okay i'll be doing some configuration today for a very basic httpd server, ideally i'd like better network hardware than what i have but for this it should be sufficient

hi all.. can somebody look at this bugs.freebsd.org/bugzilla/show_bug.cgi?id=287754 there is no progress on it and its been a month and six days ? just would like a status update to see if anytihng is going on in the background i can't see

and how has freebsd been for the past month? should i look at the freebsd website for updates or any new devices added so i can update ?

oxbar: i am not sure if it would get tackled.. in bugzilla but it may be of interest to follow this wiki page: wiki.freebsd.org/LaptopDesktopWorkingGroup

there is an on-going effort, with funding, to tackle the laptop "problem"

i know this does not give you an answer to your internal speaker problem, defined in the bugzilla.. but this starting point may give you some guidance on how to investigate further.. or if anything see if the soundcard "stuff" has been put on the docket somewhere.

Thanks.. i will be afk again i don't know how long but will keep checking.. Thanks for the link i will bookmark it

Anyone have some guidance on wireguard and speed issues? I have a VPS where when I connect from home (500/500 fiber), I hit a max of 1Mb/s down, but then 250Mb/s or so up.

pf rules are minimal and pf is doing the NAT (I send all traffic over the VPN). CPU is not anywhere near maxed out. Vultr promises they don't throttle (I mean, the more I use, the more they bill).

A wireguard tunnel on a pfsense box at another location on Verizon fios lets me do about 200/200, but I'm not sure what odd tweaks they might be doing there.

spork_css_: A direct, non-encrypted download from the same Vultr VPS provides different numbers?

using iperf on the same port number and using either TCP or UDP gets me around 150Mb/s in either direction

if the vps is freebsd, my experience is tso/lro/csum murder wireguard performance on VPSes somehow

and if you flip them off with the appropriate sysctls things improve a lot

@mewt - thanks, I just set that up and rebooted and now I'm in a network maintenance window at that colo, so gotta wait to see if it does anything.

This sounds very familiar, I think I saw that when researching some other issue I was having when setting things up on pfsense.

i hope it helps. it helped me but i am only saying it for that reason, it's not super reasoned

@mewt - that totally did it - wireguard speeds now match raw speeds.

thanks so much!

found (a) bug about this: bugs.freebsd.org/bugzilla/show_bug.cgi?id=277718

V_PauAmma_V: thanks a lot

/j #¤mpd

witness_lock_list_get: witness exhausted

aren't we all

i have a bin that i always need running. it runs in userspace (for now) and i need some custom behavior so using daemon to supervise isn't an option, so i'm making my own supervisor bin. anything i should know? i basically plan to start supervisor with path to bin, it starts bin then monitors it and restarts it if it detects it's not running

anymore

sorry i got d/c

kerneldove, maybe read lockf(1)? You could set a crontab to run every X minutes. . . something like: lockf -t 2 /var/run/your_program.lock your_program --program-options

why not run the supervisor permanently? jc

My first use lockf was when I had a script that ran as a background process via crontab every 10 minutes, but sometimes the runtime was longer than 10 minutes - lockf made sure I didn't end up with multiple copies of it running. . . but it has other uses too.

i'm curious what this 'custom behavior' is that daemon(8) doesn't implement, in case it's something that could make sense

The program's PID (amongst other things) gets stored in the lock file. . . if the specified lockfile's PID doesn't match or isn't running, then the program gets (re-)launched, overwriting the lockfile.

i want the supervisor to 'pass through' the output of the bin as well as adding its own, so i can redirect it to a file then tail that and get full output, both the bin's output as well as the supervisor's output such as "bin not running, restarting..."

maybe daemon can do that?

kevans do you know?

kerneldove: no, but today's daemon(8) doesn't seem to actually have any output anyways. I don't see a reason we couldn't add a -v --verbose option or something

like, just thumbing through daemon.c again, everything's more or less an error

do you know how daemon monitors the status of whatever it's supervising? like does it just search pgrep for the bin name or something better?

