Installation instructions for distcc -*- indented-text -*- distcc is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version. distcc comes with ABSOLUTELY NO WARRANTY, for details see the licence. Please report any problems to distcc@lists.samba.com. QUICK SUMMARY ============= 1. Build and install ./autogen.sh # If "configure" does not already exist. ./configure make make check # Optional! Should have python >= 2.4 installed. make install # You may need to use "sudo" for this command. make installcheck # Optional! Should have python >= 2.4 installed. Repeat installation for each server machine. 2. Run distccd on each server machine, with --allow options to limit access. 3. On your client machine, set DISTCC_POTENTIAL_HOSTS to the list of distcc server machines: export DISTCC_POTENTIAL_HOSTS='localhost red green blue' 4. Build your software using distcc: cd ~/my_sources/my_project pump make -j40 CC="distcc gcc" DETAILED INSTRUCTIONS ===================== Prerequisites ------------- To build distcc you need GNU Make A C compiler It is also highly desireable to have Python >=2.4 To use "pump" mode, or to run the full test suite with "make check", or to run the benchmarks. You can optionally have Python >=2.2 To run a partial test suite with "make check" (it leaves out the pump-mode functionality, which requires Python 2.4). libpopt If this is not found on your system, the popt version that is part of the distcc distribution will be statically linked in. linuxdoc SGML tools To rebuild the documentation from its SGML source. autoconf >=2.5 To rebuild the configure scripts if you edit configure.ac. To build the optional GNOME monitor (--with-gnome), you need the GNOME2 development libraries, and in particular gtk+ >=2.0 libgnome >=2.0 libgnomeui >= 2.0 libglade >= 2.0 libpango The monitor can also be built without GNOME desktop integration (--with-gtk), in which case you need only gtk+ >=2.0 libglade >= 2.0 Preliminaries ------------- distcc can be installed and used without requiring root access. Adjust directories appropriately when installing. Configuring distcc ------------------ Note that the default GNU "sysconfdir" is /usr/local/etc. You may want to change this to /etc. $ ./configure --sysconfdir=/etc You can set an installation prefix if you want to put distcc in /opt: $ ./configure --prefix=/opt/distcc/ distcc needs to be installed on all client and server machines. If you would like a graphical client-side monitor to show running jobs, you can choose either --with-gnome or --with-gtk. If you would like to try running distcc over IPv6, use --enable-rfc2553. You must have a reasonably recent operating system or this is likely to fail in complex ways. Building distcc --------------- The simple way to build distcc is to just do ./autogen.sh # if "configure" does not already exist in this directory ./configure && make but the recommended way is to build in a different directory tree than the source tree: mkdir obj cd obj ../autogen.sh ../configure && make You can optionally run "make check" afterwards to verify that everything built OK. Installing distcc ----------------- If your system supports RPM packages, you can build RPMs with "make rpm" and then install them with "rpm -i packaging/*.rpm". If your system supports Debian packages, you can build them with "make deb" and then install them with "make install-deb". Otherwise, you can use the regular "make install". But it is preferable to install via an RPM or Debian package if possible, because those methods will do a bit more of the setup work. You can optionally run "make installcheck" afterwards to verify that was installed OK. This works regardless of which installation method you used. Starting the daemon ------------------- This stage is only required if you want to run distcc over TCP sockets, rather than SSH connections. TCP is faster but less secure and should only be used on trusted networks. In TCP mode distccd can run either from inetd or as a standalone daemon. Running standalone is recommended. If you installed via the debian or RPM package, then the daemon will be installed as a service, running as a standalone daemon, and it should get started up automatically, so you can skip the rest of this section "Starting the daemon". The rest of this section only applies if you installed via "make install". To run standalone, run a command like this, either from the command line or from the system startup scripts. distccd --daemon If the daemon is started from an rc script, then make sure that it sees a PATH setting which can find any compilers installed in nonstandard directories. You should create a "distcc" account on server machines so that distcc can run with minimal privilege. It is not necessary for this account to own any files or have a home directory. If this account doesn't exist, distccd uses the "nobody" account. By default distccd writes messages to the "daemon" syslog, which