[Setup-tool-hackers] Re: Please read and comment: Setup Toolswhite paper.
- From: Bradford Hovinen <hovinen helixcode com>
- To: Miguel de Icaza <miguel helixcode com>
- Cc: setup-tool-hackers helixcode com, chema celorio com, lauris helixcode com, lupus debian org
- Subject: [Setup-tool-hackers] Re: Please read and comment: Setup Toolswhite paper.
- Date: 17 Dec 2000 13:36:41 -0500
On 26 Nov 2000 14:55:41 -0500, Miguel de Icaza wrote:
>
> Hello guys,
>
> This is a white paper on the Helix Setup Tools, I am trying to keep
> an accurrate representation of the vision of the Helix Setup Tools.
>
> This post serves two purposes: to notify a few people on what the
> plans are for the Helix Setup Tools and what the vision is, and to
> request comments and input from the actual hackers on what is
> accurrate and what things are not.
>
> For instance, I am not sure of what the status is for the work that
> Bradford has done. Ideally in Jannuary, when we move to the new
> offices, we should try to get Hans, Arturo and Paolo over here (and
> hopefully we can get Bradford at some point).
I'd like to come in January, before the new term starts. My thought is
perhaps to ride the train up shortly before LWE, spend some time here,
and come down to NY with you for the conference. Might that work?
> Anyways, if you are on the CC list for this, it means that this
> document is of relevance to you. Please, send me comments and
> suggestions before we publish this document.
Ok. These comments are rather picky, in my usual style, so you may heed
or ignore them at your discretion.
>
> Best wishes,
> Miguel
>
> The Helix Setup Tools
>
> Miguel de Icaza (miguel@helixcode.com).
>
> (C) 2000 Helix Code, Inc.
>
> * Abstract
>
>
> The Helix Setup Tools is a set of tools aimed at making system
> administration of desktop Unix systems simple. Currently
> configuring a Unix system requires a lot of Unix knowledge,
> the Helix Setup Tools lower the barrier to setting up Unix
> systems, and address this in a consistent way across Unix systems
>
> * Features of the Helix Setup Tools
>
> * Targeted end users.
>
> * Do not impose a new configuration database on the system.
>
> * Single user interface across different Linux distributions, Unix variants.
>
> * Clear split between backend and front-end.
>
> * A consistent user interface for system administration across
> platforms.
>
> * Provides clustering support to manage groups of computers.
>
> * Time travel.
>
> * Location management.
>
> * Provides desktop services through CORBA.
>
> * History
>
> The GNOME project wants to provide a complete suite of
> applications, tools, components and libraries to allow free
> Unix systems to compete on the desktop space with systems like
> Windows and the MacOS.
>
> When the GNOME project started, we decided that we would not
> tackle the problem of writing system administration tools, as
> we expected the Unix vendor to know more about their system
> and expected them to provide the right set of tools for
> configuring and setting up their system. So GNOME would
> provide the same desktop experience across operating systems
> and we would expect the vendor tools to be integrated into
> GNOME in some way or another.
>
> As time passed a number of problems with the above approach
> arised, and they were not helping the GNOME project achieve
> its goal:
>
> * Lack of consistency across platforms: different Unix
> strains shipped with a different set of tools to
> perform system administration for no good reason.
> Some people shipped COAS, other people shipped
> LinuxConf, some people write their own system
> administration tools.
>
> * Interface inconsistency: the various tools would be
> written using different GUI toolkits which did not
> help with consistency. Some other tools would be
> targeted to have multiple front-ends and would
> choose a set of user interface elements that could
> be implemented in both a graphical version, a text
> version and maybe an HTML version.
>
> Although the idea seems sound, in practice the
> result is a poor user interface that is not as rich
> as it could as it targets the common platform,
> rather than taking advantage of each presentation
> system.
>
> * Incomplete tools: the tools would not be complete,
> so from time to time it would be necessary to open
> up a Unix terminal and type in some commands as
> root.
>
> * Tools would only run as root: some of the system
> administration tools in some systems would only run
> as root. To a non-seasoned user, this would mean
> logging out and re-logging in as root. Eventually
> the user would just choose to run always as root, as
> he does not have to deal with this problem.
