19 Sep 2014    caspar 20140919



caspar - Makefile snippets for common tasks


In a Makefile, write either
include caspar/mk/caspar.mk
include caspar/mk/docbook.mk
include caspar/mk/pod.mk


Caspar offers Makefile snippets for common tasks, like installing (configuration) files, or typesetting LaTeX, DocBook XML and DocBook SGML documents.

The typesetting functionality is delivered by docbook.mk and pod.mk. This is documented in caspar-typesetting(7).

The installing-stuff functionality is delivered by caspar.mk. (That's what the rest of the manual will talk about.) It enables one to run 'make install' from within a tree which typically holds configuration files, managed using Subversion (or git or any other version control system, for that matter).

It is useful in cases like this: all configuration files of some host are under version control, and, after commiting a change to CVS, you want to be able to easily install the new configuration file on the host.

With caspar, all you have to do is specify the hostname in one place, and specify the name of the target directory in each CVS directory.

It is comparable with other tools for Unix system administrators like puppet and cfengine. Main difference: the caspar code consists of less than 100 lines of GNU Make.


Within a CVS tree, create a file include/install.mk, with contents like e.g.
csp_UHOST = root@some.host.somewhere include caspar/mk/caspar.mk
. Within each other directory of the CVS tree which holds files, create a Makefile, which looks like e.g.
csp_DIR = /some/dir/ectory/ include ../../include/install.mk
. If you'd like to use the install-recursive target too, in directories which hold subdirectories (but not files), you'll have to create a Makefile which looks something like
include ../../include/install.mk
. From within the CVS tree, one can call:
make <filename>-diff make <filename>-install make diff make install make load make make install-recursive
Calling make <filename>-diff shows the diff between the local file and the file as currently installed on the remote location. Calling make install (or make) now will scp all files in the current directory to the remote location. The install-recursive target descends down the tree, and calls make install in each subdirectory.

Of course, you'll have to be able to ssh directly as root to the remote host to get this working (if you'd like to use csp_UHOST = root@some.host.somewhere). If you don't like this, and would like to have a PermitRootLogin no in your /etc/ssh/sshd_config, you can use csp_sucp(1). See below.


The variables one can set in a calling Makefile are:

user@host, reachable using $(csp_PUSH) (which is scp by default)
space separated list of user@host items, reachable using $(csp_PUSH)
directory on host, reachable using function $(csp_PUSH)
make function for pushing files to remote location. Usually, this is a wrapper around a script or program. The function will be passed 4 arguments: [user@]host, remote directory and local filename. [user@]host will be set to all elements of $(csp_UHOSTS); directory will be set to $(csp_DIR). Currently, $(csp_scp_FUNC), $(csp_cp_FUNC) and $(csp_sucp_FUNC) are supported as push plugins. If csp_PUSH is unset, the default $(csp_scp_FUNC) is used.
the `load' target depends upon these targets.
the `build' target depends upon these targets.
cp binary, just "cp" by default
scp binary, just "scp" by default
script wrapping sudo in ssh, "csp_sucp" by default
extra files which should be installed. Can be used to include files starting with a dot.
files which should never be installed. Set to Makefile .%.swp %~ #%# pod2htmd.tmp pod2htmi.tmp by default.
extra files which should never be installed; added to list in csp_TABOOFILES.
files which should be installed, even if in initial csp_TABOOFILES list. Removed from csp_TABOOFILES list.
directories to exclude in install-recursive target. set to CVS .svn by default.
see csp_TABOOFILES equivalents.
override csp_UHOSTS: don't push to csp_UHOSTS, but to the intersection of this space separated list of user@host items and csp_UHOSTS.

The following variables might get phased out or removed soonish:

extra arguments to pass to cp invocation, none by default
extra arguments to pass to scp invocation, e.g. '-i .ssh/id_rsa-root'


Some examples:

Using csp_UHOST
This is the simplest way to use caspar. Makefile is

csp_UHOST = root@some.host.somewhere csp_DIR = /etc/ include caspar/mk/caspar.mk
Now, running "make" will scp all files in the current directory to root@some.host.somewhere:/etc/.

More hosts, not scp but sudo via ssh: using csp_PUSH
Makefile is

csp_UHOSTS = root@some.host.somewhere root@some.other.host csp_PUSH = $(csp_sucp_FUNC) csp_DIR = /etc/ include caspar/mk/caspar.mk
Now, running "make" will use csp_sucp(1) to install all files in the current directory to both root@some.host.somewhere:/etc/ and root@some.other.host:/etc/. If a file named fstab is present in the current directory, running "make fstab-install" will just install that one file. If you need to sudo(1) to another user on the remote host, add something like
csp_XARG = postgres
. (If such a username is not supplied, sudo (and csp_sucp) use the default: root.)

