lw: I think that's the main goal of some WIP by markj: github.com/markjdb/bricoler

Title: GitHub - markjdb/bricoler

hello, I'm using unbound as a local resolver here with some static zone. Is there a way when I'm creating a jail to automatically register it's DNS name without having to modify the unbound file?

mage unfortunately not that i know of

the hostname is had by dhcp and with ipv6 you don't really want dhcp

the good news is that since your working with jails you can create a custom script to extract the internal ip of each jail and update your zone file

if you do this let me know, i'd like a copy of the script too

why is that necessary?

mage: yes, you can do that with dynamic DNS updates - except i'm not sure if that works with unbound, since it's not intended to be an authoritative server; bind or knot might work better

although it would probably be easier to just do this once rather than configure ddns, unless you're creating a lot of new jails

compiling freebsd over ssh, yay or nay

actually is there a way to speed this up, we'll be doing this a lot...

presumably meta mode should help

Soni: over ssh or sshfs

dstolfa: over ssh, we're not *that* masochistic

Soni: why would that not work? people ssh into build machines for freebsd all the time

you just ssh in, fire up something like tmux and kick off a build

just run the build in tmux in case your ssh connection dies. and yes, use meta mode

because we're not using tmux

well, if your connection is stable it doesn't matter

tho, to be fair, mosh -> ssh -> vm is probably fine

alternatively you can probably hack something up with nohup

is "running on the same machine" stable enough?

unless you plan on killing the connectivity from the vm to the host or vice-versa, probably :P

alright, btw, when rebuilding libc...

is there a way to avoid rebuilding everything?

-DNO_CLEAN, WITH_META_MODE, ccache for the most part

beware of edge cases and all that

(did we pick the worst project)

i recommend not using NO_CLEAN if you're using meta mode, it's not really necessary and it does break things sometimes

(basically the point of meta mode was to make NO_CLEAN unnecessary)

all 3 of those break in some edge cases, so just generally beware

also check the various WITHOUT_* options in src.conf (if you're not using pkgbase) as not building things you don't need can improve build time a bit

best build time improvement: Not building LLVM, but that might building things more difficult.

meena: how do you do that

lw: i'm not sure it's possible lol

i'm annoyed at having to build 20 identical copies of llvm for every src tree, ccache doesn't even seem to help with this

oh

zroot/src 169G 475G 61.3G /src

to not build llvm - WITHOUT_TOOLCHAIN

the system's not terribly usable if you do that, however

it's useful for some deployments but i think you'll struggle on a regular server/desktop

it will be bad if you do delete-old and delete-old-libs after that, yes

noted

lw: thanks, I'll look into dynamic dns updates

(the use case is Gitlab CI/CD which creates jails on the fly when a push/merge is made)

i'm not sure if you're using DHCP for this (someone else mentioned it?) but if you are, most DHCP servers can do this for you

yes, I'm still using ISC DHCPD

but I was wondering on the unbound side

i'm fairly sure unbound can't do this, you need a real authoritative server for the zone. but you could run both on one host and have unbound forward queries

does NSD supports this?

i don't think nsd can do dynamic dns

it seems like there are some third-party options that can update the zone file on disk in response to dynamic updates so nsd can reload it

(in that case, you could also keep using unbound...)

yeah..

I'm using Saltstack under the hood so maybe I could just add this to my orchestration script

thanks

i have never used this and know nothing about it but: rubygems.org/gems/dyndnsd/versions/3.8.0

Title: dyndnsd | RubyGems.org | your community gem host

oh wait, i think this is a client, never mind

the fun never ends

there should be a video game in 'make buildworld' where you have to complete a level to build the next directory, that would make it more fun

good grief

does modifying libc require rebuilding llvm?

no

assuming you already built llvm once for the initial build

ugh, i wanted to stop FreeBSD-utilities from depending on FreeBSD-zfs, but fstyp(8) needs libzfs_core...

i recently started using bhyve-vm and set up a windows 10 vm with "cpu=8" in the config file, but it only shows me and uses 2 threads, anyone got an idea what's up with that?

phryk: I'm not too familiar with bhyve, but does the "cpu=8" parameter specify the amount of CPU sockets? Because due to licensing, Windows will limit itself to a certain amount of sockets, depending on the edition you have, but if you configure to instead give it cores instead, it will use all of them

(as in, one CPU socket, 8 cores)

Dooshki: sample.config says "# specify the number of cpu cores to give to the guest" about this parameter

also let me look at the activation status, i think that might actually be it.

