if there is duplicate pkgs provided by two separate pkg repos, how does pkg prioritise?

which will it pick?

assume both packages are equal?

I assume otherwise its the latest version?

polarian: there is a priority variable in the /etc/pkg/*.conf area so for each entry, if you set priority: <some number> the higher the number it sets that as "default"

so if you have 3 repos, repo 1 - priority: 100 repo 2 - priority: 50 3 - priority; 2 5

it will ook at the 100 one first and pull from there

man pkg.conf goes into a little more detail on it for you, which was my reference to get the answer :)

another angle of the same question is under man pkg-repository

ivy remember the ordering of rules in pf.conf, last match wins unless 'quick' is used. So if you have 'pass ....' then a later 'block ...' that matches will win

yelnah: yes. the issue wasn't related to that.

ek, hmjsp: wireguard doesn't have a concept of server and clients, only peers and they can't choose an IP address themself, it has to match on both sides of the tunnel, else routing won't work

voy4g3r2: thanks, sorry I didn't RTFM

polarian: its all good.. i recently went through a similar situation.. while i was building llama.cpp it was picking ports and not what iw as building.. it was about priority :)

What card are you using for llama.cpp?

nvidia gtx 3050 w/8 gig

This is on FreeBSD? How do you find the lack of CUDA support?

works well for the inference testing and agent playing

Huh, good to know

Best I have done is run claude code in a jail...

I'm gonna give it a try. I did GPU passthru with a Linux VM but such a horrible experience that I abandoned it. I don't want to sysadmin a Linux box.

Shirkdog: yeah that is not the same

voy4g3r2 Which driver are you using?

I'm going to give nvidia-drm-kmod a shot.

if this works well I owe you a coffee :)

voy4g3r2: 100% I am aware :) I had spent time trying to get pass-thru to bhyve for nvidia but just moved on.

Wise choice lol

I had a 32GB RAM windows laptop with an RTX 3060 and still was not impressed with running models locally...but will look at this again.

I'm in a similar boat. I have an RTX 3060 SUPER with 16GB of VRAM. I had to jump hoops to get passthru working, including a custom patchset to base, then when it worked, the LLM performance was terrible. Not just speed, but its ability to reason. Anything better was out of reach. But curious to see how it works on FreeBSD native this time.

I mean, that sounds consistent with my understanding; local models are mostly a joke

hi

I believe there is a way to pair things down to be very useful...but in the same way that you could process network traffic if you tune a driver to only look at the information you care about.

o0x1eef and Shirkdog: Yes, models are just models and a dance.. the harness is helpful and focusing on what you want to achieve.. big models will NOT work with either of those vram sizes.. but seeing how you can iterate without burning tokens.

and that would be a fun use case for FreeBSD

i mean, idk, you might do better to write your AI tools to run entirely on the CPU and main RAM, slower though it will be

I didn't even remember the correct card :) It's - nvidia0: <NVIDIA GeForce RTX 4070 SUPER> on vgapci0

setting up a new server, after hitting "Install" in the guided zfs-on-root menu, it hung for about 8 minutes then finally just dropped me to the login prompt. Logged in as root, ran zpool list and nothing was created, anyone know if this is an issue?

zfs-on-root setup should create a pool "zroot" which should get listed by zpool list, so no idea what happend for you, what does mount say?

XD same, didn't check mount - it looked like the installer shell as nothing was configured. reinstalling now to see (this happened with pkgbase if it makes a difference).

hm, how did you login as root? the installer shell wouldn't have the root password you choose in the installer

correct, root & no password

so far same thing, hanging, waiting for it to drop me to the shell..

just checking, you did choose some disk for the zpool?

hah yes

when it hits shell again, I'm going to try with the "Distribution Sets" option - assuming there will be an upgrade path to pkgbase for 16. My other machine on 14.3 I've refrained from upgrading because the path from 14.3 to pkgbase was hairy last I checked

interesting.. worked this time around

that's cuz you're insulated

Path to pkgbase is easy _if_ you upgrade to 15 first

Pkgbase at 14.3 and then upgrade to 15 is less easy

word thanks

Hi guys, I’m a bit confused about the FreeBSD dump command when using it to back up an entire filesystem to an external SSD.

Do I understand correctly that there is a limitation or quirk that effectively prevents backing up directories like /usr and /home to an external drive as part of the same backup?

On my ssd disk with freebsd, i have placed everything in one partition.

i feel like now there are better ways to back up than dump

but if you have a single partition / then you only need to run dump once

if you had a classic, / /usr /var you would need to run the command dump a total of three times

Ok, I use Restic for backups on my Debian system.