Overriding csp_UHOSTS: csp_UHOSTS_SKIP
If one or some of your hosts are temporarily unavailable, and you'd like to push your files to the hosts which are alive, you can temporarily override your csp_UHOSTS. E.g., when some.other.host is not available:

% cat Makefile csp_UHOSTS = root@some.host.somewhere root@some.other.host csp_DIR = /etc/ include caspar/mk/caspar.mk % make install csp_UHOSTS_SKIP=root@some.other.host scp hosts root@some.host.somewhere:/etc/ scp fstab root@some.host.somewhere:/etc/

Overriding csp_UHOSTS in a smart way: csp_UHOSTS_SUBSET. Using multiple groups of hosts. Recursive make made easy.
If you have lots of subdirectories holding information for lots of groups of hosts, while this run you just want to install for a small group (or 1) hosts, csp_UHOSTS_SUBSET is useful. Suppose your casparized tree looks like

Makefile apache/include/install.mk apache/etc/apache2/Makefile apache/etc/apache2/envvars php/include/install.mk php/etc/php4/apache/Makefile php/etc/php4/apache/php.ini grub/include/install.mk grub/boot/grub/Makefile grub/boot/grub/menu.lst logrotate/include/install.mk logrotate/etc/Makefile logrotate/etc/logrotate.conf nrpe/include/install.mk nrpe/debian/etc/default/Makefile nrpe/debian/etc/default/nagios-nrpe-server
The file apache/etc/apache2/Makefile is:
csp_DIR = /etc/apache2/ include ../../include/install.mk
(all other Makefiles are similar). The file apache/include/install.mk is
csp_UHOSTS = root@a root@b include caspar/mk/caspar.mk
The file php/include/install.mk is the same. The files grub/include/install.mk and logrotate/include/install.mk are
csp_UHOSTS = root@d root@e root@f root@g include caspar/mk/caspar.mk
The file nrpe/include/install.mk is
csp_UHOSTS = root@d root@e root@f root@n include caspar/mk/caspar.mk
The toplevel Makefile is
dirs = $(patsubst %/Makefile,%,$(shell find * -mindepth 1 -name Makefile)) all: for i in $(dirs); do $(MAKE) -$(MAKEFLAGS) -C $$i; done install for i in $(dirs); do $(MAKE) -$(MAKEFLAGS) -C $$i install; done load for i in $(dirs); do $(MAKE) -$(MAKEFLAGS) -C $$i load; done
(we don't feel like sticking a Makefile in all non-leaf nodes of our tree).

Now, when running "csp_UHOSTS_SUBSET='root@e root@f root@m root@n' make" in the toplevel, caspar just takes the intersection of csp_UHOSTS_SUBSET and csp_UHOSTS for each csp_UHOSTS list. So, caspar will not push anything for apache/ and php/. The files grub/boot/grub/menu.lst and logrotate/etc/logrotate.conf will get pushed to root@e and root@f only. The file nrpe/debian/etc/default/nagios-nrpe-server will get pushed to root@e, root@f and root@n.

This is often better than just overriding csp_UHOSTS on the commandline (or in your shell's environment): if the intersection of the original csp_UHOSTS and your new csp_UHOSTS is empty, chances are big you've just forgotten to clean your environment.

Creating remote directories if needed
Makefile is

csp_DIR = /some/dir/ectory/ csp_PUSH = $(csp_scpmkdir_FUNC) csp_UHOST = root@some.host.somewhere include caspar/mk/caspar.mk
Now, before calling scp, caspar will run 'mkdir -p' to create any missing remote directories.

Using csp_CP and csp_LOAD
username/etc/Makefile is

csp_UHOST = dummy csp_PUSH = $(csp_cp_FUNC) csp_DIR = $(HOME)/etc/ csp_LOAD = crontab-load include ../include/install.mk crontab-load: crontab $(csp_DIR)/crontab
while ../include/install.mk is just
include caspar/mk/caspar.mk
. Setting csp_PUSH to $(csp_cp_FUNC) causes cp(1) to get executed by "make install" (not scp(1)). Setting csp_LOAD causes "make load" to execute the crontab command. Just running "make" is OK too, since "make" calls both "make install" and "make load".