Workstation versions of Windows have only ever supported a max of two sockets.

And four for server versions, without custom HALs.

git add -i must be the least intuitive command line interface ever created

i don't know if drm-kmod, sway, or the kernel scheduler is at fault here but i'm a really annoying issue where sometimes if i'm building something (e.g. make -j12 buildworld) my cursor starts moving really slowly and the desktop becomes unusable

I have a box where 'pkg update' says it's updating the repo catalog and that everything is up to date. . . but there's no way it can be.

There some forced check or setting I can poke at?

CrtxReavr: if pkg says it's up to date it's up to date, it doesn't randomly hold back updates - are you sure you're using the right repository?

lw, No. . . I'm not.

i will look at your url in a sec but i can't move my mouse due above issue...

CrtxReavr: why do you think there should be updates?

'cause it hasn't been updated in a "coon's age."

CrtxReavr: one thing you can try is replace pkg+http:// with http:// in /etc/pkg/FreeBSD.conf -- you might have landed on an out of date mirror

wait, no

use pkg0.nyi.freebsd.org

Title: pkg0.nyi.FreeBSD.org

that's the master mirror

And that gets set wher?

where

/etc/pkg/FreeBSD.conf

you're meant to copy that to a new file in /usr/local/share/pkg/repos to edit but you can just edit in-place for testing

url: "pkg+http://pkg0.nyi.freebsd.org/${ABI}/quarterly",

Like that/

?

use http:// not pkg+http://

pkg+http:// is for automatic mirror selection using SRV records, but you want to force it to use that mirror

Okay, now it's doing something. . .

Hmmmm - okay, not quite.

this is giving me flashbacks to when i used to run a cvsup mirror

Title: View paste G2KA

Remember when the billionth second broke cvsup?

That was such a huge pain in the ass.

Y2K had nothing on the billionth second.

are you really sure there are meant to be updates? like, take a random package and look it up on freshports and see what the latest version for FreeBSD:13:amd64 is

CrtxReavr: what's your config look like rn?

What about that SRV record error?

oh, that's because of mirror_type

i don't think that makes a difference but you can try... (one sec)

CrtxReavr: just comment out mirror_type in FreeBSD.conf

that should fix the SRV thing

what lw said

Okay. . . it's gone. . . still no updates. . .

CrtxReavr: show 'pkg info'?

How about 'pkg version'?: termbin.com/01tv

I mean, I can do 'pkg info' but that's prolly less helpful.

that works too

oh qutebrowser crashed

CrtxReavr: just checking a couple of packages (python39, curl) it seems like you might just not have updates

Yeah. . . wierd. . . but yeah.

Thanks though.

although it's odd that FreeBSD:13:amd64 quarterly has curl 8.6.0 while 14 has 8.7.1

maybe the 13 builder is lagging a bit?

ah wait, didn't a new quarterly just come out, it might be busy building that

8.7.1 went into quarterly on april 1 though

i wish pkg-status.freebsd.org was less awful

this is why i would never use pkg.freebsd.org packages, fwiw. the builds take so long you can be weeks out of date on security updates, especially on some architectures

is this a money for servers problem? why does it lag so much?

markmcb: imo it's more of a ports infrastructure problem... or maybe a poudriere problem. there's a fix landing in poudriere soon that will significantly reduce the number of packages to be rebuilt when a dependency changes, which should help a lot with this

ah that's good news

i'm not even sure if pkg.freebsd.org uses incremental builds (as opposed to just rebuilding the entire thing every time) but i assume they do

(i say 'fix' but aiui that patch has been in development about two years, it's pretty complicated to fix)

why can jail.conf have just vnet; and work? man jail says its values are inherit and new so it should be vnet = new; no?

alepzi: because the manpage is wrong, you should submit a patch

when would a jail admin wanna be able to read kernel message buffer?