I was reading this thread: forums.freebsd.org/threads/questions-about-backup-with-dump.62698 and saw people saying that dump can be quite slow, so both backup and restore may take hours.

I’m a bit confused now. I thought FreeBSD was supposed to come with “batteries included,” but dump doesn’t look very practically usable to me. Am I missing something here? ;-)

It looks to me that backups are incredibly important, and so I would expect that BSD would have included goo solution for them in the base system, am I missing something?

ZFS

Yeah, I know ZFS supports snapshots, but that is not really what I mean.

I need backups on external drives, because the laptop can be lost, stolen, or the internal disk can fail. In that situation, ZFS snapshots alone do not help much, at least as I understand it.

So I would expect there to be one clear, recommended way to make a proper backup to an external drive. But instead, I’m seeing one supposedly “super slow” method using dump, and another only partially applicable method using ZFS snapshots.

That looks a bit strange to me. A user needs one proper, reliable, and reasonably convenient way to back up the system to an external drive.

you could always try rsyn---ah shit never mind

there are dozens of other tools, but dump is not much more complicated than a 'tar -cvf'.

JurassCZ: not usable depends on how you're using it / how you expect it to work

i also use restic to a s3 style endpoint. it does de-duplication and i can restore a single file from a snapshot rather than having to restore the whole parition

I use UrBackup and have been pleasantly surprised by it's usefulness.

JurassCZ: What kind of backup, exactly, are you looking for?

Why rule out ZFS? You can take a recursive snapshot that includes all of your datasets, send it to a file, or to another machine.

o0x1eef: Exactly. I only use UrBackup for (mostly) testing (and I maintain the port, so I kinda gotta make sure it works.) Otherwise, ZFS has been super solid for me for a very long time.

But, if they're using UFS (which it sounds like), that isn't an option.

JurassCZ: Also, "dump" isn't necessarily "slow" at all. It depends on how much data you're transferring and how you're (un)compressing the data before and after transfer. If at all.

Ok didn't knew that ZFS snapshot can be send externally.

I think the question was more like, what's the standard way on FreeBSD, and I think it is ZFS. IIRC UFS is not the default.

JurassCZ: Yep. ZFS snapshots can be sent directly to a disk, to a file, to another dataset for safekeeping, etc... You can basically send a snapshot anywhere.

<o0x1eef: right, ZFS then

o0x1eef: I agree. I'm sure there are certain cases that require UFS. But, if you have no absolutely reason to use it, I would always recommend ZFS as the filesystem.

JurassCZ: github.com/jimsalterjrs/sanoid

check this utility out

zfs send / zfs receive / zfs snapshot are all your friends :) can send anywhere

Sanoid is a pretty nice tool!

I wrote all my own scripts years ago for automating snapshots and eventually migrated over to sanoid a while back. It's just too simple and I don't need to maintain my scripts anymore. :)

I'm getting lazy in my old age.

s/lazy/wise/

nah, it is the whole learn how to do something yourself and when you comes across something better, use that :)

o0x1eef: I dunno about all that!

it is like, learning how to do electrical in a house and can do basics but when it gets to large jobs let the professionals do it

voy4g3r2: I agree. I still try to help out where I can, though. I certainly don't want OSS creators to only feel used. :)

I give back!

If you're a builder/programmer it helps to realize your time is finite so you have to choose what makes you happy, and too many projects is a quick way to ruin that

ek: do unto others what you want doen to yourself

pretty simple train of thought

o0x1eef: Completely agree. That's why I try to help out where I can. If I'm reporting a bug, I like to also submit a patch along with it (when I can). Hopefully, this saves the real devs time.

Not just a, "Hey! This doesn't work!"

oh speaking of bug.. wonder what happened with that make fetch thingie you helped me with ek

submitted a report..

Sounds like being a good open source citizen :)

Oh. The refetching thing?

o0x1eef: I do my best when I can.

Whether I submit a patch or not usually depends on how out of my depth I am. I don't know how to design a language, or any of the mechanics around it like a VM, so if I find a language bug, I'll do my best to report it, but I won't try to fix something I'm clueless about. That's usually how it goes.

o0x1eef: Same. Hence I mentioned "when I can". If it's totally out of my element, I just provide as much information as I can.

Again, not just a "This don't work! Please fix! DERP!"

Usually, between a couple buddies and myself, we can at least provide a reasonable patch or a good starting point. Helps to have smart around!

Thanks

JurassCZ: Sure thing. Give ZFS a shot and if that doesn't work, we can explore more ideas.

Good luck! I don't think you'll need it... It's ZFS. But, yeah.

:D

voy4g3r2: What bug did you submit for the python fetch issue?

voy4g3r2: Never mind. I found it.

Great: people submitting bugs!