Using csp_DIR, csp_LOAD and install(1)
To install a file on the local host, create e.g. a file etc/uruk/Makefile like:

csp_UHOST = dummy csp_DIR = /etc/uruk/ csp_PUSH = $(csp_install_FUNC) csp_LOAD = uruk-load include caspar/mk/caspar.mk uruk-load: sudo invoke-rc.d uruk force-reload

Using csp_DIR and csp_LOAD, take 2
etc/Makefile is

csp_DIR = /etc/ csp_LOAD = aliases-load include ../include/install.mk aliases-load: $(csp_SSH) $(csp_UHOST) "cd /etc; postalias aliases; postfix reload"
while ../include/install.mk is
csp_UHOST = root@some.host.somewhere include caspar/mk/caspar.mk

Using csp_BUILD: building files locally
If you'd like to build some files locally from local sources, before installing the just build files, do e.g.

csp_UHOST = root@some.host.somewhere csp_DIR = /etc/ csp_TABOOFILES_ADD = sshd_config.m4 csp_BUILD = my-build include caspar/mk/caspar.mk my-build: sshd_config sshd_config: sshd_config.m4 m4 $< > $@
List all source files in csp_TABOOFILES_ADD: this way, they won't get installed on the csp_UHOST.

Using csp_sucp_FUNC and csp_LOAD
If you'd like to use csp_sucp and want a `load' target, do something like:

csp_PUSH = $(csp_sucp_FUNC) csp_UHOST = foobar.example.com csp_DIR = /etc/uruk/ csp_LOAD = rc-load include caspar/mk/caspar.mk rc-load: $(csp_SSH) $(csp_UHOST) "sudo invoke-rc.d uruk force-reload"

Adding a "check" target
If you want to do some syntax check on the remote host, before loading the just installed configuration file (and have a "make check" thing), do

csp_UHOST = foobar.example.com csp_DIR = /etc/ csp_LOAD = check my-load include caspar/mk/caspar.mk check: $(csp_SSH) $(csp_UHOST) do-check-stuff my-load: $(csp_SSH) $(csp_UHOST) do-load-stuff
This way, "make load" won't cause the file to load if the check fails (which is probably what you want). Running "make" will perform "install", "check" and "load".

Combining the csp_LOAD target with multiple hosts; building files remotely
You'll have to loop over csp_UHOSTS to execute load-command. Here's an example doing some preprocessing on the remote hosts too.

csp_DIR = /etc/ssh/ csp_UHOSTS = root@some.host.somewhere root@some.other.host csp_LOAD = sshd_config-load sshd_config-load = ssh $1 "cd $(csp_DIR); \ m4 sshd_config.m4 >sshd_config && \ PATH=$$PATH:/sbin /etc/init.d/ssh restart" include caspar/mk/caspar.mk
(Alternatively, you could explicitly specify the loop over the hosts:
csp_DIR = /etc/ssh/ sshd_config-load: for suh in $(csp_UHOSTS); do \ ssh $$suh "cd $(csp_DIR); \ m4 sshd_config.m4 > sshd_config && \ PATH=$$PATH:/sbin /etc/init.d/ssh restart"; \ done

Using the csp_TABOOFILES_{ADD,SKIP} variables; another way to perform remote builds
Using the csp_TABOOFILES_{ADD,SKIP} variables is handy if you want to install a Makefile, instead of using it: Create Makefile just as you'd like to have it installed on the remote location. Now, create GNUmakefile as e.g.

csp_TABOOFILES_SKIP = Makefile csp_TABOOFILES_ADD = GNUmakefile csp_DIR = /etc/foobar/ csp_UHOST = root@some.host.somewhere include caspar/mk/caspar.mk load: $(csp_SSH) $(csp_UHOST) "make -C $(csp_DIR)"
Now, make install and make load will do the right thing.

Using the csp_EXTRAFILES variable
Using the csp_EXTRAFILES variable is handy if you want to install files with a leading dot. E.g.:

csp_EXTRAFILES = .bashrc csp_UHOST = root@some.host.somewhere csp_DIR = include caspar/mk/caspar.mk

Overriding csp_UHOSTS
Supply e.g.

csp_UHOSTS = root@localhost root@some.host.somewhere
in install.mk, to install on multiple hosts. Run
make filename-install csp_UHOSTS=joe@otherhost
to install filename as joe@otherhost, instead of the default as given in install.mk. If you want to enable passing csp_UHOSTS as a shell environment variable, you'll have to use conditional assignment in your Makefile:
csp_UHOSTS ?= root@localhost root@some.host.somewhere
This allows it to run
% export csp_UHOSTS=foo@bar % make filename-install
to install on foo@bar.

