Re: Managing a network of Debian machines?
Nelson Minar ([EMAIL PROTECTED]) wrote: : What's the right solution? Assume disk is cheap, bandwidth is fairly : cheap, but sysadmin time is really expensive. If Debian could provide : some solution, it would be a big help to Linux administrators. The solution running here is: 1) /, /etc, /var in one partition locally on each machine. Configuration is still too machine dependent. 2) /usr mounted via NFS. The disk really hosting the /usr files is a DEC-Alpha but each Linux machine could serve as well. 3) /usr/local is a symbolic link to /local on each machine. Here, machine specific applications are stored. They should not be touched by Debian packages. Due to the strict separation between configuration files and programs in Debian, this works quite well. All programs except for those installed in /usr/local have to be maintained only centrally. The netload due to the program loading on execution is still relatively small compared to the load produced by the real data which are also often treated via NFSi here. Hope this helps. Best regards -- Volker Volker Ossenkopf, E-mail: [EMAIL PROTECTED] Astrophysikalisches Institut und\\/// Universitaets-Sternwarte Jena ( . . ) Tel.: 03641/630324 --oOo--(_)--oOo- -- TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word unsubscribe to [EMAIL PROTECTED] . Trouble? e-mail to [EMAIL PROTECTED]
Re: Managing a network of Debian machines?
On Sat, 23 Nov 1996, Nelson Minar wrote: I've been a user of RedHat for the last year and a half. RedHat in general is a nice distribution, but the only reason I really use it is for RPM, the package manager. One thing that RPM cannot really help with is managing a whole network of workstations. Say I have ten Linux machines with a package manager I want all ten to stay synchronized, to have the same version of all packages. How do I do this? Something I've been thinking of doing is mounting an nfs filesystem which contains the packages to install on each client. This filesystem could be either common for all clients, for some clients or individual to each client. On each client there should be a script started with crontab that ran dpkg on all packages on that nfs mounted directory. All you have to do is to copy the packages you want installed to this directory on the master server. One problem is how to handle dependencies on each client. You must have the possibility to specify that some packages should be installed before others. And I do agree with you; managing a small network with debian machines is doable by mounting a cdrom from one machine to all the others and then run dpkg with rsh. But with a larger network this will become tedious. Cheers, Bengt-Ove Johansson! -- This message was delayed because the list mail delivery agent was down.
Re: Managing a network of Debian machines?
[EMAIL PROTECTED] (Nelson Minar) writes: One solution would be to automate the package updates This is pretty easy to do with dpkg. The two important commands are dpkg --get-selections [pattern ...] get list of selections to stdout dpkg --set-selections set package selections from stdin Presumably if you've got a room full of machines running Debian, you're willing to mirror ftp.debian.org. You install the packages you want on one machine, get the selections to a file, and set the other machines' selections with that file. Then you just do dpkg -iGROEB [1] on your mirror on each machine and they'll all get installed to match the master machine. One sticky point - configuration of packages - which you'll still have to do on each machine. Packages in the unstable snapshot are going to start using a neat tool called cfgtool. Among its capabilities is the ability to get and set configuration information in the same manner that dpkg can get and set selections. That'll be in Debian 1.3; a stable release of that is 3 - 4 months away. The other solution, one I sort of like, is to NFS mount as much as [...] but entails quite a big network cost. Quick calculation: 100 Mbit/sec ethernet, ~50% efficiency, about 5MByte/sec. You said only ten machines which leaves about 500KByte/sec bandwidth/machine. Bearable, but if the lab gets 2x or 3x bigger, it'll be unusable. [1] Equivalent to dpkg --install --refuse-downgrade --recursive --selected-only --skip-same-version --auto-deconfigure, which probably makes sense. Guy -- TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word unsubscribe to [EMAIL PROTECTED] . Trouble? e-mail to [EMAIL PROTECTED]
Re: Managing a network of Debian machines?
