Re: How to consistently install a set of packages?
On 06/06/2013, Andrei POPESCU <andreimpopescu@gmail.com> wrote:
> On Jo, 06 iun 13, 14:39:22, David wrote:
>>
>> Because some packages create users and groups, I want to be sure that
>> these packages are installed in the same sequence on every PC so that
>> numeric uid & gid in /etc/passwd and /etc/group end up the same on all
>> PCs. This will make admin easier when keeping the shared data in sync
>> if the numeric uid and gid are the same in every OS.
>
> For my curiosity, could you please provide an example where *system*
> users need to be synchronized between different machines?
Hi Andrei
Thank you for taking an interest in my question.
What I do with my 3 machines is apparently a little out of the ordinary
because it does not follow the typical client server model everyone is
used to, but it does work for me very effectively. Instead, any single
machine of mine is capable of taking the role of any other, to provide
redundancy.
Currently there is one desktop and two notebooks on a home LAN
with a ADSL modem/router. There is only a single user, me. I do not
serve anything to the internet. The notebooks are clones of one
another except for unique hostnames and static ips, so we can ignore
one of them. They are duplicates to provide redundancy and multiple
independent screens at my workstation, for example monitoring or
low-power logging of IRC. The desktop is the master copy of
everything, the notebooks are very similar to it, having the same
installed apps, config, and shared data.
Each machine has multiple OS root partitions managed by grub1,
each with /usr and /var. There is a different home dir for each OS with
its own dotfiles (lets say that to keep this story simpler), but most of
my work data files are bindmounted under it from a common data
partition.
Also some apps are configured to use this common data partition,
for example the approx cache. This common data partition is also
manually synced by either rsync or unison across all machines.
This is all working nicely for me for several years. I like the redundancy
and flexibility. However I need to update the base OS so I'm moving to
try wheezy with lxde on one partition, then I will try sid on another.
Now the example you requested, lets consider 'approx'. I want approx
on every OS partition on every machine to use the same numeric
uid/gid, because the approx cache is on the synced data partition.
So across all machines there is only ever one approx process running,
but it can easily be started on whichever machine is most convenient,
after the shared data partition is manually synced.
What I have said regarding approx would also apply to other servers.
For example, I might want to play with a http server, that I could run on
any machine, but it would have data available from the synced data
partition. Another possible candidate would be dnsmasq.
This is just for convenience, and definitely not for concurrent use,
because the sync is manual. For example, I prefer to work on the desktop,
but if I have to go somewhere, I just unison all the data onto one of the
notebooks, go somewhere and work happily , then unison again to the
desktop master when I return.
Or, if the desktop is down for maintenance, one of the notebooks
easily takes any LAN server roles.
The above is a simplified explanation however I hope it conveys the
essence of what I do and presents a use case. If I put all this in my
initial email it would have been too long, no-one would have read it!
Reply to: