unixwitch: i was told turn off things on my bridge'd nic, but can't recall why now: "-tso4 -rxcsum -txcsum"

i had to do something like that on a linux to resolve/work around an issue where the nic (an intel nic) would periodically bounce

networking is dark magic

unixwitch: have you seen blog.rlwinm.de/the-correct-way-to-c…ridges-in-freebsd-for-ipv6-and-ipv4 ?

Title: The correct way to configure bridges in FreeBSD for IPv6 (and IPv4). — blog.rlwinm.de

... is the output of `pw usershow` intended to be stable (unchanging from version to version)?

Ellenor: if you want stable output, try getent passwd

Also, the easiest way to know if the output is stable is to look at the git history of the file that's generating that output

if it's been stable for >5 years, breaking it would be considered bad form

meena, that doesn't give the user's login class, which VVD needs for some reason. (`pw usershow` does)

what other things can getent query?

ethers, group, hosts, netgroup, networks, protocols, rpc, services, shells, and utmpx, per its man page

I wrote a C program that breaks out the specific field, rather than needing to use `pw usershow` and `awk`, but I suspect that's a capital-W Wrong solution to the problem.

somebody should add libxo output to pw

I'd rather become an alcoholic

yeah, there's definitely more fun things to do with life

I really badly want libxo for ifconfig

but i also don't wanna actually poison my brain with the API

(-:

hmm, maybe it's because I'm the weird one but adding libxo to ifconfig and pw sounds like a fun little side project

it's really mostly cuz other things were more important

Also, i need to look into making pw work without root, so that build/install world can better work without root

meena, how about adding 802.11v support? :>

jbo: you would make people happy by adding libxo support to commands

I wish I would have known about libxo before I wrote my own thing for that a couple of years ago :/

last time I checked the libxo docs it seemed fairly easy to use. I assume that adding libxo support to ifconfig and pw is mostly an exercise in patience and endurance

Won't that double the size of the code?

meena: regarding that bridge post, why would the bridge interface have an ip? shouldn't a bridge be just like a layer 2 switch?

babz: it's one very thin c library why would it double the size of ifconfig?

nimaje: i don't know, ask crest

meena, without knowing any details I assume that it will inflate the code size considerably

(code size from the memory segments perspective)

I was talking about the source code

hm, I expect libxo support to be mostly replacing printf and similar calls to be replaced by libxo functions plus some call to some init function at the start

But maybe it's just link against this lib an now all the text output is automatically converted to xml :)

wouldn't that be nice

no, you do have to replace the printfs

it's gonna be xo_emit() all over the place :D

ah, I think the use case where you want the bridge to have an ip is when you bridge multiple physical interfaces, so the bridge is your computer and the use case I had in mind was to connect vnet jails to the outside, so the bridge is only internal

yeah, libxo conversion can be a bit dirty unfortunately

kevans, would it actually wrap existing printf() calls?

wrap as in: replace printf() with xo_emit() ?

jbo: yes, but there are edge cases sadly

understandable

the thing I dislike about this sort of thing is that the XML output is basically a "meta output". that was the whole reason why I wrote a very ugly serializer once.

for the most part it works though. the biggest things to pay attention to are when things are flushed (because this can mess up the interaction between xo_emit() and printf output) and that you're actually producing valid data

that sounds like somebody would have a fun time reviewing any potential contribution along those lines :)

unixwitch, ping

ugh I just ran fastboot(1) thinking it was the android command

adventerous

Howdy, I have a failed upgrade to 14.0. My kernel is 14, but userland is 13.2. I have updates pending, which if I install using freebds-update install, I get a missing library when executing lots of things and the system fails to boot. Is there a way to complete the upgrade, or is a reinstall necessary? Also, future me is wondering if a boot environent would help.

+1 for using boot environments

having the kernel on 14.0 while userland is still 13.2 is actually part of the regular procedure

what exactly is failing?

did you follow the update guide that shipped with the release notes?

also, technically you should update the bootloader AFTER installing the new kernel but BEFORE booting into it.

