hello

Hi.

blerp

hello there. Sorry skered I guess no.

morning, i got an old 11.3 server up that suddenly shows high load 12+ on an 8core cpu. But I can't see why...process load is minimal, disks are OK, I see no errors whatsoever in dmesg. Any ideas how to investigate this? Im not too used to freebsd. Thanks

systat reports high interrupt load hm

can someone help me interpret this vmstat result? pastebin.com/raw/n21DYSAc does it look like ix0 is the cause of this? ix1 has high count too thoguh

dirtwash: do you got two ix(4) interfaces. and you have setup queueing on them?

meena, how do you setup queueing? I have same interfaces as dirtwash but I don't have the same problem (14-CURRENT here)

meka: pf seems to be the only way to add support to it

and so maybe it's just blocklistd's queues?

Wait, I thought you're talking about multiqueue card and activating it's queues

interrupt count is usually a function of the traffic that an interface is receiving, if it's doing high-rate traffic handling it's usually a good idea to switch to polling

huh? how do you do that?

meka: I didn't know that was a thing

meena, I don't know much, either, I just know that enabling ATLQ in the kernel forces all ethernet drivers to switch to single queue, so I thought you know something I don't (which is not hard)

ALTQ/dummynet is about shaping traffic.

10:50 <meena> huh? how do you do that? ➡️ ifconfig ix0 polling

For example, if you're doing VoIP + lots of HTTP(S), you want to reserve a (relatively small) amount of bandwidth for the VoIP, because it's very sensitive to latency (and delta latency) - so you configure network queues such that VoIP is highly classified whereas HTTP(S) is classified as bulk traffic.

meena: no, you need device polling enabled in the kernel, otherwise it won't do anything.

oy, why isn't that in GENERIC yet?

Because it's not universally a benefit.

There's no point in using polling if you aren't doing near-linerate routing.

It's ultimately about reducing invulntary context switches at high load.

aye

this continues to be a very strange bug report bugs.freebsd.org/bugzilla/show_bug.cgi?id=270528

Title: 270528 – more swap is getting used

good morning

meena: if by strange you mean useless, sure

docs.freebsd.org/en/articles/problem-reports hasn't been followed, and there's no real evidence of there being a bug

Title: Writing FreeBSD Problem Reports | FreeBSD Documentation Portal

by no evidence of there being a bug, i mean there's no control that it's being compared against

that was my first comment

a crappily written bug report

"some stuff happened"

does at least mention releng/13

i'm not sure what relevance the branch has

besides which, there's no such thing as releng/13

well eliminates 11 12 and 14 at least

not really

there's no control, so you can't eliminate anything

for all we know from the bug report, the user is comparing against linux or windows

the two Operating Systems: oxide.computer FreeBSD and Ubuntu

Title: Oxide Computer Company

don't oxide target illumos?

🤷🏻‍♀️

i will say, i do love their rack scale designs that they've been talking about

rectifier in the middle, every machine is powered through direct current bus bars

they borrowed the concept from the hyperscalers, because the hyperscalers don't make their designs available to anyone else

(also it's part of opencompute, so anyone's free to copy the idea as far as i understand it, it's just that nobody has)

"borrowed"?

borrowed cf opencompute.org/documents/ocphl-r-copyleft-v10

i mean, opencompute probably didn't invent it either, to be honest

the idea of bus bars providing dc power isn't exactly new

but ocp is the first implementation of it i know of on a modern hardware platform

anyway, it's a neat idea in terms of not wasting power, we'll see how the rest of it plays out

debdrup: now you just need to get applications to suck less power…

meena: rctl, see cputime. ;)

set_loginclass(random()); + rcctl on that…

passing random pointer?

yuripv: a random string, something that doesn't exist in /etc/login.conf

if you pass a non existing class name to limits( 1) it says: (paraphrasing here, cuz I'm away from the computer again) this class doesn't exist, I'll go with default

what's rctl(8)'s underlying syscalls?

truss it! :D

apropos -s2 rctl

src/sys/kern/kern_racct.c is probably a good place to start

yuripv: i've read github.com/freebsd/freebsd-src/blob/main/usr.bin/rctl/rctl.c and it's just rctl_add_rule(2) but according to the docs it only adds the rule to the database…