typically ends up in /var/log/messages or /var/log/daemon. You can set IP-based access control using the --allow and --listen options, in either inetd or daemon mode: distccd --allow 10.4.20.0/24 distccd does not need to run on machines that will only act as clients. See the manual for more information. Editing the daemon configuration files -------------------------------------- If you installed via the RPM or Debian package, then you will have an /etc/init.d/distcc script which reads several configuration files to decide if and how to invoke distccd. These configuration files are /etc/default/distcc /etc/distcc/clients.allow /etc/distcc/commands.allow.sh If you installed via the RPM or Debian package, then you should edit those configuration files, especially the clients.allow file. If you installed via "make install", then those configuration files will be installed, but they will NOT be used, unless you manually install the init.d/distcc script from packaging/RedHat/init.d/distcc and tailor it for your system. If you start distccd manually, rather than via the init.d/distcc script, then distccd won't use any configuration files; the allowed client IP addresses will be determined by the --allow options that you pass to distccd and the allowed commands will be determined by the DISTCC_CMDLIST environment variable; see the distccd(1) man page for details. See also the doc/example directory and its README file, which has examples of the configuration files that you need. Setting up the host list ------------------------ On the client machines, store a list of servers names in ~/.distcc/hosts or /etc/hosts, or in the DISTCC_HOSTS environment variable. If you're using TCP connections, it should look like this: localhost red green blue For SSH connections localhost @red @green @blue The hosts should be listed in descending order of speed. localhost should normally be first, unless it is signficantly slower than another machine. If you many hosts (say ten or more), it's probably better to leave localhost out of the list. The host list also needs ",cpp,lzo" after each host if you're using pump mode - see README.pump for details. Another alternative if you're using pump mode is to set the DISTCC_POTENTIAL_HOSTS environment variable; in that case, the pump script will use "lsdistcc" to set DISTCC_HOSTS, automatically eliminating hosts that are down or inaccessible or that don't have distccd running. See the distcc(1) manual and the pump(1) manual for more information. Using distcc ------------ Distcc will only improve performance if your build is parallelized. So you need to use the "-j" option to make, or its equivalent with your build tools. If your build contains too many sequential steps, e.g. if your Makefile contains all: for subdir in $(SUBDIRS); do make -C $$subdir all; done then you may need to rewrite your Makefile to get better parallelism. This is especially important if you're using pump mode. Using pump mode --------------- To use pump mode, invoke your build command via the "pump" script, e.g. export DISTCC_POTENTIAL_HOSTS="localhost red blue green" cd ~/my_sources/my_project pump make -j40 CC="distcc gcc" my_target Pump mode assumes that source files are not modified during the build. If this assumption does not hold, you should not use pump mode. You can disable pump mode by not using the pump script. In that case, you need to set DISTCC_HOSTS rather than DISTCC_POTENTIAL_HOSTS. export DISTCC_HOSTS="localhost red blue green" cd ~/my_sources/my_project make -j40 CC="distcc gcc" my_target Creating the masquerade directories ----------------------------------- The easiest way to use distcc is in "masquerade" mode, where it is installed on the PATH to "catch" calls to the compiler and redirect them over the network. Other options are discussed in the manual. For instance, you could create the directory named /usr/lib/distcc and populate it with links. # mkdir /usr/lib/distcc # cd /usr/lib/distcc # ln -s /usr/bin/distcc gcc # ln -s /usr/bin/distcc cc # ln -s /usr/bin/distcc g++ # ln -s /usr/bin/distcc c++ If you installed via the RedHat or Debian package, then this masquerade directory is already set up automatically in /usr/lib/distcc. Do this for all compiler names that you use. Then, to use distcc, a user just needs to put the directory /usr/lib/distcc/bin early in the PATH and distcc will handle the rest. export PATH=/usr/lib/distcc/bin:$PATH Use with ccache --------------- The best way to use is to set up a similar masquerade directory for ccache and put it on the PATH before distcc. NOTE: This use of ccache is incompatible with use of distcc's "pump" mode. (If you're using "pump" mode, it might be possible to use ccache on the distcc server machines. But we haven't tried that setup.) Complete the survey ------------------- Once you have distcc working for your own application, please complete and mail in the survey in survey.txt.