How many Gentoo boxes do you maintain?

Tony0945 · Posted: Wed Feb 17, 2021 4:24 pm Post subject:

My machines are all on a LAN so i just mount portage as a samba share. I can see that you wouldn't want to do that over the internet.
Easy way is to just tar up portage and e-mail it. assuming you have sufficient bandwidth. With all these python surprises I now tar up my portage before syncing so I can revert.

gtwrek · Posted: Wed Feb 17, 2021 4:34 pm Post subject:

eccerr0r · Posted: Wed Feb 17, 2021 5:12 pm Post subject:

As I refer to the portage tree frequently and to try to keep machines somewhat independent of each other, each machine has its own portage. Distfiles I tend to purge on most machines except for one machine that I keep all that I download. I sync to one machine's portage tree.

Not sure it helps much either way however, same amount of effort to maintain all machines overall, but it does reduce bandwidth used so it.

Now the question is whether I should share the USE flags (including portage.use, portage.accept_keywords, etc.) between all machines where I can... would it make it easier to maintain them all even if they have different world files? Probably not, but it was an idea...
_________________
Intel Core i7 2700K/Radeon R7 250/24GB DDR3/256GB SSD
What am I supposed watching?

Hu · Administrator Joined: 06 Mar 2007 Posts: 21715

eccerr0r · Posted: Wed Feb 17, 2021 6:33 pm Post subject:

I ended up thinking sharing USE flags/Accept_keywords may not save much effort overall because if I end up setting specific PYTHON_TARGETS for dependencies for software I don't use on other machines (like I don't want to have MythTV installed on all machines)...

Don't know. But yes, I was hoping that all machines would fail the same way even if they don't have the same software installed, therefore reducing the amount of effort debugging merge issues, but it'd reduce flexibility of portage which would increase other problems.
_________________
Intel Core i7 2700K/Radeon R7 250/24GB DDR3/256GB SSD
What am I supposed watching?

Tony0945 · Posted: Wed Feb 17, 2021 8:28 pm Post subject:

Regarding sharing use flags: yes and no. I have quite a lot of base use flags, usually shut off, i.e. -systemd -avahi ... that I put in make.conf and another set in package.use that are tailored to each package as I don't have all of the same packages on every machine.

I thought of putting the big list from make.conf into a custom profile, but didn't understand the wiki on how to do it nor the help I got from forum. I'm not a very geometric guy so I still don't understand "profile to the left" and "profile to the right".

figueroa · Posted: Wed Feb 17, 2021 8:53 pm Post subject:

I wouldn't share USE flags here. My server and desktop are not sufficiently alike even though they started out the same.

I emerge --sync from my server. I could share distfiles but I don't think I would gain much in bandwidth. And, if my server were to go off-line, I don't want my desktop's portage to be dependent upon it.

I do heavily use NFS and find it perfectly reliable on my LAN, and the LAN at the school. Through NFS I use the server's web server, extra storage, and it's downloads directory. These are mounted full-time. Others I can mount on-demand.
_________________
Andy Figueroa
hp pavilion hpe h8-1260t/2AB5; spinning rust x3
i7-2600 @ 3.40GHz; 16 gb; Radeon HD 7570
amd64/23.0/split-usr/desktop (stable), OpenRC, -systemd -pulseaudio -uefi

eccerr0r · Posted: Thu Feb 18, 2021 6:01 am Post subject:

Fortunately there wasn't too much interaction needed for my netbook, though there was some (like that *#@$ gcc-config CHOST issue which I can sort of ignore because it's not causing build failures though stuck with gcc-9.2).

Finally on Feb 20 at just shy of 5 days (4 days 18 hours), my Atom N270, *with distcc help* minus about 9 hours of idle time due to merge death (sci-libs/lapack having problems with ninja/cmake and dev-qt/qtmultimedia getting amd64 object file back from distcc) and an "unexpected" "power outage" (oops, compiling on battery power does not last long) it completed ~780 total packages to update.

2021-02-16T14:16:22 >>> sys-devel/gcc-9.3.0-r2: 12:51:37
2021-02-18T07:21:21 >>> dev-lang/rust-1.47.0-r2: 28:09:38
2021-02-20T04:44:34 >>> www-client/firefox-78.7.1-r1: 12:11:08

Now when do we do this again ...
_________________
Intel Core i7 2700K/Radeon R7 250/24GB DDR3/256GB SSD
What am I supposed watching?

mimosinnet · Posted: Sun Feb 21, 2021 4:22 pm Post subject:

teika · Posted: Wed Mar 03, 2021 7:05 am Post subject:

I'm not sure if Gentoo is for me, after 15 years of using Gentoo...

mouacyk · n00b Joined: 23 Mar 2009 Posts: 23

So there are no solutions yet to produce multiple binaries from the same compile, with gcc or clang? I've been searching over the years to no avail.
_________________
OCed24/7: 9900K5.0GHz//16GB4000MHzC16//1080Ti2.10GHz
OCed24/7: 1680v24.5GHz//32GB2333MHzC10//750Ti[GHz

nahsi · n00b Joined: 03 Jul 2017 Posts: 3

I maintain 5. When using ansible it doesn't matter if it is 2 or 20.