Title: freebsd-src/rctl.c at main · freebsd/freebsd-src · GitHub

probably good that it got renamed, because that's a handful to type and remember

meena: wiki.freebsd.org/Hierarchical_Resource_Limits#Implementation_details

Title: Hierarchical_Resource_Limits - FreeBSD Wiki

hrl sounds more fun to say than rctl tho

yes, yes it is

meena: pastebin.com/XtQcV2r4

Title: $ grep -oE 'rctl_[a-z_]+' usr.bin/rctl/rctl.c | sort -urctl_add_rulerctl_get - Pastebin.com

so i'm not sure what you meant previously; also see the apropos command above :)

i wonder why that manual page doesn't have an alias for rctl(2) which is .Xr'd in rctl(4) and rctl(8)..

i don't see it: man.freebsd.org/cgi/man.cgi?query=r….0-CURRENT&arch=default&format=html

Title: rctl(4)

anybody care to help setting up the steam client? it reports 'authorization required but no authorization protocol specified' and also 'cannot open display:' i've follow the instruction and am running `steam' after the `steam-install' command on the `steam' dedicated account. I've setup a password for it, is that why it wont work?

meena: right, that's what I was saying; cgit.freebsd.org/src/tree/lib/libc/sys/Makefile.inc#n455 needs a link to rctl.2

Title: Makefile.inc « sys « libc « lib - src - FreeBSD source tree

and then rctl(4) and rctl(8) needs to have cross-references to it with .Xr rctl 2

ah

I misread you: i thought you meant: rctl(4) and rctl(8) reference rctl(2), but there's no such alias

fair

no harm no foul

gzar: do you allow this user to open windows in your session? e.g. unix.stackexchange.com/questions/61…user-when-display0-is-on-first-user ?

Title: x11 - How to enable xhost access from second user when display:0 is on first user? - Unix & Linux Stack Exchange

where is the rctl database?

very likely not, thanks, will check this out

meena: it's an in-memory thing that gets read from rctl.conf(5), iirc

not an on-disk thing like what's found in /var/db/

if the word dataset didn't have connotations from zfs, that might be a better word for it

it's not a database in the relational sense, more in the array sense

database as in organized collection of structured data, i guess is the way to say it properly?

i'm not saying racct/rctl uses arrays, obviously, since i haven't looked at it in probably a decade?

thanks ridcully_ , that did the trick

ridcully_: read your answer and started scrolling after someone with steam issues =D

debdrup: I want thinking /var/db, but more like /etc/pwd.db and their like

oh well, it seems like steam wont start either way

seems to be stuck in some 'assertion failure' loop or something

gzar: what's dtruss say?

i dont have dtruss

gzar: then install dtrace-toolkit

bugs.freebsd.org/bugzilla/show_bug.cgi?id=270546 this seems… dubious.

Title: 270546 – nullfs prevents zfs from freeing space

Aside from putting it inside of a jail, are there other options to limit a process to a specific network interface? (assume the process doesn't support configuration for that)

I was looking at the rctl man page, but didn't see a resource listed for netif.

In my case, the issue is that I use an app that utilizes UDP broadcast. My machine has a primary interface (re0) and a bridge0 interface (for jails, etc). Problem is that the app is is broadcasting on all interfaces, which is causing it to screw up due to duplicate broadcasts (one going out on re0, and another going out on bridge0).

I could use pf to block it, but I don't want to restrict the ability of bridged jails to be able to broadcast if necessary. Ideally it would be done via some host mechanism so that said process only sees re0, while everything else by default can see bridge0.

Oh, just hit me, can devfs.rules + devd accomplish this?

_xor: maybe you could use vlans to get seperate layer 2 multicast? Or maybe there some equivalent to virtual ethernet in freebsd that you could bind the process to? not very good with networking just thought id put out some thoughts

ohh udp multicast so not layer 2 multicast then sorry

hope do you set something at

as RT lists.freebsd.org/archives/freebsd-hackers/2023-March/002060.html ?

Title: Re: Periodic rant about SCHED_ULE

how…

let's try this again: how do you set a process / process group (service) to be RT?

real time priority?

man 1 rtprio

i.have been waiting years for this question

rtprio waited for yeara to become useful

lol

_xor: you're also waiting for your turn, isn't it?

rtprio: thanks!

rtprio: I'm surprised it's in section 1

it's/you're

i installed fbsd but there is no boot loader

linsux: FreeBSD has a thing called loader(8) and it generally has a menu, unless the timeout is set to 0

does it boot?