I think I lost shell on one of these boxes during the upgrade. ld-elf.so.1: Shared object "libcrypto.so.30" not found, required by "sshd" is the error. I'll double check the release notes.

pro tip: when updating remote systems always use a tmux session (or similar). this way you can jump back into it after loosing the connection

if the system still boots I'd try to re-install the kernel just to make sure that everything is dandy there

Yes, remembering to use mosh+tmux again :)

Well, I'm booted on it now, but nothing is usable. zfs binary fails to execute.

frebsd-update rollback seems to cure the missing libs.

ld-elf.so.1: Shared object "libcrypto.so.30" not found, required by "zfs"

but... what

Those instructions look like what I've been using for years, so I think the lost shell in the middle is to blame. freebsd-version -kru shows 14.0 14.0 13.2

If it were me I would definitely use the Boot Environment to boot back to the previous which is guarenteed to be consistent.

I dont think I've eve setup boot environments. So unless that's done for me, not sure.

zale23, libcrypto.so.30 would be part of the kernel AFAIK. are you sure that you installed it successfully?

Try: bootctl list

libbe_init("") failed.

is this even a ZFS system?

It's automatically configured if the system is using zfs and freebsd-update

jbo: ah okay, I thought that was part of the sytem.

Definately a zfs system.

but if it's a ZFS system then you should have a boot environment automatically created for you with freebsd-update

but now I go freebsd-update rollback and I can zfs list again.

I think that feature was added in 13.x or something

rwp, bootctl list? isn't that bectl list?

Yeah, the output above is from bectl list

(after the rollback)

So can I completely restart the upgrade to ensure a clean kernel etc?

Jail are working fine :)

is rollback using BEs under the hood? I usually just select the BE in the loader

I'm almost never in front of this system while its booting unless it doesn't.

if this was a source update I would just re-install the kernel. not sure how you do that with freebsd-update

Future me will read wiki.freebsd.org/BootEnvironments unless there is a better doc for understanding.

Title: BootEnvironments - FreeBSD Wiki

zale23, there are two ways to use BEs: create a new BE, boot into it and then do work there and stay there if it works out. choose previous BE if it doesn't. The other option is what I usually do: create BE, but stay in the current one. Only use the newly created one if something failed.

AFAIK the latter option is also what freebsd-update does automatically nowdays

I don't mind compiling a kernel, I'm only using generic. If I started the compile for the kernel, would I also need to compile userland? Currently already have a ports tree with poudriere on 14 and 13.2.

I dunno I would personally try to fix the binary udpate

I never had one fail so I don't know how :D

Sounds like I should have a be currently, but don't. Hmm.

if it's definitely a ZFS system I would check whether you have the BE and boot into that.

how do you now that you don't have one if libbe is failing?

Yeah I guess I don't. The rollback fixed the rest of the commands which were failing for libcrypto reasons. Are you thinking that I'd be able to see the boot environments from the loader?

yes, the load shows the available BEs

By default?

yes

it's an option in the menu

where you select multi-user mode, single-user mode etc.

I don't think it was present. I had a rough go geting this thing to boot. Single user -> remount root, rollback but took a bit to learn that's what I needed.

now you make me doubt whether I'm talking shit

Dunno :)

I can reboot, but its like a 10 minute boot before I get to the loader prompt.

Okay maybe like 6 minutes.

Yeah, so no updates pending, fetch -> install -> ld-elf.so failing for libpcap, libcrypto and other reasons. Further fetching fails and rollback required.

if they exist there will be a submenu, but it shouldn't show you anything bectl list doesn't

zale23, you made me boot a machine... it is definitely listing the boot environments if you have more than one.

in the loader menu that is. entry 8.

BEs without being able to specify one in the load would be half as useful :D

Okay, I'll reboot tomorrow and check it out.

Makes sense, that's what I thought BEs were for.

yes

and afaik since 13 freebsd-update auto-creates BEs on install or something

So, the goal is to boot the environment before I started the upgrade, then start the upgrade again from 13.2 -> 14.0, correct.

I don't trust it so I always create one myself explicitly.

yes.

lol

Just bectl create?

yeah. I usually do bectl create 20230102 and move on

And what is that, a zfs snapshot?

/usr/local is still only one copy?

afaik it's not just a snapshot

it's a clone

or maybe it just becomes a clone once you activate it - not sure.

