Simple Apache configuration
- Simple Cherokee Configuration
- Simple Lighttpd Configuration
- Simple LiteSpeed Configuration
- Simple Nginx Configuration
Simple Apache configuration
There are two FastCGI modules commonly available for Apache: mod_fastcgi
and
mod_fcgid
(preferred). The latter is more up-to-date.
The following sections focus on the FCGI specific setup, see also TracModWSGI for configuring the authentication in Apache.
Regardless of which cgi module is used, be sure the web server has executable permissions on the cgi-bin folder. While FastCGI will throw specific permissions errors, mod_fcgid will throw an ambiguous error if this has not been done. (Connection reset by peer: mod_fcgid: error reading data from FastCGI server)
Set up with mod_fastcgi
mod_fastcgi
uses FastCgiIpcDir
and FastCgiConfig
directives that should be added to an appropriate Apache configuration file:
# Enable fastcgi for .fcgi files # (If you're using a distro package for mod_fcgi, something like # this is probably already present) <IfModule mod_fastcgi.c> AddHandler fastcgi-script .fcgi FastCgiIpcDir /var/lib/apache2/fastcgi </IfModule> LoadModule fastcgi_module /usr/lib/apache2/modules/mod_fastcgi.so
Setting FastCgiIpcDir
is optional if the default is suitable. Note that the LoadModule
line must be after the IfModule
group.
Configure ScriptAlias
or similar options as described in TracCgi, but
calling trac.fcgi
instead of trac.cgi
.
Add the following to the Apache configuration file (below the FastCgiIpcDir
line) if you intend to set up the TRAC_ENV
as an overall default:
FastCgiConfig -initial-env TRAC_ENV=/path/to/env/trac
Alternatively, you can serve multiple Trac projects in a directory by adding this:
FastCgiConfig -initial-env TRAC_ENV_PARENT_DIR=/parent/dir/of/projects
Set up with mod_fcgid
Configure ScriptAlias
(see TracCgi for details), but call trac.fcgi
instead of trac.cgi
. Note that slash at the end - it is important.
ScriptAlias /trac /path/to/www/trac/cgi-bin/trac.fcgi/
To set up Trac environment for mod_fcgid
it is necessary to use
DefaultInitEnv
directive. It cannot be used in Directory
or
Location
context, so if you need to support multiple projects, try
alternative environment setup below.
DefaultInitEnv TRAC_ENV /path/to/env/trac/
alternative environment setup
A better method to specify path to Trac environment is to embed the path
into trac.fcgi
script itself. That doesn't require configuration of server
environment variables, works for both FastCgi? modules
(and for lighttpd and CGI as well):
import os os.environ['TRAC_ENV'] = "/path/to/projectenv"
or
import os os.environ['TRAC_ENV_PARENT_DIR'] = "/path/to/project/parent/dir"
With this method different projects can be supported by using different
.fcgi
scripts with different ScriptAliases
.
See this fcgid example config which uses a ScriptAlias directive with trac.fcgi with a trailing / like this:
ScriptAlias / /srv/tracsite/cgi-bin/trac.fcgi/
Simple Cherokee Configuration
The configuration on Cherokee's side is quite simple. You will only need to know that you can spawn Trac as an SCGI process. You can either start it manually, or better yet, automatically by letting Cherokee spawn the server whenever it is down. First set up an information source in cherokee-admin with a local interpreter.
Host: localhost:4433 Interpreter: /usr/bin/tracd —single-env —daemonize —protocol=scgi —hostname=localhost —port=4433 /path/to/project/
If the port was not reachable, the interpreter command would be launched. Note that, in the definition of the information source, you will have to manually launch the spawner if you use a Remote host as Information source instead of a Local interpreter.
After doing this, we will just have to create a new rule managed by the SCGI handler to access Trac. It can be created in a new virtual server, trac.example.net for instance, and will only need two rules. The default one will use the SCGI handler associated to the previously created information source. The second rule will be there to serve the few static files needed to correctly display the Trac interface. Create it as Directory rule for /common and just set it to the Static files handler and with a Document root that points to the appropriate files: $TRAC_LOCAL/htdocs/ (where $TRAC_LOCAL is a directory defined by the user or the system administrator to place local trac resources).
Note:
If the tracd process fails to start up, and cherokee displays a 503 error page, you might be missing the python-flup package.
