andyf how far are we about switching to gcc 10?

I think we could do it. OmniOS has already switched and the only thing that has showed up as a problem since is the boot cpuid thing you already fixed. There are probably other issues but we won't find them without switching.

There's one issue I had marked as a pre-requisite, but having gone through it I don't think it is, let me dig it out

illumos 14421 (fenix) is the master issue

FEATURE 14421: use GCC 10 as default primary compiler (In Progress)

↳ illumos.org/issues/14421

and illumos 14749 (fenix) is the last piece I was looking at

FEATURE 14749: gcc10 build should use -ftoplevel-reorder (New)

↳ illumos.org/issues/14749

The thing is that gcc10 does not strip unused static functions by default, unless you use -ftoplevel-reorder, and if you use that then you lose a load of global statics which are only read too (due to constant folding)

So I played around with different optimisation options etc. but in the end I don't think we want to strip all of the static global variables

We definitely should go through and remove the static qualifier from any which are supposed to be tunables anyway, but not as a blocker for switching to gcc10.

ping tsoome

Testing the upstream merge which includes 14970 seems to break nfs-server-in-a-zone for TWO of my three zones.

The workaround is to set mountd_remote_dump to true, which seems to cure the problem.

breaks how?

nfs/server goes into maintenance.

I see one possible problem that might be specific to SmartOS. Diving into it now.

does it tell why it cant get online?

What's WEIRD is that ONE of my nfs-serving zones isn't affected when mountd_remote_dump is false.

No.

mountd just returns '1'.

ok, if you get name from hostname commands, does it resolve to ip?

this code does add 2 cases when we exit(1) -- from gethostname() and from getaddrinfo()

Hang on.

WOW! Okay, this might be operator error on my part. Please hold...

we also do: syslog(LOG_ERR, "gethostname() failed"); and syslog(LOG_ERR, "getaddrinfo() failed");

Sonofabitch, that's it.

So my two failing ones were build2 and build3. Their /etc/nodename values were build-2 and build-3, but those don't have DNS entries.

Fix the DNS, fix the problem. Alternatively I could have fixed /etc/nodename to be build2 and build3.

Okay, I think I'm set, and I've bonus-tested your new feature. :)

(The one that worked already was "build" )

The background is that nfs related commands do rely on hostname having ip address -- see showmount(8) for example. Thats why I did add check for IP for hostname. Altho, maybe we should ignore failing getaddrinfo()...

I see them:

/var/adm/messages:2022-09-22T11:10:47+00:00 smartos-build-3 mountd[11959]: [ID 115315 user.error] getaddrinfo() failed

Okay, crisis averted, no need to seize up the release. Thank you tsoome!

you are welcome

(I was looking for them in syslog not messages.)

it was nice from Delphix to write test for btree, but why bugs in it....

fortunately smatch did find those.

that could be turned into a template: "it was nice from <insert-name> to write <insert-product>, but why bugs in it" :D

:D

I hope it is not c++;)

or /* XXX: how this even work? */

It was working, just leaking a bit;)

did anyone here ever run illumos with X on a thinkpad w530 with nvidia k2000m gpu?

Woodstock: I successfully ran OpenIndiana on a Thinkpad T530, I can't recall the specific nvidia chip it has but might be able to check later today.

mini-survey: how many folks' build pipelines would be inconvenienced if mail_msg was instead named mail_msg.txt (to make for easier handling in webmail clients and the like)

pmooney does not matter for me:)

.txt is fine

I'll have to look... would appreciate a diff so I can test it on a branch build, however.

Looks like I'd need to change a line in smartos-live:

tools/build_illumos: mail_msg=$PWD/log/$log_dir/mail_msg

I'm generally in favor of mail_msg.txt, however, given Mail.app's historical munging of vanilla mail_msg.

It's probably easier than getting people to set their MIME types properly when sending in RTI emails..

There's tooling that would need adjusting in several places in OmniOS but it's all possible

andyf: that presumes a mail client which even lets you set the mime type

^^^

(and one which heeds it on the receiving end)

Also ^^^

I think mail_msg.txt would be nice, I think the last few times I sent an RTI I added .txt extension

If we're going to rename it, probably "summary.txt" would make more sense?

also we should do illumos.org/issues/14516

→ BUG 14516: build summary should have an explicit success/failure message (New)

what if the result is at the bottom?