alepzi: some interesting messages only go to kern msgbuf

like: someone else is using my ip address on <interface>

or arp: <ip> moved from <mac> to <mac>

or <intf> link state changed to <UP|DOWN> (although that's probably uncommon in jails)

or messages from programs that exit due signals

or various NFS errors (i think NFS can run in jail now, although maybe only the server part)

is that /var/log/messages?

it does end up in /var/log/messages but only because syslogd can read it from kernel msgbuf

i think

actually not 100% sure there

how do i use "depend" in a jail conf? i tried to put depend = "testjail"; in the conf of testjail2 but service jail start says cannot start testjail2 depends on undefined jail "testjail"

are you using /etc/jail.d?

jail.conf.d, whatever it's called

each jail has a file in /etc/jail.conf.d/

like testjail.conf, testjail2.conf

yeah that doesn't work depend, there's a PR for it

damn

either stop using depend or put them all in jail.conf

fwiw, i don't like depend and don't think it works well

why not?

just because a jail is started doesn't mean its services are available

better to make each jail able to start individually, perhaps with some extra rc sripts if needed

e.g. my KDC jails have an rc.d script to wait for LDAP to become available because otherwise the KDC doesn't start (this is a KDC bug really, but...)

nice

i noticed the handbook has you set up a jail so its console output goes into a file on the jail host. why not have it go into the jail's /var/log like a normal system?

the handbook can't consider every possible configuration someone might want, if you want the console log in the jail, just configure it like that

just read all of man jail

lw, I enabled EXPO, it's building even faster now ^^

jbo: the fuck's an expo

AMD's version of XMP

ah

is that different from AMP?

dunno, first time doing AMD stuff

everything is different for me atm :p

i think my AM4 board just does XMP but that's pretty old, it's probably changed since then

I am trying to figure out why the CPU has so few clock freqs

dev.cpu.0.freq_levels: 4200/5040 2200/1540 1500/975

jbo: a little birdy (diizzy) told me Zen 4 support is bad and doesn't work properly, how are you finding it?

is that normal in team red land?

not sure about your fancy TR but my Ryzen has only 3 also: dev.cpu.0.freq_levels: 3400/3740 2800/2800 2200/1980

lw, I can't speak for that yet. I setup the box a few days ago and only since today I am actually using it as a desktop/workstation.

interesting yours go down to 1500, i wish the 2700X in my server did that

if you keep using that language I'm gonna ship a TR your way just to make it staph

no stepping between 2.2GHz and 4.2GHz is a bit weird though

I was surprised to see those freqs. form Intel I'm used to seeing more freqs. sometimes the CPU would just clock around 650 MHz or something :D

yeah

maybe this is what diizzy meant?

dunno, diizzy is just jelly

like i wonder if they switched to a new API or something and this is the old API

I hope to find all of this out soon (tm)

Title: 278311 – amdtemp: Does not recognize AMD Threadripper 7960X

i saw that commit in my inbox and though, oh jbo has one of those, oh it was jbo who reported the bug

you are the only freebsd 7960X apparently, congrats

don't thank me, thank $employer

maybe I should start doing something useful with it soon

i can send my ssh public key, let me know

lw, would it help src people if I send experience summary to the ML or something like that? or is that considered noise?

jbo: i don't consider myself 'src people' but i think reporting hardware experiences is useless in general, someone might be searching for it later

i'm not exactly sure what is the best list for that though

uhhh

i mean it's USEFUL, not useless

damned keyboard

maybe amd64@

lol'd - I am/was genuienly confused now

let me try that again

did you learn how to paste yet?

jbo: yes you should send it to the list because it's useful info for people who will be searching for experiences with this cpu on freebsd

jbo: i keep leaving these glasses of whiskey next to my keyboard and mouse and they start behaving weirdly and not typing/pasting properly, i don't know why it keeps happening

did you know there's an entire mailing list for Enlightenment on FreeBSD? the 3 people still using Enlightenment must be having a great time there

also, freebsd-women@, has not had a post since 2022, that's probably a sad reflection on something

amd64@ seems pretty dead too, i was complaining yesterday there are far too many mailing lists

hardware@, had 21 posts in the whole of 2023

jbo: so in summary, yes you should mail the list, but as to what list, your guess is as good as mine

lw, got some whiskey right now o/

ack

maybe just mail -current, assuming you're running 15.0 as god intended

(god = kirk mckusick)

lw, is there a way in C++20, to convert a constexpr integer to a hex string_view representation during compile time?