Python-flup is a dependency which provides trac with SCGI capability. You can install it on debian based systems with:
sudo apt-get install python-flup
Simple Lighttpd Configuration
The FastCGI front-end was developed primarily for use with alternative webservers, such as lighttpd.
lighttpd is a secure, fast, compliant and very flexible web-server that has been optimized for high-performance environments. It has a very low memory footprint compared to other web servers and takes care of CPU load.
For using trac.fcgi
(prior to 0.11) / fcgi_frontend.py (0.11) with lighttpd add the following to your lighttpd.conf:
#var.fcgi_binary="/usr/bin/python /path/to/fcgi_frontend.py" # 0.11 if installed with easy_setup, it is inside the egg directory var.fcgi_binary="/path/to/cgi-bin/trac.fcgi" # 0.10 name of prior fcgi executable fastcgi.server = ("/trac" => ("trac" => ("socket" => "/tmp/trac-fastcgi.sock", "bin-path" => fcgi_binary, "check-local" => "disable", "bin-environment" => ("TRAC_ENV" => "/path/to/projenv") ) ) )
Note that you will need to add a new entry to fastcgi.server
for each separate Trac instance that you wish to run. Alternatively, you may use the TRAC_ENV_PARENT_DIR
variable instead of TRAC_ENV
as described above,
and you may set one of the two in trac.fcgi
instead of in lighttpd.conf
using bin-environment
(as in the section above on Apache configuration).
Note that lighttpd has a bug related to 'SCRIPT_NAME' and 'PATH_INFO' when the uri of fastcgi.server is '/' instead of '/trac' in this example (see #2418). This should be fixed since lighttpd 1.4.23, and you may need to add "fix-root-scriptname" => "enable"
as parameter of fastcgi.server.
For using two projects with lighttpd add the following to your lighttpd.conf
:
fastcgi.server = ("/first" => ("first" => ("socket" => "/tmp/trac-fastcgi-first.sock", "bin-path" => fcgi_binary, "check-local" => "disable", "bin-environment" => ("TRAC_ENV" => "/path/to/projenv-first") ) ), "/second" => ("second" => ("socket" => "/tmp/trac-fastcgi-second.sock", "bin-path" => fcgi_binary, "check-local" => "disable", "bin-environment" => ("TRAC_ENV" => "/path/to/projenv-second") ) ) )
Note that field values are different. If you prefer setting the environment
variables in the .fcgi
scripts, then copy/rename trac.fcgi
, e.g., to
first.fcgi
and second.fcgi
, and reference them in the above settings.
Note that the above will result in different processes in any event, even
if both are running from the same trac.fcgi
script.
Note It's very important the order on which server.modules are loaded, if mod_auth is not loaded BEFORE mod_fastcgi, then the server will fail to authenticate the user.
For authentication you should enable mod_auth in lighttpd.conf 'server.modules', select auth.backend and auth rules:
server.modules = ( ... "mod_auth", ... ) auth.backend = "htpasswd" # Separated password files for each project # See "Conditional Configuration" in # http://trac.lighttpd.net/trac/file/branches/lighttpd-merge-1.4.x/doc/configuration.txt $HTTP["url"] =~ "^/first/" { auth.backend.htpasswd.userfile = "/path/to/projenv-first/htpasswd.htaccess" } $HTTP["url"] =~ "^/second/" { auth.backend.htpasswd.userfile = "/path/to/projenv-second/htpasswd.htaccess" } # Enable auth on trac URLs, see # http://trac.lighttpd.net/trac/file/branches/lighttpd-merge-1.4.x/doc/authentication.txt auth.require = ("/first/login" => ("method" => "basic", "realm" => "First project", "require" => "valid-user" ), "/second/login" => ("method" => "basic", "realm" => "Second project", "require" => "valid-user" ) )
Note that lighttpd (I use version 1.4.3) stopped if password file doesn't exist.
Note that lighttpd doesn't support 'valid-user' in versions prior to 1.3.16.