Okay, but a specific filesytem version for the kernel specifically or other filesystem paths also?

it creates a clone unless you specify the create param like default@snapname

yeah that makes sense. if it weren't a clone it wouldn't be all that useful

reason as I understand it: a snapshot is read-only. that wouldn't be particularly fun to boot from :D

man zfsconcepts :)

Thanks for the pointers.

good luck!

you might be intrigued to hear that loader semi-recently learned how to boot from a snapshot, too

jbo, Yes, sorry.

If you have two zfs pools on the same machine, is there a better way to transfer the data from one to the other than using zfs send | zfs recv? it feels like there should be a more efficient mechanism

Shouldn't native file system transfer using send|recv be expected to be the most efficient?

what do I need to install to mount NTFS?

it just feels like there'd be overhead from linearising the data on the pool just to unpack it again, when everything is already in memory in the same kernel

darwin, docs.freebsd.org/en/books/handbook/disks/#using-ntfs

Title: Chapter 20. Storage | FreeBSD Documentation Portal

thanks; I found it

what do I need to do to set 1280 2MB huge pages and enable 1GB huge pages? I've seen a script do this on GNU/Linux but don't know it'll run for FreeBSD (the script is by the xmr-stak-rx programmer)

yeah, that doesn't run on *BSD

i followed a similar page but only could enable mounting NTFS on one of the Lenovo X1 Yoga ThinkPad 20s (FR, UB) (the UB)... the other said fuse doesn't exist even though I installed fusefs-ntfs which got the dependencies

fuse: failed to open fuse device: no such file or directory

directory to mount in is there

okay, I fixed mounting the second NTFS

i'm so shocked fstab uses one SSD/M2/NVMe name and gpart uses something totally different

as a thing, how long should chromium take to start from a HDD?

I think things should be a little quicker

probably not long unless you have many tabs

nah, after a reboot for example

it has to load everything, takes about 15 - 20 seconds :/

I remember on Linux you had to tell hdparm to do some thing

and it would speed up read writes a lot on HDDs

I do not know about a similar thing in FreeBSD

Hi there :) Anyone maybe has an idea, what the minimal size for autotrim was before OpenZFS (e. g. with FreeBSD 12.4)? Currently it's 32kb (vfs.zfs.trim.extent_bytes_min: 32768)

danel1_: It was the same 32Kib back in that time.

Well, let me check, I might be wrong.

I was searching in the code but wasn't able to find something.. the only stuff i found was in cgit.freebsd.org/src/tree/sys/cddl/…fs/zfs/trim_map.c?h=releng/12.4#n79

Title: trim_map.c « zfs « fs « common « uts « opensolaris « contrib « cddl « sys - src - FreeBSD source tree

but there is no minimal limit set as far as i can see

danel1_: what is the old sysctl variable name for that?

i haven't found one..

are you sure it had auto trimming

danel1_: I have a nagging suspicion that autotrim was a new pool property in 13.0 introduced with openzfs

i am - "vfs.zfs.trim.enabled: 1"

it was just controlled by a sysctl knob

also see: forums.freebsd.org/threads/how-to-enable-autotrim-for-zfs.86091

Title: Solved - How to enable autotrim for zfs? | The FreeBSD Forums

Title: src - FreeBSD source tree

danel1_: sorry, I am just unsure it refers to automatic trimming

has anyone managed to get the freebsd 14 installer booting on the pi4 with edk2? the few times i've tried the entire system resets early on in kernel boot

made changes to pucdata.c and compiled a module. When running kldload puc.ko it fails and suggests its already loaded or in kernel. When trying to do a kldunload it complains about finding the file. Do I need to build a custom kernel to incorporate my changes in pucdata.c?

drobban:puc is built into GENERIC

oki doki. thanks for clarifying Remilia. What would the simplest way forward be, if I want to "try" my changes out.

make kernel

Will I need to make a complete build and boot the new kernel, or is there a way to compile a module and try loading it?

you can use INSTKERNNAME to give it a different name

ooh thanks

you cannot unload the kernel

hehehe =)

Naively tought I might me be able to build a "submodule" loading a "different" module doing the almost same thing but just adding my extra to it

drobban: `make -jX kernel INSTKERNNAME=kernel.custompuc` would give you /boot/kernel.custompuc which you can then use in the boot loader