Problems: some packages need hand editing of some config files in /etc. This could be handled by cfengine, which can be run by the same cron job after dftp. Another problem is that I *think* dftp can only do ftp. This is a nuisance when your upgrade center doesn't have anonymous ftp. If this is really the case, I think we should ask the author of dftp to include support to use a directory instead of ftp only. 'cfengine' actually comes with it's own cron.daily script that also supports config files stored using RCS or CVS. Running 'dftp' from that same script will cause problems since it eventually calls 'dpkg' to do the work and that can require user input. 'dftp' will run off any mounted filesystem as well as over FTP. Brian ( [EMAIL PROTECTED] ) --- In theory, theory and practice are the same. In practice, they're not. -- TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word unsubscribe to [EMAIL PROTECTED] . Trouble? e-mail to [EMAIL PROTECTED]
Re: Managing a network of Debian machines?
Nelson Minar ([EMAIL PROTECTED]) wrote: : run the package upgrade command on every machine. : : One solution would be to automate the package updates, run a cron job : on all machines that keep them in sync with some master list of : package versions. This isn't very efficient, but would be acceptable. : Does someone have such a script for Debian? : Use rdist. It's a tool to syncronise filesystems of different systems. Updating only one master system and setting up a cron job to update the other systems is quite easy. You can specify files (and groups of files) that should not get updated (e.g. network initialization and ip address) It works even over the secure ssh. Andreas -- Andreas Tack Real-time-Voice-Connect: +49 (0)721 608 2326 [EMAIL PROTECTED] FAX: +49 (0)721 607102 http://ttk1.ciw.uni-karlsruhe.de/mitarbeiter/ib09/index.html -- TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word unsubscribe to [EMAIL PROTECTED] . Trouble? e-mail to [EMAIL PROTECTED]
Re: Managing a network of Debian machines?
There's no solution at the moment :-( :-( :-( I have the same problem. The situation is even worse when the machines are slightly different, and have a few different packages :-( I have a suggestion, that I've not yet tried but I'll do soon. dftp can make a list of packages that need upgrading. You run it in one machine and download the upgrades in a specific directory from an ftp site, and install them. The other machines will have dftp started by cron. Configure it to look in that upgrade directory, so the new packages will be automatically installed. This method can work even with different packages in different machines. Problems: some packages need hand editing of some config files in /etc. This could be handled by cfengine, which can be run by the same cron job after dftp. Another problem is that I *think* dftp can only do ftp. This is a nuisance when your upgrade center doesn't have anonymous ftp. If this is really the case, I think we should ask the author of dftp to include support to use a directory instead of ftp only. Carlos -- TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word unsubscribe to [EMAIL PROTECTED] . Trouble? e-mail to [EMAIL PROTECTED]
Managing a network of Debian machines?
I've been a user of RedHat for the last year and a half. RedHat in general is a nice distribution, but the only reason I really use it is for RPM, the package manager. One thing that RPM cannot really help with is managing a whole network of workstations. Say I have ten Linux machines with a package manager I want all ten to stay synchronized, to have the same version of all packages. How do I do this? In a traditional Unix installation one way to do this is to only install the vendor's default Unix on each hard drive. Since commercial Unices don't update frequently, this means that you have to sync each machine every year or so. Then all the packages you *really* care about (say, emacs or perl) get installed in an NFS mounted /usr/local. NFS takes care of the synchronization, since there's only one copy. But that doesn't work for RedHat, and I imagine doesn't work for Debian. If I upgrade emacs, for instance, it's going to upgrade in /usr/bin, a directory traditionally not NFS mounted. I would have to run the package upgrade command on every machine. One solution would be to automate the package updates, run a cron job on all machines that keep them in sync with some master list of package versions. This isn't very efficient, but would be acceptable. Does someone have such a script for Debian? The other solution, one I sort of like, is to NFS mount as much as possible on all the machines but one. Clients NFS mount /usr (maybe even the whole root disk) from a central server that is maintained by hand. This duplicates the /usr/local NFS setup of typical machines, but entails quite a big network cost. What's the right solution? Assume disk is cheap, bandwidth is fairly cheap, but sysadmin time is really expensive. If Debian could provide some solution, it would be a big help to Linux administrators. PS: the web archive of these mailing lists on www.debian.org isn't working right. Messages aren't being split correctly. -- TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word unsubscribe to [EMAIL PROTECTED] . Trouble? e-mail to [EMAIL PROTECTED]
