grr.. figures the day after my UPS battery dies the power starts flickering (from storms)

[illumos-gate] 16601 Remove CERRWARN overrides from aggr(4D) -- Dan McDonald <danmcd⊙mi>

we did bump usernames/groups to 32 characters, right?

illumos 2989 (fenix)

BUG 2989: LOGNAME_MAX should be increased to 32 (Resolved)

↳ illumos.org/issues/2989

Looks like it

ok..i couldn't find it

probably should add mention of that to the ticket i just filed as well

[illumos-gate] 4467 kadmind: segfault in acquire_accept_cred+0x140() -- Dale Ghent <daleg⊙oc>

this patch did took its good time to mature:P

Anyone see those new AMD Epyc 4004 series? Look like they wanna kneecap Xeon E.

no, I'm sticking to XEON for now, the used ones are way better supported and cheaper

for example I upgrade two X4-2 with E5-2680v2 where the CPUs were ~35EUR each

When I say "Xeon E" , I mean the modern ones that are equivalent to the "E3" of old.

ah

sorry :)

danmcd: I'm also still on sparc, so like 10y behind everyone else ;)

But you're not wrong about the used market being quite the opportunity. Now granted, speaking of kneecapped, some of those will need modern ucode and OS to mitigate some of the spectre/meltdown damage, but also some of those ones are immune to other damage introduced in Skylake times.

in the end it's always a thing were you have to decide which risks you want to accept

v2 of my home server was on a Low-Powered Xeon E3 (1265L v3), its replacement HDC v3 (which I run now), is a Ryzen 3700X.

I'm thinking what I do and don't need when/if I jump my home sever to a brand new v4 (which at my upgrade schedule can happen in 2026).

The Epyc 4004 seems to be targetting me. Though if I'm being honest, current setup is QUITE nice, and may benefit not from full replacement but a series of upgrades.

The 20C/40T X4-2 boxes I have run at 200W and I run a whole OpenShift Cluster on each of them...

I'll very likely going to run these servers for another 5 years

10GigE that supports 2.5/5BaseT (would love 4x X550s e.g.), and a better case (I chose poorly for HDC 3, my own fault), e.g.

Since 2008 I've had three home servers, each in HDC use 6 years. HDC 2 (the Xeon E3-1265L) is now a Triton Head Node for my at-home test cloud.

danmcd: do you run just a headnode or a bunch of boxes? bonus question: how much ram per node?

I have four nodes: the 32GB RAM Xeon E3 is the headnode. The rest are 128GB RAM used 1U 2-socket Haswell-E (Xeon E3-2xxx v3) servers. They're linked up with Netgear blue GigE switches (one for admin, one for external, one for underlay).

Oops, Xeon E5-2xxx v3... typo.

nice

Useful for observing real-world interactions.

I have more compute than I need for sure (and note: the CNs in Kebecloud are owned by MNX).

yeah, that's why I run OCP at home...I train people on it. and my labs are more flexible than the training environment

danmcd: eenfach.de/~olbohlen/serverrack_front.jpeg and eenfach.de/~olbohlen/serverrack_rear.jpeg

Ooof. I don't even have a proper rack. :)

I have some more room, so I could set that up ;)

I *do* wish (and I keep banging this drum) there was cheaper 10GigEBaseT switches, akin to the GigE classic of Netgear's blue GS1xx series.

I may have to "settle" for 2.5 or 5 in that arena.

danmcd: I think I told you already which netgears I have, right? 24x1G copper + 4xSFP+

(Problem is with old/used equipment, the 10G NICs on there are 1G or 10G only, no 2.5 or 5)

I'm still searching for a 16xSFP+ switch

1G copper is the "problem" I wanna solve. I mean, multiple GigE and faster uplink is nice, and if FiOS delivers > gigE to my house I'll have to consider that.

and the Silkworm has "just" 4Gbit, but that's way enough for home

I mean...who has a SAN fab at home anyway

is it just me or is it broken: src.illumos.org/source/xref/illumos…i/e1000_mac.c?r=49b78600?r=49b78600

that....is empty?

so, it is broken:)

itis broken

I've pinged some folks about it already before this popped up.

it's the second parameter

remove the second ?r=xxxx

at least as a workaround

The double r=xxx is likely the problem.

Good catch @jbk

if opengrok is generating those, it's a problem obviously.. but at least you can workaround it to get whatever info you need

on a completely different subject....

Yes?

danmcd: I just picked up (so recently that I haven't yet powered on) a Mikrotik 12-port switch (8 10GbaseT, 4 combo SFP+/10GbaseT). I'll let you know how it works.

i'm contemplating putting together an IPD if people think there's merit to this -- when dealing with nested filesystems on zfs where an ancestor is shared out (but not the children individually)... we've run into a few situations where somehow clients are able to get files into the mountpoint directory on the parent filesystem

I've heard BAD THINGS about Mikrotik's horrid software. OTOH if it's Just A Switch and not in a management path, maybe it's not so bad.

(keep going jbk...)

we think smb or nfs (probably smb) client has the parent dir 'open' (being a bit loose here)

and when the child gets mounted, it's still able to write to the parent dir instead of the child

my thought was to add a 'mountpoint' attribute that could be set on a directory

that would prevent most operations unless explicitly cleared

since I've also personally seen where sometimes if a child fails to mount for whatever reason, you can end up with files in the mountpoint dir

which if you don't notice, tends to bite you in the butt later when you try to mount

jbk: oh, as in something that persists when nothing is mounted on top?

that way, even if it isn't mounted, writing stuff to it would fail

yeah

just like we can mark a file as archive/immutable/appendonly/etc. with attributes

add one that can be set on directories

"immutable" on a directory wouldn't be far from what you want.

it wouldn't be required to mount a filesystem somewhere, but on directories marked as such, it would basically restrict what could happen

since the root vnode of the filesystem 'replaces'/overlays the mountpoint vnode, once mounted, everything would work as is...

i haven't looked at hte specifics of immutable if that would work, just been throwing around ideas since it's something that's been hit in various ways

BTW are you running with SMB traverse_mounts set to false?

no

i don't think though that SMB re-validates the path every time

so if it's holding the directory vnode on the parent, and then the child is mounted

i'm not sure it's going to update it's state to now look at the mounted root vnode at the same spot

i thought you could maybe flag the vnode of the dir on the parent once something's mounted, but then i've had similar issues just because some child dataset didn't mount for whatever reason (e.g. missing passphrase), and you end up with data in there as well, so was leaning towards something more presistent (but obviously could be turned off as desired)

although.. the description of immutable (which only appears with ls(1) and not chmod.. might be worth a bug on that) it might work to just mark it immutable...

i'll need to test that

danmcd: re: mikrotik: from reviewing their docs, a lot of linux quirks show through. this is going to be a pure-L2 switch not doing anything more exotic than hosting a few vlans, with its only ip address on my not-routed mgt net.

Yeah... I'm shooting for pure L2 switch capabilities as well.

jbk: I suspect immutable is indeed what you want. Someone set it on some files in a shared directory somehow recently and it took me a while to figure out why I couldn't do anything with them haha

I could imagine an enhancement though where ZFS would do something with the immutable bit when it's arranging nested datasets

yeah, i was thinking that

[illumos-gate] 16567 clean up yacc grammar -- Toomas Soome <tsoome⊙mc>