>
> * Poor human interfaces: the tools would have little
> or no attention paid to the usability aspect.
> Sometimes just a bunch of widgets put together, with
> no way to help users achieve the task they are
> trying to achieve.
>
> * Application-based configuration, rather than system
> configuration: sometimes configuration tools for a
> specific package would be written, with no attention
> paid to the interaction of this component with the
> rest of the system. Why should a user need a tool
> to export directories with NFS, another for
> AppleTalk, and a third one for Samba?
A couple of these items are very similar and could be combined, or at
least grouped together. E.g. inconsistency and poor human interfaces,
and to some degree incompleteness.
Should we include something about tools that expect the configuration
files to be in an overly restrictive format, causing extreme
difficulties when they are used on files that have been manually edited?
I understand that Linuxconf has this problem, and it's something that we
specifically address by inline editing.
> * Helix Code's Setup Tools
>
> As part of Helix Code's strategy to improve the GNOME desktop
> we wanted to address the problem above, to drive GNOME into a
> position where it can be successfully used by non-Unix
> experts.
>
> While we decided to start work on the Helix Setup Tools, we
> came up with a design that would satisfiy the needs of end
> users, and at the same time would be a good tool for seasoned
> system administrators to use.
>
> Some of the constrains set on the Helix Setup Tools are:
>
> * To be a set of cross-platform tools, the same set of
> tools should work across the various Unix strains
> and have the same user interface. It does not
> matter if you are using Solaris, Slackware, Debian,
> or Red Hat: the system administration tools will be
> the same, and they will do the same things from the
> user perspective.
>
> * The tools will have to parse and analize the system
> configuration as used in traditional Unix. Then
> when changes are made, the tools should do inline
> editing of those files
>
> * There will not be a central database layered on top
> of the Unix system. The Unix configuration files
> are the master files that will be queried and
> manipulated.
>
> * Information on the system can at any time be parsed
> and analized to produce an XML-based reprensentation
Should be "representation"
> of the system state.
>
> * The functionality to parse and analize a system, as
> well as the functionality required to make changes
> to the system will be exported as well as CORBA
> interfaces that will become system services.
>
> Given that the state of the system can be at any point
> can be snapshotted and turned into an XML description of the
> system, a number of possibilities are open by layering new
> tools on top of the core infrastructure, we will get back to
> this later.
I'm not sure I'd include this last paragraph, since all it's doing is
saying that there's something else included elsewhere in the document.
The material later on should be sufficient.
> * The Helix Setup Tools structure.
>
> Helix Setup Tools is a component-based architecture. New
> components can be added to the system when new features are
> added to the system. At the time of this writing the Helix
> Setup Tools have the following modules: file system, memory
> management, name resolution, network configuration, exported
> shares management, time and date and user management.
>
> Each Setup Tool consists of two parts: a GNOME-based
> front-end, and a backend.
>
> The backend is typically written in Perl, or C. Each backend
> program has a number of well defined parts.
>
> 1. Operating System detection.
>
> 2. Parsing and analisys of the system configuration.
>
> 3. Generation of an XML-based representation of the
> state that was parsed and analized on step 2.
>
> 4. In-place modification of the system files based on
> a new configuration definition.
>
> 5. Parsing of an XML-based definition of the system.
>
> That looks more complex than it actually is. The bottom line
> is that every Setup Tool can be invoked with the --get flag
> and you would get an XML-based snapshot of the system for the
> domain that the Setup Tool administers. Or you can feed a new
> system configuration by passing the --set argument to a Setup
> Tool.
>
> For example:
>
> bash$ nameresolution-conf --get
> <?xml version='1.0' encoding='ISO-8859-1' standalone='yes'?>
> <!DOCTYPE nameresolution []>
>
> <nameresolution>
>
> <!-- Configuration starts here -->
>
> <hostname>erandi.helixcode.com</hostname>
> <workgroup>MYGROUP</workgroup>
> <description>Samba Server</description>
>
> <searchdomain>helixcode.com</searchdomain>
>
> <nameserver>140.239.238.2</nameserver>
> <nameserver>132.248.10.2</nameserver>
>
> <statichost>
> <ip>127.0.0.1</ip>
> <alias>erandi.helixcode.com</alias>
> <alias>erandi</alias>
> <alias>localhost.localdomain</alias>
> <alias>localhost</alias>
> </statichost>
>
> <!-- You shouldn't have to modify anything below this line -->
>
> <order><hosts/><bind/></order>
> <hostmatch return='all'/>
>
> <!-- You cannot modify anything below this line -->
>
> <!-- End of configuration -->
>
> </nameresolution>
> bash$
>
> The backend and the front-end communicate by exchanging these
> XML-based representations.