I would expect it to use wait(2) or whatever it is now.

fork + exec, then kqueue(2) to watch for SIGCHLD

-v --verbose would be cool so i can keep track of crash&restart events

but i also need the bin's output to be passed through, dunno if that's an option for additions too

-o or -S

because i want to then redirect this combined output to a file

kerneldove: that's what --output-mask and --output-file already do

it would put both the bin and daemon's output into the file?

you would just want -v and that -v to either go to the file by default or add an --output-mask bit to adopt it to the file

there is no daemon output to go into the file today, so I guess technically yes since that's an empty set

how fast could the -v feature be added?

does anyone know what's required to make a bhyve guest detect the cpu configuration from bhyve? for example, booting with '-c 2,sockets=2,cores=1,threads=1' turns into 'FreeBSD/SMP: 1 package(s) x 1 core(s) x 2 hardware threads' in the guest

it's an easy feature to add, my question would be how fast would it take to propagate to whatever version you're using or planning to use

shit ivy

i'm flexible

kevans: i probably am, but what did i do this time?

i'm on 14.3 fwiw

ivy: no it's me not you

kevans could you add -v and add a daemonng pkg for me to install later today?

kerneldove: no, sorry, certainly not today. my backlog is the size of a walrus

tomorrow?

probably not, walrii are pretty large creatures compared to units of work

ok i'll make my own supervisor in rust but watch out for -v to be added to daemon and maybe switch

ivy: re: reviews.freebsd.org/D51243, is that commit message updated?

oh wait, there's a (show details) button

kevans: it is

I had just misunderstood after missing the word 'type'

ivy: also, I worded my response too strongly. it was just a suggestion, you're not my mentee anymore and you're more free to just disregard =D

? :o

kevans: you're too late, des taught me to squash everything into a single commit!

but perhaps i'll do that

"No no. . . every change has to be it's own commit."

right, in some sense it's a preference; otoh, they're independently useful and a separate commit gives you a chance to advertise it as functionality for folks interested in hacking on ifconfig

SarahMalik: it's not the drama that you might picture from what I wrote here, but reviews.freebsd.org/D51243

(for context)

Ah, hm.

if I were to write it again, I would s/please split/please consider splitting/

i've long since learned that phab sends e-mails quickly and a lot of people don't go to the review to view edits, so I bother less and less

kevans: part of the issue is i've learnt (from experience) that if you try to split things into multiple commits, it significantly reduces the chance of getting enough reviews to commit

this bhyve problem is annoying me, i shall ask virtualization@ about it

also, is anyone else having problems building multimedia/jellyfin recently? bugs.freebsd.org/bugzilla/show_bug.cgi?id=288572

ivy: sure, but now it's been reviewed and you can just split + push with reviewed-by :-)

ugh the github review interface feels awful after using phab

need to open 3 tabs just to view the commit message and the diff while writing a comment

I'm learning to build pkgs with poudriere, would anyone like a build of x11/kde 'latest" ports branch ?

basically I just need to know the platform/version of FreeBSD, and the ports branch you want, I default to the latest and fbsd current but can do anything that doesn't error out ;)

having to manually add a bunch of stuff to the never mfc list is a bit tedious

ivy: I'm building jellyfin soon, so I will tell you what happens if I remember

ivy: krb5 between kdenetworking and jellyfin has had a file conflict, so I might have to do something to config on one of them

ivy: are you on the 'latest' ports branch, and what fbsd version?

ivy: oh, never mind -- I see you listed that in the details already, my bad ;)

sponix2ipfw: ports 7a3f3e7020294fb03a1c1c06b72740524281e0b8 (that is latest from a few hours ago), src 29af6d2e2ec9fe8df7cf1e1a0bf3597028831b18 (main from a few weeks ago)