Conditional configuration is also useful for mapping static resources, i.e. serving out images and CSS directly instead of through FastCGI:
# Aliasing functionality is needed server.modules += ("mod_alias") # Set up an alias for the static resources alias.url = ("/trac/chrome/common" => "/usr/share/trac/htdocs") # Use negative lookahead, matching all requests that ask for any resource under /trac, EXCEPT in # /trac/chrome/common, and use FastCGI for those $HTTP["url"] =~ "^/trac(?!/chrome/common)" { # Even if you have other fastcgi.server declarations for applications other than Trac, do NOT use += here fastcgi.server = ("/trac" => ("trac" => ("socket" => "/tmp/trac-fastcgi.sock", "bin-path" => fcgi_binary, "check-local" => "disable", "bin-environment" => ("TRAC_ENV" => "/path/to/projenv") ) ) ) }
The technique can be easily adapted for use with multiple projects by creating aliases for each of them, and wrapping the fastcgi.server declarations inside conditional configuration blocks. Also there is another way to handle multiple projects and it's to use TRAC_ENV_PARENT_DIR instead of TRAC_ENV and use global auth, let's see an example:
# This is for handling multiple projects alias.url = ( "/trac/" => "/path/to/trac/htdocs/" ) fastcgi.server += ("/projects" => ("trac" => ( "socket" => "/tmp/trac.sock", "bin-path" => fcgi_binary, "check-local" => "disable", "bin-environment" => ("TRAC_ENV_PARENT_DIR" => "/path/to/parent/dir/of/projects/" ) ) ) ) #And here starts the global auth configuration auth.backend = "htpasswd" auth.backend.htpasswd.userfile = "/path/to/unique/htpassword/file/trac.htpasswd" $HTTP["url"] =~ "^/projects/.*/login$" { auth.require = ("/" => ( "method" => "basic", "realm" => "trac", "require" => "valid-user" ) ) }
Changing date/time format also supported by lighttpd over environment variable LC_TIME
fastcgi.server = ("/trac" => ("trac" => ("socket" => "/tmp/trac-fastcgi.sock", "bin-path" => fcgi_binary, "check-local" => "disable", "bin-environment" => ("TRAC_ENV" => "/path/to/projenv", "LC_TIME" => "ru_RU") ) ) )
For details about languages specification see TracFaq question 2.13.
Other important information like the mapping static resources advices are useful for non-fastcgi specific installation aspects. ]
Relaunch lighttpd, and browse to http://yourhost.example.org/trac
to access Trac.
Note about running lighttpd with reduced permissions:
If nothing else helps and trac.fcgi doesn't start with lighttpd settings server.username = "www-data"
, server.groupname = "www-data"
, then in the bin-environment
section set PYTHON_EGG_CACHE
to the home directory of www-data
or some other directory accessible to this account for writing.
Simple LiteSpeed Configuration
The FastCGI front-end was developed primarily for use with alternative webservers, such as LiteSpeed.
LiteSpeed web server is an event-driven asynchronous Apache replacement designed from the ground-up to be secure, scalable, and operate with minimal resources. LiteSpeed can operate directly from an Apache config file and is targeted for business-critical environments.
- Please make sure you have first have a working install of a Trac project. Test install with “tracd” first.
- Create a Virtual Host for this setup. From now on we will refer to this vhost as TracVhost. For this tutorial we will be assuming that your trac project will be accessible via:
http://yourdomain.com/trac/
- Go “TracVhost → External Apps” tab and create a new “External Application”.
Name: MyTracFCGI Address: uds://tmp/lshttpd/mytracfcgi.sock Max Connections: 10 Environment: TRAC_ENV=/fullpathto/mytracproject/ <--- path to root folder of trac project Initial Request Timeout (secs): 30 Retry Timeout (secs): 0 Persistent Connection Yes Connection Keepalive Timeout: 30 Response Bufferring: No Auto Start: Yes Command: /usr/share/trac/cgi-bin/trac.fcgi <--- path to trac.fcgi Back Log: 50 Instances: 10
- Optional. If you need to use htpasswd based authentication. Go to “TracVhost → Security” tab and create a new security “Realm”.
DB Type: Password File Realm Name: MyTracUserDB <--- any name you wish and referenced later User DB Location: /fullpathto/htpasswd <--- path to your htpasswd file
If you don’t have a htpasswd file or don’t know how to create the entries within one, go to http://sherylcanter.com/encrypt.php, to generate the user:password combos.
- Go to “PythonVhost → Contexts” and create a new “FCGI Context”.
URI: /trac/ <--- URI path to bind to python fcgi app we created Fast CGI App: [VHost Level] MyTractFCGI <--- select the trac fcgi extapp we just created Realm: TracUserDB <--- only if (4) is set. select realm created in (4)
- Modify
/fullpathto/mytracproject/conf/trac.ini
#find/set base_rul, url, and link variables base_url = http://yourdomain.com/trac/ <--- base url to generate correct links to url = http://yourdomain.com/trac/ <--- link of project link = http://yourdomain.com/trac/ <--- link of graphic logo
- Restart LiteSpeed, “lswsctrl restart”, and access your new Trac project at:
http://yourdomain.com/trac/
Simple Nginx Configuration
Nginx is able to communicate with FastCGI processes, but can not spawn them. So you need to start FastCGI server for Trac separately.
- Nginx configuration with basic authentication handled by Nginx - confirmed to work on 0.6.32
server { listen 10.9.8.7:443; server_name trac.example; ssl on; ssl_certificate /etc/ssl/trac.example.crt; ssl_certificate_key /etc/ssl/trac.example.key; ssl_session_timeout 5m; ssl_protocols SSLv2 SSLv3 TLSv1; ssl_ciphers ALL:!ADH:!EXPORT56:RC4+RSA:+HIGH:+MEDIUM:+LOW:+SSLv2:+EXP; ssl_prefer_server_ciphers on; # (Or ``^/some/prefix/(.*)``. if ($uri ~ ^/(.*)) { set $path_info /$1; } # it makes sense to serve static resources through Nginx location /chrome/ { alias /home/trac/instance/static/htdocs/; } # You can copy this whole location to ``location [/some/prefix]/login`` # and remove the auth entries below if you want Trac to enforce # authorization where appropriate instead of needing to authenticate # for accessing the whole site. # (Or ``location /some/prefix``.) location / { auth_basic "trac realm"; auth_basic_user_file /home/trac/htpasswd; # socket address fastcgi_pass unix:/home/trac/run/instance.sock; # python - wsgi specific fastcgi_param HTTPS on; ## WSGI REQUIRED VARIABLES # WSGI application name - trac instance prefix. # (Or ``fastcgi_param SCRIPT_NAME /some/prefix``.) fastcgi_param SCRIPT_NAME ""; fastcgi_param PATH_INFO $path_info; ## WSGI NEEDED VARIABLES - trac warns about them fastcgi_param REQUEST_METHOD $request_method; fastcgi_param SERVER_NAME $server_name; fastcgi_param SERVER_PORT $server_port; fastcgi_param SERVER_PROTOCOL $server_protocol; fastcgi_param QUERY_STRING $query_string; # For Nginx authentication to work - do not forget to comment these # lines if not using Nginx for authentication fastcgi_param AUTH_USER $remote_user; fastcgi_param REMOTE_USER $remote_user; # for ip to work fastcgi_param REMOTE_ADDR $remote_addr; # For attchments to work fastcgi_param CONTENT_TYPE $content_type; fastcgi_param CONTENT_LENGTH $content_length; } }
- Modified trac.fcgi:
#!/usr/bin/env python import os sockaddr = '/home/trac/run/instance.sock' os.environ['TRAC_ENV'] = '/home/trac/instance' try: from trac.web.main import dispatch_request import trac.web._fcgi fcgiserv = trac.web._fcgi.WSGIServer(dispatch_request, bindAddress = sockaddr, umask = 7) fcgiserv.run() except SystemExit: raise except Exception, e: print 'Content-Type: text/plain\r\n\r\n', print 'Oops...' print print 'Trac detected an internal error:' print print e print import traceback import StringIO tb = StringIO.StringIO() traceback.print_exc(file=tb) print tb.getvalue()
- reload nginx and launch trac.fcgi like that:
trac@trac.example ~ $ ./trac-standalone-fcgi.py
The above assumes that:
- There is a user named 'trac' for running trac instances and keeping trac environments in its home directory.
/home/trac/instance
contains a trac environment/home/trac/htpasswd
contains authentication information/home/trac/run
is owned by the same group the nginx runs under- and if your system is Linux the
/home/trac/run
has setgid bit set (chmod g+s run
) - and patch from ticket #T7239 is applied, or you'll have to fix the socket file permissions every time
- and if your system is Linux the
Unfortunately nginx does not support variable expansion in fastcgi_pass directive. Thus it is not possible to serve multiple trac instances from one server block.
If you worry enough about security, run trac instances under separate users.
Another way to run trac as a FCGI external application is offered in ticket #T6224
See also: TracGuide, TracInstall, ModWSGI, CGI, ModPython, TracNginxRecipe