X = number of jobs

that would be a bad idea in my opinion

thanks Remilia

you could also make a custom kernel configuration based on GENERIC with `nodevice puc` and then test modules

drobban: for convenience you can edit /boot/loader.conf to include something like kernels="kernel kernel.old kernel.custompuc", this will let you choose the kernel through the loader menu

kernel="somekernel" would default to kernel named somekernel

or MINIMAL, then there's every chance that it's just built as module

but getting MINIMAL to actually boot might be more fun than some people would actually like to have

ahaha

Remilia: thanks for the heads up on the -j flag =)

-j24 made some difference =D

ooh nice.. done =)

drobban: using parallel jobs sometimes can break the compilation process so if you encounter inexplicable build failures try without but last time this happened to me was like 10 years ago

o it seems to be done without errors

hopefully they will be able to choose the default kernel if stuff breaks

anyone familiar with NIS?

really need my freebsd client auth with NIS

i made linux clients connect but bsd gets rpc failure

second thing, i need to configure my tcp connectivity, i get low tcp transfers over my tun device

# ypwhich -d vlepy.nis.local 10.8.0.1

ypwhich: host is not bound to a ypmaster

# domainname

vlepy.nis.loca

l

i just can't bind my freebsd client to the nis master, the linux client i have works well

hummm. that sucked. copy pasting internet doesnt seem to be 100% solution =D

maybe ypbind ?

will have to read datasheet

nimaje i have ypbind running

i have ypbind running but it's not binding

ok

i made that working

:DD

i made ypbind to bind :)

the problem was the nis server works in a remote network that can't make broadcasting work

so i had to specify -S and -m as ypbind settings

now i have to start openvpn right after networking starts

i appreciarte any hints for that

the rc script doesn't work for you?

we would like to run ipv4-only apps on an ipv6-only freebsd (kernel) build

with an ipv6-only network that supports NAT64, obviously

(why would you do an ipv6-only kernel build?)

unixwitch, ping

nimaje it does work but it needs to run earlier, it needs to run before ypbind

so i read around on BE (boot envs) and i wanna start testing it out. but if i check freebsd-update and pkg upgrade, and there's none available, how can i test BE? because i think i have to create a BE, then do an OS or pkg upgrade, then set BE as active, then reboot and select the new BE and see the system as it was before the upgrade

bectl list

polyex: you can make a new be anytime

tried that and it said libbe_init("") failed

lol

its just integrated with the updater

yes, but you can do it yourself

polyex: that seems bad, broken, even

ya and i never typed bectl on this vm before lol. 13.2 p9

want me to try to debug it just tell me what to type

so i made a new VM and i notice when i boot it there's only 7 option in the menu. but shouldn't there be 8 where 8 is the bectl menu?

disconnected

nimaje: why not...?

it's an ipv6-only network, the ipv4 stack is unnecessary

oh after i created a BE then rebooting shows option 8

hey anyone know if RhodiumToad is ok? hasn't been around a lot

2023-12-23 20:20:17 < RhodiumToad> one of the issues for example is that curl updates very of

ten, and rebuilding curl forces a rebuild of rust, and hence all rust-dependent ports

Hasn't been *THAT* long.

fuck ya! glad he's alive and well

what all parts of hier are included in a "boot environment"?

paths

polyex42, generally, I'd say the / prefix.

so not /usr or /root?

/ has the tools to get the host booted, on the network, and filesystms mounted.

/usr has base system user utils and daemons.

by / do you mean anything under / that isn't in a subdir of /?

/root is just root's default homedir location. . . really shouldn't be part of your boot chain.

polyex42, basically yes. . . /bin, /sbin, /lib, /libexec, /etc, et al.

you said / as the tools, but those are all subdirs

so a BE is everything under / except /usr /home and /var?

FreeBSD used to default towards pushing you to make /usr, /tmp, & /var their own volumes.

Moving away from that is a big mistake, IMO.

is boot environment not defined somewhere? i didn't see it in man bectl

ya i remember that

now the default is just 1 big partition

CrtxReavr: How recently was that the case? I thought only various bits under /usr tend to not be on / lately.

polyex42, I said the "the / prefix."

and that's bad for all the old reasons that if /tmp fills up then critical dirs don't have room to work?

what's the / prefix?