You need to invest some time and effort once in the beginning to put everything in Ansible and after that it is a breeze.

https://github.com/nahsi-homelab/ansible/tree/master/roles/base

gtwrek · Posted: Mon Oct 04, 2021 6:16 pm Post subject:

Admins at my place of employment use Ansible (for non-Gentoo boxes). I've been curious about the product, but haven't really dug in to learn much.

How well does Ansible mesh with Portage? They both seem to have many overlapping roles. I'm curious how one would setup a workflow using both...

nahsi · n00b Joined: 03 Jul 2017 Posts: 3

gtwrek · Posted: Mon Oct 04, 2021 7:01 pm Post subject:

@nahsi

Thank you very much for your detailed reply. Some things to think about there for those of us who maintain (too many) boxes...

pjp · Administrator Joined: 16 Apr 2002 Posts: 20067

EDIT: I guess I took too long to finish this before nahsi replied.

I don't know how well it meshes with portage, but ansible uses YAML config files to "do the same thing" on multiple boxes. To ensure that some software was installed, the ansible instructions would use apt, yum or (theoretically) emerge to install the software. The actual installation is handled by the package manager (at least in the limiited capacity I tested ansible).

There is a learning curve, but it seems useful for some number of systems. There's a point at which people become concerned with scalability, but I don't have experience with that situation (you could probably search for discussions about that... I believe they seem to be more "web-scale" related issues).

Other factors beyond learning curve include the audience of who is maintaining the tool / configuration. Chef, Salt and others rely more on programming (python, ruby or maybe something else). And while that may be "more powerful", YAML is more comparable to maintaining config files on a system, so it at least seems easier to get started unless you are already comfortable enough with programming in general or at least the specific language used by a given tool.

YAML itself has its own issues. There are some "idiosyncrasies" that are not intuitive and possibly not clearly defined. At least when I was evaluating it, "improper" indentation could cause non-obvious problems, which in one case, I found difficult to track down (the non-obvious part).

Due to the learning curve, I'd recommend trying to be as "generic" in approach to implementing the config files and their file system layout as possible. That way if you find that your initial solution isn't working, it is easier to redo. I don't have a succinct explanation, but I can offer an example.

You want all systems to use the same sshd_config file. You define the ansible configuration file and then deploy it to all systems. Hopefully sshd_config didn't contain anything that could prevent sshd from starting, or prevent previously working systems from reconnecting. Updating everything at once is likely not what you want to do. Now you devise a scheme to divide servers into groups. How? Web servers seems like an obvious starting point. OK, do you want all web servers to break? What about web servers in a single app? So now you might have web-app_name-odd-servers and web-app_name-even-servers so that at least half of the app_name service shouldn't break.

Suddenly that starts to look more difficult to manage than "promised" by the conceptual simplicity of configuration automation tools. Where they become useful depends on any given situation. With only a few or a handful of systems, the benefits may be minimal. Continuing with the web server example, if you have mostly unique servers, then you're going to have 1 (or likely more) configuration files to manage each "unique presence" of web server. And of course that applies to any type of server or system, not only web servers.

At some point, managing individual configuration files is only so useful, so the concept extends into playbooks. So now you have individual ansible configuration files (that instruct how to install sshd_config or some software package) and separate files that delineate the types of systems. Depending on the size of environment, that may make sense to define the types of systems in a monolithic file. Or you may want to separate that into web-servers (containing references to app_name-even, odd and their respective server names), foo-servers, etc. Don't forget that you need to maintain the "one true sshd_config" file somewhere.

Finally (at least for this example), you probably want to incorporate multiple sets of instructions into a playbook (which I can't recall if that uses more than one file per playbook). In a given playbook you might then ensure the sshd_config is installed, then make sure various software packages are installed, and maybe have some other customized solutions that do something unique to your environment. Keeping in mind that you should only alter a configuration file if it is necessary or install a piece of software if it is necessary (the term used is idempotent if you're not familiar with it).

So that's a relatively simple example / description of the learning curve and the additional workload in developing and maintaining the convenience that automated configuration tools provide. I liken it to when virtualization was new. Virtualization didn't simplify administration as at least some people thought it would. It added some conveniences and advantages, but it also increased complexity. If you have a large group of servers that are configured identically, then tools like ansible offer obvious advantages. But as soon as you have a bunch of servers providing unique services, the advantages aren't as easily achieved.

I'm not suggesting to avoid using them, only that it is important to understand what they do and what's involved by adding the complexity of relying on them.
_________________
Quis separabit? Quo animo?

Massimo B. · Posted: Thu Nov 24, 2022 8:50 am Post subject:

I manage 3 machines that is my main desktop, my remote desktop and my mobile laptop. For keeping all Gentoo installations almost identical I started How to share /etc/portage and world file.
_________________
HP ZBook Power 15.6" G8 i7-11800H|HP EliteDesk 800G1 i7-4790|HP Compaq Pro 6300 i7-3770

DavidF · Posted: Thu Nov 24, 2022 6:20 pm Post subject:

I currently have two. My desktop with openrc and laptop with systemd. Loving being back on Gentoo!

DawgG · l33t Joined: 17 Sep 2003 Posts: 866

about 30 real servers, 15 printservers and 200-300 clients in 18 locations at work.
almost all of my private stuff runs gentoo, too, altough i have to admit some boxes run devuan/debian.
_________________
DUMM KLICKT GUT.