[gentoo-server] gentoo and portage in a large scale environment

[gentoo-server] gentoo and portage in a large scale environment

[Joe Rizzo]

I hope I am posting this to the correct mailing lists.  The
gentoo-servers and gentoo-admin mailing lists seem appropriate. 

I am looking at deploying a large scale installation of systems running
Gentoo.  (Large scale being 200-300 systems)  Currently, I am mainly
concerned with how to handle portage and packages.

My current thinking is to have a limited portage tree that only contains
ebuilds needed by the systems.  Included in this portage tree will be
ebuilds for custom software packages.  The systems will sync off of this
custom maintained portage tree.  I would like binary packages to be
available from a central repository and not have gcc on the systems.  I
would like to avoid mounting a network file system.  Am I on the best
path?

Functionally, systems will be imaged the minimal base image via
systemimager.  After that, packages will be deployed via emerge.  The
packages will be precompiled and available.  On going, packages will be
updated and deployed via emerge.

Please provide experience or ideas on:
1)  Creating and maintaining a minimal gentoo image?
2)  Managing portage and packages for a large scale gentoo environment?

Thanks,
Joe Rizzo

-- 
gentoo-server@gentoo.org mailing list

Re: [gentoo-server] gentoo and portage in a large scale environment

[Jason Stubbs]

Joe Rizzo wrote:
> I am looking at deploying a large scale installation of systems running
> Gentoo.  (Large scale being 200-300 systems)  Currently, I am mainly
> concerned with how to handle portage and packages.
> 
<snip>
 >
> Please provide experience or ideas on:
> 1)  Creating and maintaining a minimal gentoo image?

I'll leave this one to those with experience, but I'll at least offer 
the advise of "test test test!" ;)

> 2)  Managing portage and packages for a large scale gentoo environment?

Although it pains me to say it, I'd suggest not using emerge's remote 
binary package feature. Instead, you'd be much safer to rsync the binary 
packages as well and use portage with the local package repository. To 
be quite honest, that particular feature in it's current implementation 
is basically unmaintained and has several obscure bugs that hit often. 
This will be rectified in the next major portage release, but until then 
the ideas and methods you outlined would be your best bet.

As for the actual rolling out of upgrades and configuration changes, 
I've heard many people rave about cfengine. I believe it is used 
extensively throughout gentoo's own infrastructure as well. I haven't 
spent the time to learn it myself yet though so can't offer any advice 
beyond that either.

Regards,
Jason Stubbs
-- 
gentoo-server@gentoo.org mailing list

Re: [gentoo-server] gentoo and portage in a large scale environment

[Lance Albertson]

[-- Attachment #1: Type: text/plain, Size: 2051 bytes --]

Jason Stubbs wrote:
> Joe Rizzo wrote:
> 
>> I am looking at deploying a large scale installation of systems running
>> Gentoo.  (Large scale being 200-300 systems)  Currently, I am mainly
>> concerned with how to handle portage and packages.

I've been wanting to deal with that myself, but haven't had a chance to
figure out the best method to do so. I'd be interested in what you come
up with in the end that works!


> Although it pains me to say it, I'd suggest not using emerge's remote
> binary package feature. Instead, you'd be much safer to rsync the binary
> packages as well and use portage with the local package repository. To
> be quite honest, that particular feature in it's current implementation
> is basically unmaintained and has several obscure bugs that hit often.
> This will be rectified in the next major portage release, but until then
> the ideas and methods you outlined would be your best bet.

Also, if you include updates from the main tree that include movement of
packages between categories, sometimes the binary packages will get
broken because of that. I'm sure Jason can point out the problem with
that, but I've been bit by that a few times myself.

> As for the actual rolling out of upgrades and configuration changes,
> I've heard many people rave about cfengine. I believe it is used
> extensively throughout gentoo's own infrastructure as well. I haven't
> spent the time to learn it myself yet though so can't offer any advice
> beyond that either.

Yes, if you want to do simple config pushes, cfengine would be the
perfect fit. We do use it throughout all of Gentoo's infrastructure and
makes our life much easier (though, sometimes its a pain). Just make
sure you read up on their documentation (which is great), and maybe add
yourself to their help mailing list.

-- 
Lance Albertson <ramereth@gentoo.org>
Gentoo Infrastructure | Operations Manager

---
GPG Public Key:  <http://www.ramereth.net/lance.asc>
Key fingerprint: 0423 92F3 544A 1282 5AB1  4D07 416F A15D 27F4 B742

ramereth/irc.freenode.net

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-server] gentoo and portage in a large scale environment

[Daniel Schledermann (TypoConsult A/S)]

Joe Rizzo wrote:

>
>My current thinking is to have a limited portage tree that only contains
>ebuilds needed by the systems.  Included in this portage tree will be
>ebuilds for custom software packages.  The systems will sync off of this
>custom maintained portage tree.  I would like binary packages to be
>available from a central repository and not have gcc on the systems.  I
>would like to avoid mounting a network file system.  Am I on the best
>path?
>
>Functionally, systems will be imaged the minimal base image via
>systemimager.  After that, packages will be deployed via emerge.  The
>packages will be precompiled and available.  On going, packages will be
>updated and deployed via emerge.
>
>Please provide experience or ideas on:
>1)  Creating and maintaining a minimal gentoo image?
>  
>
I've done it in quite a simple way:
1. Make an install of how you would want the base-server to look like.
2. Tar it in a tar-ball called stage4-<arch>.2005.0.tar.bz2
3. Upload to internal web/ftp.
4. Install next servers with stage4 :)
Of couse this isnt very automated, but makes for a quicker install than 
anything we've used previously.

>2)  Managing portage and packages for a large scale gentoo environment?
>  
>
I administer a smaller but rapidly growing serverfarm of about a dozen 
Gentoo boxes. I distribute custom ebuilds with the gensync-tool from 
gentoolkit-dev and portage overlay. That way all my own ebuilds resides 
in a seperate place. I've been planning to create a second portage 
overlay for "stable" packages, and then only let "emerge --sync" update 
important stuff like php, apache, mysql and exim. Though I'm not a 100% 
convinced with the longevity of such a setup. I've found that cleaning 
up broken binary-packages and revdep-rebuild stuff currently is quite a 
headache, so something like this must be done.

/Daniel
-- 
gentoo-server@gentoo.org mailing list