/, /usr/, /usr/local/ are common prefixes on FreeBSD systems.

they're just dirs tho ya?

polyex42: varies

???

prefix is what then if not 1:1 with dirs?

Think of them as an install target. . . most *nix utils require stuff under (s)bin, etc, lib. . .

is /usr/home a common prefix on freebsd systems?

polyex42: not any more

/, /usr, & /usr/local are all the same in that they have bin, sbin, lib, et al. sub-directories.

is /home a common prefix?

/home would not be a prefix. . . it's the parent dir of user homedirs.

is /tmp a prefix?

(Though some users build prefixes inside their homedirs.

Is there a /tmp/bin/ or /tmp/lib?

ok i get it

hmm

polyex42, ever build a program manually from a tarball?

don't think so

hmm

so there's 3 levels of environments, starting with / boot env for core bringup

Well, most *nix software is distributed as a .tar.gz file (less-so in the age of git).

And in that tarball is usually a "configure" script.

and if you run 'configure --help' you'll see a list of build options.

One of the very common options is the PREFIX.

Essentially the parent directory for it to be installed to once built.

ok

and the prefix has a few common subdirs like bin sbin and lib

Yes.

it's like the mac version of how to package apps but inside out. instead of 1 single bundle, it's spread into common dirs

at 3 levels of environment

at 1 of*

so the core OS itself is just programs like any others?

like the very first .../bin? that runs

hello. when do quarterly pkg mirrors get updated?

Quarterly.

right, i mean any specific date? wondering if q1 2024 will be end of march or beggining of jan (next couple of days maybe?)

Look at the dates on the download sites.

so why aren't the exact paths that are included in the boot env documented somewhere?

unixwitch ^

sddm/ui on one of my laptops is busted and i think newest versions of libs might fix it. wondering if i should wait a few days to get newest quarterly, or if that ship sailed, than just change to latest. i can wait a bit and prefer too, but if it was updated like last week, and this thing i still busted...then might as well change to latest, update and start troubleshooting if it's not in fact a

lib version issue....

ohh it's on the boot env wiki

can you blow away EVERYTHING excluded from the boot env and still reboot the system into usable state?

Prolly depends on your definition of "usable."

usable enough to be able to rebuild the system however you can

pulling stuff over network, expanding into place

Well, to rebuild the system, you need things under /usr.

ah, found this...i guess i'll try again in a week or two: "ports branched from main at the beginning of every January, April, July, and October"

sshd is in /usr, so if you have to access the system remotely. . .

but according to the wiki, only /usr/home /usr/ports and /usr/src are excluded from BE, so /usr/bin and friends should STILL be there no?

guess you're done?

so why's it a mistake you said to move away from separate volumes?

Well. . . logs and the default location for database software go to /var

Different processes write to /tmp

You don't want those sort of things to fill your / volume.

so why did freebsd core make the decision to squish them?

I think part of that was the promotion of ZFS.

zfs can't be made to have the same behavior?

I mean, a responsible admin will make those separate volumes.

But I guess when they added ZFS support to the installer, they didn't want to also prompt for the creation of separate volumes for /tmp, /var, & /usr.

so really, any root dir that grows over time should be on its own volume right?

Part of that may also be the cattle vs. pets virtualization movement.

Single purpose, disposable VMs.

but you still want them to run right

so how exactly does it protect a system to put /tmp /var and /usr on separate volumes? those fill up and then what, instead of what would happen if they weren't on different volumes?

Well, if /tmp is full, something can still log an error that /tmp is full.

But if theyr'e all the same volume.

Plus, if you have users on the system, do you want users to fill your / (and your /tmp & /var)?

prolly not. that can cause a crash right?

if nothing else the system is basically stuck

Crashes. .. or just for the system to become unusable.

ya i'm back in your camp on separate volumes

can't remember but does installer even let us make those when going the zfs route?

so /tmp can be filled by random processes, /var can be filled with logs, but how can /usr be filled?

pkubaj, hi. scribus conficts with scribus-devel - I can create patch if u want

polyex42: filled mostly with /usr/local content aka pkg. :P

polyex42, /usr/home/ :-o

well it's /home now

so i guess CrtxReavr we should put /home on its own volume too now since it's moved out of /usr?

polyex42: re: your question about boot environments, you got some misleading information

a BE is just / and everything under it that is *not* a filesystem mountpoint otherwise

use `zfs list -o name,canmount,mountpoint` to see what is and is not under / but note that /usr by default is a special exception

it has canmount=off and is therefore included in the BE

essentially the dataset /zroot/usr is empty and used purely as a node for other datasets, including, by default, /usr/home, /usr/ports, /usr/src and /usr/obj

therefore a boot environment includes the entire base system distribution (base.txz)

(and some datasets in /var, like db, for example)

polyex42: also generally see hier(7) and note that FreeBSD itself does not have a concept of 'prefix'; it is an autotools argument that tells where to install binaries/libraries

the default on FreeBSD is /usr/local

VVD: ah, thanks, if you can please do :)

oh wow

polyex42: does it make sense to you now?

BEs are very simple by design, and these canmount=off tricks help keeping it so

also BEs are a ZFS only thing, and ZFS default in the installer is to have multiple datasets inside your pool

it kinda does. kinda overwhelmed

do you agree with having /tmp /var /usr /home in their own separate volumes in case they fill up?

they are already in separate datasets, except /usr, and /home is a symlink to /usr/home

polyex42: please note that there is nothing in /usr in the default layout besides the base system, because everything else is excluded by being a separate dataset

there is no real reason to have multiple zpools

polyex42: oh and also, you can tell ZFS to keep an amount of space reserved for a dataset

polyex42: something like `zfs set reservation=16G zroot/ROOT` though I do not recall if this will affect datasets underneath

oh it will

> That is, if a quota is set on the tank/home dataset, the total amount of disk space used by tank/home and all of its descendents cannot exceed the quota. Similarly, if tank/home is given a reservation, tank/home and all of its descendents draw from that reservation. The amount of disk space used by a dataset and all of its descendents is reported by the used property.

so the command I posted above will guarantee 16 GB of space shared between all BEs

it does not limit space, just means that out of the entire pool size, 16 GB will always be guaranteed to that dataset, and other datasets will be unable to take that space

so we can change it, but the default setup would let /tmp for example fill up the system's storage right?

yes, but the base system does not need space to start and run

it can boot and run off a read only dataset and quite a few systems administrators prefer to have the / read-only

ehh

polyex42: you can also `zfs set quota=16G zroot/tmp` to limit the size of the /tmp dataset

but you're saying there's absolutely no risk of a filled up zfs causing a system to crash or not be able to reboot?

of course there is some risk depending on what you are running

am i wrong that it seems like the default zfs setup isn't as good as it could be if it prompted to set a few quotas on datasets that can grow?

you can set, change, or remove reservations/quotas whenever you want

it is not UFS, you are not cutting a slice into hard-sized partitions

what do you think about having no swap?

also re: /root, it is included in the BE but there are no issues with that because in most cases you have nothing in there except shell configuration files and the like

and you can zfs create zroot/root any time you'd like

polyex42: I prefer having swap because my machines only have 32 GB or less RAM

do you have flash storage? i heard swap is bad to have if you got that

and you can never know ahead whether something you run will fit or not, unless this is a lab-tested, static configuration

I guess if your storage is a USB pen drive, actively swapping would be bad

also bad if your storage is eMMC

but if you are using eMMC you should have most everything read only

you need to keep in mind that FreeBSD does not use swap until there really is no RAM to use

nvme

ssd

why would swap be an issue then?

i thought if swap was enabled and you got flash storage, it'll wear it out

is that another urban legend?

my Windows desktop has 18 GB of swap space and unlike FreeBSD, Windows does use it even when there is enough RAM

well if you're swapping continuously that's tons of reads/writes over and over on the flash. they do wear out

if you are swapping continuously, you are doing something wrong

swap space is your fallback 'in case of emergency'

you don't think ppl use swap as 'free ram booster'?

that sounds weird

why would you spend *more* money on something that is *slower* than extra RAM?

