it was indeed an issue with 2TB+, maybe the new data was in the 2TB+ area

user decided to add a small NVMe for boot.

tsoome_ thank you very much for your help.

I'm acting as a bridge between discord and IRC questions and knowledge bases :))

i thought a bridge displayed something in brackets on each message. e.g. [Discord user #123] hello there

FYI there seems to be a public log of this channel. I'll not disclose the URL.

how to make sure if the gpu driver is loaded or not?

ghodawalaaman, what's the GPU?

grahamperrin: it's skylake (in-built GPU)

ghodawalaaman, kldstat | grep i915kms

grahamperrin: it returned nothing

here is the kldstat output: termbin.com/ud89

Which version of FreeBSD, exactly? Packages of ports from latest, or quarterly?

it's 14.2-RELEASE

freebsd-version -kru ; uname -aKU ; pkg -vv | grep -B 1 -e url -e priority

OK, AMD64.

pkg query '%o %v %At:%Av' drm-61-kmod

I have no idea how new or old Skylake is, I guess it's Intel

that command returned nothing

antranigv you are welcome.

yes it's intel

pciconf -lv | grep -B 3 -A 1 display

a

<bsd-hardware.info/?probe=adcbf8a526#pci:8086-1912-1028-07a3> worked with i915 with 14.0-RELEASE-p5, so

you should install two things:

a) drm-kmod

then

I have installed drm-kmod, it was pretty big package

I also added kld_list="i915kms" in /etc/rc.conf

pkg add pkg.freebsd.org/FreeBSD:14:amd64/km…ll/drm-61-kmod-6.1.92.1402000_3.pkg

the most recent version of drm-61-kmod-6.1.92 is already installed

pkg add -f pkg.freebsd.org/FreeBSD:14:amd64/km…ll/drm-61-kmod-6.1.92.1402000_3.pkg

done

Now restart the OS, hopefully you'll see a flicker when the module loads. With or without a flicker:

kldstat | grep i915kms

ohk, thanks

BBL, maybe. Bye.

Hello, I want to just add a message saying "make sure GPU driver is correctly installed" in sway/wayland section. how would I edit the handbook?

also this link on the handbook is broken (github.com/freebsd/freebsd-doc/blob…ain/documentation/content/en/_index)

I have to make a Github account to contribute to the documentation.

ghodawalaaman: you don't have to do that, you can create a bugzilla account and send patches/issues there

Folks, I have a iSCSI server (NetApp), I am the iSCSI client (FreeBSD), the block device is formatted as EXT4. I can connect to iSCSI, I see the device, I can mount EXT4, but it's VERYYYYYYY SLOWWWW. Thoughts?

dch: sending patches this way is also new to me but I will try doing this way

antranigv: you're doing it via fuse I guess?

ghodawalaaman: no worries, feel free to ask here for help any time

dch mount -t ext2fs. I think that's *not* fuse.

aah yes

ghodawalaaman thank you for contributing.

antranigv: I would compare throughput of unmounted block device with dd .. vs mounted and look a bit at tcpdump in case there's any stupid MTU or similar things obviously not right

but I guess its probably just not a very optimised implementation on freebsd

dch I'll have a look with unmounted dd if=/dev/mydevice right now. If that's the case then I can copy the iscsi over dd, and then boot a linux vm for the ext parts

thanks for the tip uncle dch <3

;-)

can you just pass the iscsi device through to a linux vm directly?

why is data deep-copied when you move it from one data set to another on the same pool?

with dd I'm getting 10MB/s, but I have a 10Gbps link, and iperf shows 9.9Gbps and disk in dmesg shows 150MB/s for the speed.

faster than EXT4 mount, slower than what's needed.

jemius because datasets are indeed different filesystems. they start at inode 0, etc, basically a different filesystem indeed.

makes sense

could someone verify this pull request?

I have just made the pull request for adding a warning while installing sway

antranigv: 150MB/s is not far off a single NAS-level SATA disk throughput. whats the backing vdevs on the netap?

so is it common pracitce on FreeBSD/ZFS to give each user his own dataset as home directory? I guess it's handy for example for limiting maximum homedir size

jemius: very common, not just for quotas. some users have multiple datasets, with different retention or replication (snapshot) policies, or different performance / tuning settings

I see.

Defragmentation is not a thing on ZFS, is it?

zfs datasets are awesome

if i have zroot on mirror and physically remove one drive before powering up, will it boot normally?

(both drives have the bootcode)

futune: yes, as long as your boot environment (EFI partitions, etc.) are redundant -- FreEBSD doesn't set that up automatically

the pool will just import in a degraded state

hmm, that is true, and they are, but there is also a non zfs geom mirrored swap partition

will that assemble in a degraded state too? or halt the boot in some way if it fails?

and as for the zfs part, if i then power down afterwards, reconnect the drive, and power up, will it be easy to resilver it again?

i would strongly recommend not putting zfs on a geom mirror, because geom doesn't understand checksums, which means ZFS cannot fix any data corruption in the mirror

no, no, zfs is not on a geom mirror

the swap partition is, because i dont want swap on zfs

ah, i see what you mean

yes, as far as i know, gmirror will still come up if one side is missing, although i've never actually tried that -- so i don't know how you recover that off hand

(from what i remember though, you just use a gmirror command to overwrite one side of the mirror with the other)

that seems fine, as long as i can recover - its just swap, no permanent data

right, mirrored swap is just about having the system not crash if a root disk dies, but it should still boot if a disk is missing

i have done this extensively on Solaris but sadly not so much on FreeBSD

yeah exactly

i am mainly worried about the zfs resilver after putting the drive back

that should be pretty straightforward, resilver does not interrupt service at all other than using a few iops

awesome, thank you ivy

although, "a few iops" can actually be quite a hit to performance -- if that's a concern there are sysctls you can tune to adjust how fast the resilver runs

if it's just the root being slow it should be fine for me

this whole adventure is actually about replacing disks in a much larger array, but I don't have enough ports, and I don't want to temporarily degrade the critical data

dch I am okay with the iSCSI showing itself as 150MB/s, but I still get 10MB/s when actually doing dd :D

antranigv: maaybe check block size of dd vs whatever the network can support?

sounds like a mess anyway

if you set copies > 1, does the file system then ensure that the redundant blocks are spread across the partition? I suppose it does, because having them next to each other would be very dumb

dch indeed. we're moving to the NetApp to the other supercomputer datacenter, so I need to take of my files. My FreeBSD and NetApp are 1KM afar, but the fiber has been giving us stable ~10Gbps.

jemius: as far as i know, yes, but it's not as useful as you would think -- if you have two striped vdevs with copies=2, and one breaks, you can no longer import the pool at all

or in other words, copies is not a way to just mirror certain filesystems

ivy, what does "two striped vdevs" mean?

antranigv: I expect its the latency then. can you try the perf at the local DC for comparison?

jemius: nobody relies on copies>1 in practice. best to use mirrors or raidz combos, drives tend to fail in unexpected wats

I had bit errors on certain sectors on a dying drive once.

Anyways, why provide a feature that is not useful

jemius: a vdev is the lowest level of of zfs building blocks - it can be a physical block device like a disk partition

you can stripe 2 vdevs together, which yields a faster vdev but with less redundancy

you can mirror 2 vdevs, which results in redundancy but slower writes, and faster reads

etc etc

You're simply talking about how you can organize your pool. A RAID of mirrors and so on

jemius: the feature is useful, just only in very specific circumstances -- it's intended for the situation where you have, say, a laptop with a single disk, and you want to protect against some data being corrupted by bad blocks

"two striped vdevs" mean a zfs pool configured as a stripe across two disks (akin to a raid0) -- some people think they can use copies=2 in this situation to provide redundancy, but you cannot

Ah. No no, I'm not doing that. I have a 2-mirror

I have four MBR partitions, in the first one a 36G-sized FreeBSD partition (created with Fedora's fdisk), then I created the BSD partition scheme in that partition with "gpart create -s bsd" but when I try to add the slices with "gpart -t freebsd -s 10G" I get "gpart: invalid argument", why?

asarch: what is the exact command that fails ?

Well, it is: "gpart -t freebsd -s 10G" but it seems because I did not add the MBR bootcode to the partition with "gpart bootcode -b /boot/mbr"

...first

how would gpart know where to add this slice ?

moreover there's no type "freebsd"

asarch: take a look at this nice gpart primer by W. Block wonkity.com/~wblock/docs/html/disksetup.html

I mean, freebsd-ufs

"gpart -t freebsd -s 10G" - the syntax is wrong, please take a look at gpart(8)

Bingo!

Thank you!

asarch: I would not normally use BSD partition type these days, just `-s gpt`

Thank you very much! :-)

I know, but actually is an old PC

theres nothing wrong with it per se, just gpt is basically standard on all OS now

make sense

With no UEFI support at all

asarch: UEFI and GPT partition types are IIRC orthogonal, but its if its a very old box then that makes sense

A very old HP box

It is still alive

yes, using GPT if you have no MS Windows there will be best, both FreeBSD and Linux will boot fine

but if the disk has already MBR laout and place for one primary partition for FreeBSD, they follow old MBR/BSD partiotioning scheme

s/partition/slice