(probably easier to implement)

rather than having to cat it again, I assume

We already do that for the full build time

mm, true

I think it being at the top is useful

I guess another question is: should it still be "sending" email

ha ha

Probably not

as someone who has "mail<cr>d<cr>" in muscle memory

no mailx? :D

But yes, it seems somewhat anachronistic to send mail by default

well, I guess that's flag-controlled

time to PR a change to the defaults shipped by omnios, haha

[illumos-gate] 14485 bhyve needs better cpuid control -- Patrick Mooney <pmooney⊙pc>

I like all these ideas

mail is often unimplemented by the local admin these days when all the world is a firewall

I have nginx working but i use dyndns so no way to get email on my domains and i am savvy enough to own some

gigabit fiber is $65 here with random IP addresses and $300 for a /24

so i have a great connection but my servers need dynamic ip, i can't afford static... perhaps a dynamic ip redirector should be part of core?

and sorry but it's time for project pitchfork

at any rate i strongly prefer illumos for a server, my raspi running linux is ok but...

[illumos-gate] 14997 unnecessary jmp in desctbls_asm.s -- Dan Cross <cross⊙oc>

pmooney: The thing with the flag is, will we still generate the mail_msg if we're not mailing?

That's why I have never tried turning it off haha

jclulow: it appears that the flag only controls sending of the mail, not generation of the mail_msg

ah good

indeed

why would you not have a mailer configured even if mailto:root is effectively /dev/null?

i wish to understand your reasoning

There is really no reason to have a mailer running at all if you are not intending to send mail or read mail

uhm, it's traditional?

If you go back far enough, panicking instead of returning EIO is traditional

mailto:root is for stuff you would rather ignore oif lazy :)

indeed

But sendmail is a security and configuration nightmare, and the better mailers are all similarly complex -- except for the ones that basically just use a smarthost unconditionally, but then for those, you need a smarthost

It's really hard to send SMTP mail in 2022 without a bunch of ongoing work

sendmail is indeed a nightmare, i usually use whatever is the current smail3 derivative

The only thing on the system that routinely sends mail is cron, and we could instead provide an option to write that into a tree of log files and then there would be no requirement

sorry, listening to ilatian/german musik

ilatian, is that a new word, it's mo-Do

who actually reads log files?

i know you should

but who does in 2022?

our unix boxes are mostly toys by now

please send me a deprecating answer

I read them when something breaks

i call here when something breaks and usually someone like danmcd has it fixed in under 30 minutes :) thanks danmcd

i try to fix it myself but you guys are too cool to let me

and yes you are supposed to chuckle at that

someone even taught me to use the kernel debugger which could seriously use some symbolic features

It has symbolic features, it just doesn't have source-level debugging features

i can read assembly but random 64 bit addresses are odd'

jclulow i am not insulting

i should be able to fix it

Most of the addresses you see probably don't _have_ a symbol

They're often the address of a stack frame or something from kmem_alloc() etc

i have not seen that much JIT but i digress

anyway i don't understand at all how you get away with such few locking structures

and i must rest for q5h

i am old and we had to lock every object before use

anyone recently run zfs-tests?

my attempt is stuck with zfs hung due to testpool with faulted device or somesuch. time to trigger panic and get dump....

I've run some, but on SmartOS so there are ones that get skipped that might not get skipped in upstream illumos, etc. etc.

understandable.

Which test is hanging for you tsoome?

last one in log was setting the gzip levels, but I'll try to get some bits out from dump once I get the update done:)

> It has symbolic features, it just doesn't have source-level debugging features

jclulow: what if Robert reads this while on the plane?

Fortunately he is I think watching a movie and trying to sleep

Robert you enjoy to have a peaceful vacation :)

I hope he isn't reading this on a plane!

What does the last line in your log say? I can see how it corresponds to my didn't-panic run (and determine if it's one that got disabled or not).

This is why I used this full name, hopefully he doesn't alert on that, haha

you know drobbins (funtoo) stopped trusting zfs like 2 years ago... :)

there was a data loss bug iirc and as if btrfs is better

and it could have been the linux porting team's fault so who knows, stuff breaks

00000000

oops

now you know my client uses insecure passwords

nice.. wasn't too bad to add mDNS support to smbd (right now just advertising _smb._tcp)

now to deal with shares..

Woodstock: my T530 has a different GPU: NVS 5400M though FWIW it works well on OI