>
> The front-end would provide the best user interface possible
> to edit the above configuration, and then when the user is
> ready to accept the changes, invoke the backend with a new XML
> representation and pass the --set argument to it.
Inconsistency here: backend vs. front-end. Should it be hyphenated or
not?
> The front-end does not need to know anything about the
> underlying OS, unless a feature would be present or hidden,
> but besides these things, the front-end is a generic tool: it
> does not matter what is running on the backend.
I'd say: it does not matter what is running on the back end, provided
that both parts speak the same language. Something to that effect.
> * The root-manager.
>
> The root-manager is a suid application that would authenticate
> a regular user to have administration privileges and run
> applications on his behalf as root.
I think it's a bit misleading to say it's suid, because it runs through
something else that performs authentication and gets root privileges.
Saying it's suid might make some people nervous about certain security
problems that don't really exist.
Also, you're using a couple of different tenses here: "is a (should be
an) suid application" and "would authenticate". I'd change the latter to
"authenticates". Also "administration" should really be
"administrative".
> The setup tools are integrated with the GNOME control center,
> and blend with the rest of the configuration tools for the
> user's desktop. The first time the user tries to launch an
> application that requires root privileges, the root manager
> will be invoked, and will stay around to launch any other
> Setup Tools that the user activates without asking for the
> root password a second time.
Another verb tense problem, plus passive voice: "tries to launch" vs.
"will be invoked". I'd change the latter to "such and such invokes the
root manager". Also, make sure to be consistent about whether you
hyphenate "root manager".
> * The upper layers
>
> The fact that the Helix Setup Tools can create snapshots of
> the system at any point, and can change the system to match
> an XML-based definition of the system is the foundation for
> two extra features: Location manager and Time Travel.
I'd change "Location manager" to "location management" and de-capitalize
"Time Travel". Also, what about clustering?
> Neither of them have been finalized. They are both works in
> progress, an we hope to release these tools integrated with
> the rest of the GNOME desktop in the near future. Snapshots
> of these are available on demand (FIXME: we should put these
> tarballs in ftp.gnome.org and ftp.helixcode.com).
Re the FIXME: Agreed, once I have something packageable.
> * Location management
>
> Location management is an application of the Helix Setup Tools
> that is of most interest to Laptop users. For example, here
> is a typical setup, the one I am faced with:
>
> * Laptop at the Boston office: uses a static IP
> address; local DNS server; has to use a Postscript
> printer hooked up to a machine on the network; Time
> Zone set to EST; News server is news.harvard.net.
> Mounts NFS partitions from Helix network.
>
> * Laptop at Boston appartment: uses DHCP based IP
> address; DNS server and news server are my providers
> servers; Printer is an HP 4 that requires
> client-side Postscript to HP conversion, attached to
> a Mac computer.
>
> * Laptop at Mexico appartment: uses PPP for internet
> access; DNS server and news server are UNAM's
> servers; No printer defined; Timezone set to
> America/Mexico_City.
>
> * Laptop at Nuclear Sciences Research facility at
> UNAM: DNS server is local DNS server, news server is
> news.unam.mx; Default printer is Postscript
> attached to Linux machine; Timezeon set to
> America/Mexico_City.
>
> * Traveling in the US: ISP-provided PPP; No news
> system; no default printer; Timezone changes;
> Dynamic DNS setup.
>
> The idea is to present the user with a panel applet, or
> integrate into the panel a button that allows users to switch
> their location configuration by just selecting a location from
> the list of locations.
"The idea is" isn't really great diction for a formal paper. I'd suggest
rewording to have something more active, that refers explicitly to the
location manager. Also, try to avoid reusing the word `location' three
times in a single sentence.