(more when you consider running costs)

if you get a cheap vps for $5 and you get 20GB disk but only 1GB ram, you run 1GB swap continuously so now you have 2/19

if your normal workload requires swapping you need to add RAM

because swapping on a VPS is a major performance hit

so if swap is for emergency use, when is it enough? like eventually we run out of disk

on a vps would you disable swap or keep it but small as the emergency backup?

sorry, I do not understand your question

if you have less than 8 GB RAM you generally need swap

nvm i gotta think more

unless you have fine-tuned your set-up to stay within the available RAM

so the point of swap is not to use it, but to give yourself the capacity to use when it's needed AND to what, monitor it so you get alerted you need to add more ram?

for example, my server has 32 GB RAM and I tuned my software so that there is 4 GB RAM usually left unused

(and I have 32 GB swap space)

polyex42: in my opinion, something like that

but this applies to servers

like what, what i said?

yes

what do you use to monitor swap use and alert?

with servers, if your software configuration is static, you do not need swap

but if you use it for various different things and install/uninstall stuff often, it is good to have

I do not use anything specific, I have Prometheus+Grafana

ok

for a workstation though you always need swap

always?

because you might open Krita and load a 8 GB PNG file

or maybe do some NLE

you really do not want your window manager / DE to get OOM killed, do you

i guess not

but then you know you got a prob and just buy more ram no?

Remilia: you know, rhel8 changed swap behavior, they are now aggressively swapping. their argument "everyone is on ssd now, so we can swap more often"

i built a new server for a customer, which started swapping 30% nightly triggering alarms

so they'd rather replace ssds more often than buy more ram i guess demos?

yep, bizarre eh?

polyex42: the thing with workstations is that workloads on them are irregular unlike with servers

ya

true Remilia

ram doesn't wear out. but this whole "swapiness" thing is poorly documented... i just gave up and monitor a different statistic now

which stat you monitor now?

swapping for a task that goes above installed RAM once in a blue moon is fine

well, i use SNMP for monitoring, and on leerniupoteringnux you can use hrStorageTable and the row for "Virtual Memory"

btw Windows swaps actively, and I have an NVMe drive

one that I got quite a while ago

I never really got swapping

what is it for, really?

is anyone able to successfully build security/p5-Net-SSLeay on 14 + LibreSSL?

polyex42 so I got this 970 PRO NVMe drive in July of 2018 and it has been my system boot drive + swap drive for the Windows desktop ever since

there's an existing bug, bugs.freebsd.org/bugzilla/show_bug.cgi?id=272553, but i can't seem to get it to work with any of the proposed solutions

Title: Invalid Bug ID

with active swapping

its remaining lifespan is 95%

so I guess swapping is Really Bad

it really swaps huh?

yeah it exhausted 5% of the lifespan already

isn't 970 pro SSD not nvme?

huh

"2018" "already"

hmm

:)

polyex42: you're right samsung.com/us/computing/memory-sto…d-970-pro-nvme-m2-512gb-mz-v7p512bw

Title: SSD 970 PRO NVMe M.2 512GB Memory & Storage - MZ-V7P512BW | Samsung US

5% over 6 years is what, 120 years?

'970 PRO NVMe® M.2 SSD 512GB'

it's not

huh ok

is it shaped like a stick of ram, or a small HD? like 3.5"

or i guess 2.5

it is shaped like an m.2 2280

ok a stick, it is nvme then

ok sweet good to know it doesn't get destroyed by swapping

the Samsung website does not know anything about non-NVMe 970 PRO

or 970 in general

870 is SATA though

I think 9xx are NVMe and 8xx are SATA

do you prefer shallow or deep BE's and why?

I am going to leave this question unanswered because you can figure this out yourself

huh?

deep BEs include /usr/local

like i can make my own opinion eventually but i wonder what yours is

hmm

so heavier to create new BE's i guess

it is not about 'heavier'

well it just uses more disk right?

use deep BEs if you want to keep your entire package set with the BE

note that in this case you really should also keep database files etc. inside the BE as well

it's kinda like a thin vs thick jail

you do not want to run an old version of, say, MariaDB with a datadir that was upgraded

are shallow BE's even useful? because what if you rollback a bad freebsd-update, but your pkgs are still borked?

ya exactly

exactly what?

mismatches in package versions and stuff

shallow BEs are useful because they let you have a base system to fall back on if shit hits the fan

and that tends to happen before you upgrade all packages

ok so they're not really useful for knowing 'i can rollback my system to a known fully good state' it's more 'i can at least get back to a bootable state to pull data off' ?

plus packages are built for every -RELEASE

just reinstall them with the old BE

so they're not really useful for knowing 'i can rollback my system to a known fully good state' it's more 'i can at least get back to a bootable state to pull data off' ?