I am _not_ running -current

lw, also, are you running powerdxx on your AMD machines?

no, i use the normal powerd on both amd64 and arm64

and i do not know about constexpr but maybe check if std::to_string is constexpr (i suspect not)

because reasons?

because... it's what's in base and it seemed to work fine

wasn't powerdxx supposed to be "better"?

what does powerdxx do better?

aye

lw, can you explaine those freqs to me?

it seems to be three pairs

what are the two number?

how to report mailing list spammer le-fay.org/tmp/30d/Screenshot.from.…024-04-16.at.21_18_27.112234313.png

base/boost? but then why is the magnitude/order different?

jbo: first number is cpu speed (MHz), second number is supposed to be power consumption (mW) but i'm extremely dubious about its accuracy in general

also not clear if that's mW per core or total for the package

well doing a stress -c 48 results in 560W power draw (of the entire machine/box).

nvidia-smi says 40W

that is with stock settings, no OC

why did we think it was a good idea to start the build from our phone, now we can't reboot our phone until the build finishes

i'm pretty sure someone suggested using tmux. oh yeah, that was me

yeah well; it's building libclang now

oh right we have our cpu scaling set to powersave

we could probably have been done already but instead we were running it at half the speed

jbo: remember when you told me there are 4,000 src committers? i have reliable information that this is fake news

finally got around to fixing my alc(4) issue llfw/freebsd-src ec630c5

Title: alc(4): disable MSI-X by default on Killer cards · llfw/freebsd-src@ec630c5 · GitHub

don't even use that card anymore (it's crap) but it annoyed me enough at the time that i thought i should do something about it

ok now what

i did a pkg upgrade and my sway status bar vanished

I have questions

who would like to answer some questions?

jbo: no one

questions are bad

:<<<<<<<

you will accept what we tell you

not sure whether I understand poudriere.conf BASEFS=

Title: Don't ask to ask, just ask

I set ZROOTFS=/ssd0/poudriere where ssd0/poudriere is a dataset

so I guess I want to set BASEFS=/ssd0/poudriere ?

no

ZPOOL=ssd0

ZROOTFS=/poudriere

BASEFS=/poudriere

that... makes sense

thanks <3

but create the zfs before you run poudriere: zfs create -o mountpoint=/poudriere ssd0/poudriere

wait... why the mountpoint?

is there something wrong with just mounting it under /ssd0/poudriere?

no but then you have to set $ZROOTFS and $BASEFS to /ssd0/poudriere and i think it gets complicated

one of those is reletive to $ZPOOL and i think the other isn't, can't remember which

yeah so I just did what you said but without mountpoint= and it still dropped the jail in /usr/local/poudriere/jails

must be AMD's fault

are you sure it actually did that, it's a bit confusing

it'll create filesystems in /poudriere then mount them in /usr/local/poudriere

wat

zfs list

ssd0/poudriere/jails 1.70G 898G 96K /ssd0/poudriere/jails

ssd0/poudriere/jails/140Ramd64 1.70G 898G 1.70G /poudriere/jails/140Ramd64

lw, it actually seems to set mountpoint= automatically then?

yes exactly, that's what it does

although, wait, maybe not

i think one of those options tells it where to create the filesystems and the other says where to mount them

but i can't remember which is which

now we get back to why I asked

but like, just put them on /poudriere, it works

ugh i need to find a new window manager

what do people use on wayland, hyprland? apparently some freebsd people wrote that?

on wayland I once used hikari

worked nicely

oh, hikari, that's what i was thinking of

lw, how can I confirm under FreeBSD that the TR is actually doing the boost freq thingy?

it's always reporting 4.2 max

i don't think freebsd has any way to monitor that

i'm not even sure it's exposed to the OS at all

trying to build hikari

[main-amd64-main] [2024-04-17_00h54m00s] [balancing_pool] Queued: 231 Built: 0 Failed: 0 Skipped: 0 Ignored: 0 Fetched: 0 Tobuild: 231 Time: 00:00:24

FUCKING PORTS

that's good right?