> The location manager should allow the user to choose which
> settings change, and which settings do not change.
There shouldn't be a comma there at "... change, and which ..." I'd
expound a lot here, based on what we discussed in August. Perhaps
something along these lines:
The location manager allows the user to choose which settings vary from
location to location. When creating a new location profile, the user
specifies a location from which to derive its settings. By default, the
location manager assumes that all of the settings are identical to those
of the inherited location. The location manager then employs a type of
copy on write policy with respect to configuration changes that the user
makes in the new location.
Anyway, something along those lines. The above paragraph undoubtedly
needs some cleaning up and/or clarification.
> We want to store the differences between two configuration
> through a system like XPath (only the flat nodes are recorded
> for simplicity now, for non flat nodes, we will just consider
> its parent to be the last root node we can handle).
I don't recall our deciding to use XPath to store configuration
differences. In August, we decided on a system whereby there would be a
per-user and per-system configuration archive containing several
directories, one for each location. Each location's directory would then
contain the XML proper as well as a rollback log and a roster of all
configuration items managed by that location.
That's what I've implemented, anyway.
> * Time Travel and Undo.
>
> Each time a change is made to the system configuration, a
> filter sits between the front-end and the back-end and
> archives all the changes.
>
> Consider the case where an end user has a working system
> configuration. One day the power user niece comes home, plays
> with the computer, and figures he can `improve' the setup in
"niece" is feminine (the daugher of one's sibling, or the illegitimate
daughter of an official of the Catholic Church).
> the machine. By the time the niece is gone, the machine is no
> longer working, and your mum is lost. She has no idea what
> happened, but the machine has stopped working.
>
> Time travel allows the user to restore the configuration to
> the state it was on a given date. The archiving system can be
> asked to retrieve the configuration that was in effect at any
> time in the past since the installation of the Helix Setup
> Tools and restore that configuration.
>
> The Undo system is layered on top of this, given that we have
> entire snapshots of the changes that were done to the system,
> we can move back and forth between the various changes made on
> the system.
>
> * Clustering
>
> We are looking at clustering from the point of view of
> managing a set of computers: performing the same kind of
> configuration and tuning that would be done in a
> machine-per-machine basis.
I would avoid personal pronouns like we or I. Try rewording to refer to
the design of our system.
> Typically network administrators would need a set of changes
> to be applied to a set of computers in a network. The Helix
I'd change this sentence to something perhaps a bit more concrete, like
"Consider the case where a network administrator is managing a computer
lab with several dozen workstations and needs a configuration change
made to each one." You'd then want to change the rest of the paragraph
to refer to this hypothetical system administrator.
> Setup Tools will allow you to create groups of computers and
> allow you to apply changes to sets of computers or sets of
> sets (for example: changing the DNS server for all computers
> in the company; Or changing the print server for the
> engineering department; Or changing the NFS server for both
> marketing and engineering; Or changing the network
> configuration for the entire company to use DHCP).
>
> Clustering support is an ongoing development at Helix. The
> current plan is to allow the user to define a "login" method
> in a per-computer, or in a per-group basis (OpenSSH-based
> login is our current approach).
Regarding the above paragraphs: it's best to avoid enclosing things in
parantheses.
> Then the central manager would be able to fetch the
> information for all the cluster, allow the administrator to
> make changes, and have those changes be applied to all the
> machines on the network. This is done regardless of the kind
> of operating system being run, remember that the Helix Setup
> Tools do not care about the underlying operating system.
>
> This code is still a work in progress.
This is a repeat of the above "Clustering support is an ongoing
development"
> * Implementation details.
>
> * Error handling, codes being passed back and forth.
> * XML based encapsulation of data.
> * The archiving setup.
> * Integration with the gnome control center.
>
> * The current components
>
> FIXME: list the features in each one of these.
>
> memory:
> networking:
> time:
> disks:
> nameresolution:
> shares:
> users:
>
> * The components planned:
>
> Mail configuration.
> Web server configuration.
> Printer configuration/Printer spooler.
>
> * CORBA system services.
>
> The following text was taken from my `How to fix Unix' paper:
>
> * Background
>
> Configuring a Unix system traditionally has involved editing
> a system file, typically on /etc and saving the file. Then
> either the various programs and processes on the system pick
> up the changes, or you have to notify the daemons that require
> to be informed of this change.