+ some local ports/src patches that almost certainly wouldn't affect this

glad to see someone else on current

I was going to switch from Plex to Jellyfin, but it turned out the feature that they started requiring payment for, was one I don't really use, so. . . the migration got delayed.

i evaluated both and Plex appeared to require an online account even for "free" functionality, which made jellyfin the much better choice

i have x11/kde all going right now. if I wasn't already using about half my ram with tmpfs on that, i would spin up another jail and try the jellyfin build right now

Well, Plex's client always required purchase, but. . . I think I paid $5 for the Android client and $10 for the FireStick client, which seemed reasonable enough.

it's more the online stuff than the cost, i pay for Infuse anyway to use jellyfin on ios/ipad

But Plex now wants a membership with a recurring, monthly fee for "remote streaming," but I only do monthly streaming.

I broke down and did the life time Plex pass right before the prices moved UP

I will not be giving them any more money.

I was doing monthly prior, just to keep supporting Plex, but didn't want to do that at the higher price structure

Jellyfin has better hardware support anyway, even being coded in C#

Though honestly, Plex does keep getting enshitified. . . they're continually buring the menu options for playing your own media. . . trying to push you toward their own content.

Yeah, they are on a downhill slide, and it is pretty sad

death, taxes, enshitification

Is Plex publically traded?

bdrewery: I thought that was just a Microsoft thing ;)

Far from it.

Um, Google?

CrtxReavr: if so, that would explain the behavior

Any Zuckerberg project.

ivy: re: mfctracker, doesn't really look like it, but I don't think it'd be too hard to add

I'm a terrible person, I went all in on Google way back in the " don't be evil " days. and now question that, but would be hard to pull back

what does 'mfc' stand for?

Merge From Current

ie. back-porting new OS features to older versions.

SponiX: merge from current, which is when a commit is backported from main (freebsd -CURRENT) into a stable branch

oh, you all are talking real FreeBSD coding

no wonder I got lost ;)

kevans: did you find the source code somewhere?

The only time I tracked -CURRENT was during the 5.x SMP instability woes and the early 6.x dueling scheduler drama.

I got a book on C (21st Century C), but it is still too advanced for me. I need something basic

oh, github.com/freebsd/mfctracker

So, I have roblox "scripting school" teaching me lua right now LOL

whatever works

speaking of jellyfin has anyone been able to get it to use, in a jail, a gpu :) for transcoding

i keep eyeing the subproject.. they optimized ffmpeg for their use.. it is a great project.. especially with plex doing a money grab...

voy4g3r2: I run jellyfin naked on my host OS install

(As God intended!)

I have yet to learn to use jails

ivy: it's on github, freebsd/mfctracker

oh, yeah

CrtxReavr: for me, if I have an app that doesn't misbehave and color outside the lines, I don't really even consider putting it into a container

(I've now read to the end :-))

I have one python app running in podman on my Linux box, just because it acts as a Microsoft Volume License server for my oldest Microsoft Windows 11 Pro for Workstations box

Windows. . . and some Linux distros are what really needs jails.

i just converted a bunch of jails into VMs but i'm pretty unhappy with bhyve cpu scheduling so i may rethink this

It's really too bad Broadcomm shat all over vSphere.

It was an amazing and mature software suite.

ivy: Can you be a tad more specific about your unhappiness with bhyve's CPU scheduling? I'm just curious what you're seeing. I migrated to bhyve a while back and am certainly interested in any shortcomings you're experiencing. I'd like to be on the lookout if I should.

ek: so, i have 8 host CPUs, i created a VM with priority 10 (i.e., lower than default) with 7 CPUs and run make -j12 buildworld inside that VM. all my other VMs, which have 2 CPUs each, become starved for CPU time, so badly that e.g. communication between grafana and prometheus times out

ivy: Ah, okay. That's unfortunate.