Remilia, you're hinting at pkg upgrade -f ? :D

Hi all! Is there any update in FreeBSD 14 regarding WiFi support? I am looking if there is any advance in supporting the WiFi chip in the Raspberry Pi.

Ronis_BR: anything like that would be in the release notes

Title: FreeBSD 14.0-RELEASE Release Notes | The FreeBSD Project

also wiki.freebsd.org/arm/Raspberry%20Pi

Title: arm/Raspberry Pi - FreeBSD Wiki

Broadcom usually means you are out of luck

yuripv: nice find regarding lang/hs-koka. It looks very interesting :)

Anyone running 14 on i386? Any hiccups?

I haven't seen a non-for-fun i386 system in like 10 years :o

So make that 1.

I have 14 on a i386 (Intel Atom, 32 bit) netbook. Seems fine - performance is as expected for the hardware.

I guess ports is really the only issue with i386 right now?

14 i386 is 2nd tier right?

FreeBSD maintains support for i386 (x86) as a Tier 2 architecture. It is not recommended for new installations.

Support for this architecture will be removed in FreeBSD 15.0, so all users and administrators are strongly recommended to upgrade their installations to AMD64.

yeah

I don't even test package builds on i386 anymore

It is. According to the 14.0R release notes, Ports will not include support for deprecated 32-bit platforms for FreeBSD 15.0 and later releases.

I have an i386 machine but I do not think it can run anything above 7.0

it is a Pentium 75 with 16 MB RAM

This netbook has a minimal set of installed packages, but everything did install fine when I set it up.

at least 2024 will be the year of the linux desktop </SCNR>

Pentium 1 can support 128MB RAM

that board cannot support more than 32 iirc and finding SIMMs is another challenge

also would need to find an SCSI HDD because of the BIOS's 4 GB ATA limit

SIMMs heh

I have 80486DX2-66@80MHz with 2x16MB SIMM 72pin

It's turned off for decade…

I could probably dig out a PC9801

VM

with two 5.25" drives

Remilia: thanks!

jbo: referring to that awkward freebsd forum post? :P

pkubaj, reviews.freebsd.org/D43276

Title: ⚙ D43276 print/scribus{,-devel}: add CONFLICTS_INSTALL each other

does pkg upgrade create a new zfs boot env too or just freebsd-update?

polyex42: pkg does not do that

k ty

Nor should it need to

is freebsd-update the only part of freebsd that automatically creates boot envs?

polyex42: I'm not aware of any other things doing that

ty!

polyex42: pkg should have no need to do it because pkgs shouldn't affect the ability to boot

ok ya all the more reason shallow jails feel like the right choice

there any way to choose which version of a pkg we want installed?

not really, the repo only contains one version at a time

polyex42: You can build it yourself, and in doing so, use an older version of the port

would be better if we could pkg install nginx⊙6

polyex42: the package repositories already take days to build and loads of storage to host

though I understand why this might not seem like a problem to the end user

that's the wrong angle. as an end user, it sucks

I mean 'hosting every version of a package'

not 'not hosting every version'

if a pkg is upgraded in a round of pkg upgrade, but it's buggy, there's no way to roll back to the prior version that was working fine

sure is

just rebuild you say, sure, but for ppl who are strictly pkg binary basis, it's a big hassle to ask

that's all im saying

it is in /var/cache/pkg

well maybe host latest 3 versions? just something other than latest or nothing

the old pkg is in /var/cache/pkg?

so you mean 'thrice the storage'

ok then, latest 2 versions and twice the storage, but ya, that's what i mean

have you looked at your /var/cache/pkg?

oh shit

what now

so how would pkg foo be downgraded to 1 of the older versions in my cache?

well my point still stands for new installs that don't have that

pkg install /var/cache/pkg/foo-...?

so do you uninstall the upgraded but buggy port then install the old 1 from cache?

you can just use install

s/port/pkg

it will uninstall the currently installed package first

whoa

ok well that goes a long way

how often, if ever, do you manually create a boot env?

or do you just let freebsd-update do it automatically and that's it?

lately i've just been letting freebsd-update do it