>
> This is not really an ideal situation, because those files are
> easily corrupted by human error. And depending on the tool,
> you need a different way of editing the file.
>
> For example, consider /etc/passwd, to edit this file you need
> to use the vipw command, which will do some basic checking on
> the integrity of the file before you save it (as well as
> locking), but there are no equivalent commands to "lock",
> edit, and verify other system files.
>
> So depending on each tool, a user/administrator needs to:
>
> 1. Read the man page, and figure out what the correct
> procedure is for a specific piece to be edited.
> This is obviously not always documented.
>
> 2. Edit the file.
>
> 3. Now, depending on the kind of file you edited, you
> need to figure out how to notify the proper service
> that a change has happened.
>
> 4. Depending on the service, either hope that the
> settings are properly reloaded, or that some error
> is going to creep in, or that something will be
> said in some of the system logs.
>
> For the system logs case, examine /etc/syslog.conf
> to locate the place where error messages are
> logged. Then make sure you tail -f the file,
> restart the service, scan for errors.
>
> Hopefully you do not have a busy web server, or the
> error is going to be lost in the noise (or
> hopefully you edited /etc/syslog.conf, restarted
> syslog)
>
> * What we want to do
>
> Although the above `Background' description was aimed towards
> moving applications into providing CORBA interfaces to their
> functionality, and making new applications have a CORBA-based
> control facility, this does not solve the problem of fixing
> the current Unix.
>
> Although existing tools could be modified, a stop gap measure
> would be to reuse the Helix Setup Tools. Given that the Helix
> Setup Tools have already gone trough a great lenght to isolate
Should be `length'
> the different across Unix systems, and given that they can
Should be `differences'
> analize, and modify existing system settings, we think that
There shouldn't be a comma between `analize' and `and modify'.
> the backends are perfect for encapsulating the configuration
> issues of a Unix system.
>
> Each one of the Setup Tool backends will also be installed as
> a system service (only while running as root) that would be
> activated on demand to expose the system management functions
> trough CORBA interfaces.
>
> So the objective here is to add a few changes to the existing
> backends (this will be done by Helix CORBA::Perl team) to
> expose the CORBA interfaces for querying and making changes to
> the system, using the functions that are already defined in
> the code.
>
> A programmer could write code like this in Perl:
>
> use Bonobo::bonobo;
>
> $interfaces = BONOBO::bonobo::resolveReference ("interfaces");
> $name_resolution = $interfaces.getObject ("IDL:Helix/NameResolution:1.0");
> print "Hostname is: ", $name_resolution->getHostname (), "\n";
> print "Changing hostname to JollyJumper\n";
> $name_resolution->setHostname ("JollyJumper");
>
> * The Future
>
> It is possible to write other front-ends to the Helix Setup
> Tools. A web-based front-end is possible, as well as a
> text-based front end. We would like to see those happen at
> some point.
>
> Helix Code has also been working on Red Carpet, a
> cross-platform package management tool, and update tool that
> will bring the same uniformity to all systems. Red Carpet
> works currently on RPM-based systems and Debian-based
> systems. We will be extending it to support the Solaris, IRIX
> and HP-UX native package formats as well in the future.
>
> Package management is essential to maintain a system up to
> date, and to allow people to easily install software. Red
> Carpet and the Helix Setup Tools will also be integrated to
> deliver a complete, perfectly blended set of services.
>
> The tricky part is using Red Carpet for doing cluster
> management.
>
> * The Hackers
>
> The Helix Setup Tools were implmented by Hans Peter Jansson,
> Arturo Espinosa, Bradford Hovinen and Tambet Ingo. I am their
> project manager.
--
-Bradford Hovinen
"If the fool would persist in his folly he would become wise."
- William Blake, "Proverbs of Hell," 1793
_______________________________________________
Setup-tool-hackers maillist - Setup-tool-hackers@helixcode.com
http://lists.helixcode.com/mailman/listinfo/setup-tool-hackers
[
Date Prev][
Date Next] [
Thread Prev][
Thread Next]
[
Thread Index]
[
Date Index]
[
Author Index]