I definitely haven't beat my VM's up much so haven't had that happen (yet). Thanks for the info.

also, there's a weird bug where booting GENERIC on bhyve hangs for several minutes at startup: bugs.freebsd.org/bugzilla/show_bug.cgi?id=278535 - i noticed this the last time i used bhyve and apparently it's still present

since no one else seems to run into this it might be something i'm doing, but i can't think what, it's a fairly standard vm-bhyve setup

ivy: I was doing virtmanager bhyve and it seemed to let my FreeBSD build vm use more CPU that I was giving it, allowing it to starve my Host OS, leaving my Media Services in shambles

this might be fixed by better poudriere configuration inside the vm, but that too me defeats the point of having the vm in the 1st place

might be "user error" on my part though, as I am very new to byhve and am use to virtmanager with qemu/kvm on Linux

Interesting. I haven't run into either of those scenarios (again, "yet"). Thanks for the heads up.

ivy: I have a bunch of FBSD VM's and haven't seen that boot hang before. Which version(s) are you seeing this with? Any/All?

ek: 15.0

ivy: Huh. I see you're on AMD while I'm using Intel. I wonder if that may be part of it?

ek: i wondered, but no, i'm now seeing the problem on Intel as well

Or, perhaps that suggestion of lock waiting is the culprit. I wouldn't think it would be MINUTES, though. That's wild.

ivy: Ah. No bueno.

(CPU: Intel(R) Xeon(R) E E-2468 (2611.20-MHz K8-class CPU))

I'm: Intel(R) Xeon(R) CPU E5-2690 v4 @ 2.60GHz

Plenty of 15.0 VM's I use for testing builds and such since clang/llvm jumped and broke some of my ports.

ivy: Which exact snapshot of 15.0 do you see this with? I wanna try to duplicate.

Intel Xeon E5-2698 v4

sponix2ipfw: Thanks.

ek: 29af6d2e2ec9fe8df7cf1e1a0bf3597028831b18, but note the original bug report is over a year old, so it's not a new problem

ivy: Yeah. I saw it was a while back. I figured you've tried multiple different snapshots since then.

one odd thing is it only affects my VMs with a GENERIC kernel, not the cut-down bhyve kernel, but that might just be because those VMs also have more CPUs

also, i need to verify this, but it may have only started happening once i turned WITNESS off on the host

How many CPU's are you assinging?

the "big" VMs have 7 or 8, the small ones have 2

in total 3 VMs with 7 or 8 and 8 VMs with 2

Thanks.

le-fay.org/tmp/7d/sgDJSaxc77 - mostly idle, should be a very light load for this host (and indeed in normal operation load avg is ~0.2)

i thought switching from 8 to 7 guest CPUs might help the problem by leaving an idle host core, but it did not seem to make a difference

sponix2ipfw: haha, yeah well i was doing a plex / jellyfin and for the longest time i could not get the ports package to "work".. which lead me down setting up poudriere and this whole "setup"

now i have a jail.. that anyone can get to and i am like.. dang it.. how can i use my gpu to do the transcoding.. because gpu/jail/bhyve blargh

gpu passthru to a jail is a new one

i do not think it is possible.. some how i could "pass through" the /dev tree but failed

i got real frustrated with.. a bhyve.. and it SHOULD work.. debian COULD see it.. but be damned if it would work.. so i just have llama.cpp running on host and it is operational

hmm, isn't pf rdr-to supposed to change the *destination* address? it's changing the source address for me

rtprio: I thought a jail was more like a container than a vm and passing through the gpu would be easy -- I have no experience with FreeBSD jails though

a) i don't think so and b) none of my systems have video cards so

I have syslog-ng for my remote logging server. But its documentation is terrible and it doesn't have a proper manual or even a straight administration guide. Which syslog service do you use? built-in syslog? rsyslog?

ty for the link. So you prefer rsyslog.

Not sure if anyone else has one however I am happy to report the Orion-O6 runs FreeBSD-15 quite well...

(Radxa Orion-O6)