Using sudo locally for installing files
If you'd like to install files like

sudo cp foo.rc /etc/foobar/
you could set up your Makefile as
csp_DIR = /etc/foobar/ csp_UHOST = dummy csp_PUSH = sudo cp $(1) $(3) include caspar/mk/caspar.mk
This is like csp_sucp, but without the ssh wrapping: it works on localhost only.

Plugging your own install script in caspar
If your script foobar should be called as e.g.

foobar --file=fstab --user@host=joe@some.host \ --dir=/etc/ --debuglevel=3
then make sure your Makefile features something like
csp_foobar_FUNC = foobar --file=$(1) --user@host=$(2) \ --dir=$(3) --debuglevel=$(4) csp_PUSH = $(csp_foobar_FUNC) csp_XARG = 3
You can now use csp_UHOST and csp_DIR just as you're used to.

More advanced tricks
When you don't want to ssh to root@some.host.somewhere directly, you could do

sudo rsync -az /path/to/your/config_archive /etc
on some.host.somewhere (e.g. from cron).


caspar/mk/caspar.mk, caspar/mk/docbook.mk, caspar/mk/pod.mk


For caspar.mk: csp_CP, csp_LOAD, csp_SCP, csp_UHOST, csp_PUSH, ...


Very likely, GNU Make is not the best tool for doing the stuff caspar.mk is doing. For the list of reported bugs, see http://bugs.debian.org/src:caspar. See also TODO, distributed with the caspar package. (And online at http://mdcc.cx/pub/caspar/caspar-latest/TODO.)


Caspar is named after Caspar the Friendly Ghost, since that's the title of the Daniel Johnston song I was listening to when deciding to package my homegrown scripts.


Joost van Baal-Ilić


caspar-typesetting(7) csp_helper(1)

The caspar homepage is at http://mdcc.cx/caspar/ .

The document "Versiebeheer en software-packages: Waarom en Hoe" (in Dutch) describes some of the reasons why people might want to use tools like caspar.

Jeroen Hoppenbrouwers blogs about the way he uses caspar, in "Using Subversion and Caspar to maintain a Linux host".

Lots of tools overlap (partly) with caspar in their functionality. Here's a list.

ansible (http://www.ansible.com/); code is on github (https://github.com/ansible/ansible)

cfengine (http://www.cfengine.org/), by Mark Burgess e.a., builds expert systems to administrate and configure large computer networks: it delivers a very big superset of caspar's installation mechanism.

Puppet (http://reductivelabs.com/projects/puppet), also something like a configuration management system.

Here's an older list; a big part of it was collected by Ray Miller (http://users.ox.ac.uk/~raym/) of Oxford University, and published in the article "Configuration Management with Subversion, YAML and Perl Template Toolkit" in the SANE 2006 (http://www.sane.nl/sane2006/) conference proceedings. FIXME: Check urls, update

docbookmk, by Michael Wiedmann (http://www.miwie.org/docbkmake/) offers probably a superset of Caspar's docbook.mk functionality.

latex-make by the LaTeX Utils project on http://gforge.inria.fr/projects/latex-utils/ seems to provide similar functionality as docbook.mk for LaTeX documents.

Latexmk by John Collins e.a. on http://www.phys.psu.edu/~collins/software/latexmk-jcc/ is another implementation of this idea.

SUP, the Software Upgrade Protocol and it's implementation by Carnegie Mellon University offers another way to distribute (configuration)files. Beware though: between Nov 1996 and June 2004, no new release has been published. The Debian (ftp://ftp.debian.org/debian/pool/main/s/sup/) and NetBSD packages are likely still maintained, though.

PIKT (http://www.pikt.org/) is intended primarily for system monitoring, but does do configuration management too.

LCFG (http://www.lcfg.org/) is another configuration management system.

The Arusha Project (ARK, at http://ark.sf.net/) provides a framework for collaborative system administration.

Bcfg2 (http://trac.mcs.anl.gov/projects/bcfg2/) is yet another configuration management system.

quattor (http://quattor.web.cern.ch/) is a system administration toolkit for installation, configuration and management of Unix systems.

rb3 and friends, as written and used by Ray Miller e.a. at Oxford University, (http://users.ox.ac.uk/~raym/software/configuration-management/).

The svk version control system is said to be quite usable for handling configuration file management (without a separate install mechanism like caspar). See also this discussion on the Debian development list.

On the http://www.infrastructures.org/ website on automated (Unix) system administration, you can find some thoughts on managing configuration files using a version control system.