[illumos-gate] 16233 pcieadm readiness time reporting decoding -- Robert Mustacchi <rm⊙fo>

[illumos-gate] 7125 Provide execvpe -- Robert Mustacchi <rm⊙fo>

Which usb wifi dongle does illumos have support for? I have a realtek card and it obviously doesnt have drivers for it so I need an external one for my laptop

CutieMelo have no idea really, sorry.

okay

Hello! I will be giving an illumos talk today at a local event... if anyone can check my slides for mistakes, will be very happy :) antranigv.am/misc/OmniOS.pdf

I will also try to record the talk, altho it will probably be in Armenian, depending on the audience.

You could mention LibreOffice (ex-OpenOffice, ex-StarOffice) as another thing to have come, albeit indirectly, from Sun

Any chance this can get attention again? illumos.org/issues/15024

→ BUG 15024: NFS can exhaust pool threads getting RPCSEC_GSS credentials (In Progress) | code.illumos.org/c/illumos-gate/+/2402

we had this issue yet again today where the machine was basically non-responsive, with this error message: /usr/lib/nfs/nfsd[771]: [ID 791759 daemon.error] t_rcvrel(file descriptor 552/transport tcp) Resource temporarily unavailable

hrm.. i bet matt forgot about it... i can ping him after this meeting

oh, these thread names in prstat -L are fancy. I don't remember this. Is this new(ish)?

2018

which for the illumos code base, could still qualify as 'new' (relatively speaking) :)

the whole change was started because someone (I think I remember who) was whining about how smartos didn't have it

well it is _np

honestly I probably just never ran things that used it, I was just watching the build with prstat -L and noticed it for some binaries. Not sure how useful it is but it looks cool.

i can't recall which ones jlevon added, but for the bits I did, I tried to pick things where I thought it would be useful to be able to tell what was going on

It's been useful for bhyve processes

rzezeski: whenever you have a number of otherwise identical worker threads it's very handy

good to know

rzezeski: similar motivation for ptree -g (whining)

hmm, ptree -g doesn't use the box characters for me as I use a hybrid locale (mix of C and C.UTF-8)

setlocale returns something like "C.UTF-8/C.UTF-8/C.UTF-8/C/C.UTF-8/C.UTF-8"

ptree.c should probably use LC_MESSAGES rather than LC_ALL ..

(I set LC_COLLATE=C because I don't like how LC_COLLATE=C.UTF-8 sorts things)

MESSAGES doesn't feel right there, tbh.

As I thought that was just about translation and not what you can accept, but hard to say.

Perhaps "if (p != NULL && (p[5] == '\0') || (p[5] == '/'))"

intent: if any locale component ends in UTF-8, we assume you can cope with UTF-8 output on stdout.

parens not right in the above but you get the idea

sqlite3 just sends the box characters if you say .mode box without doing any other checks.

maaaybe LC_CTYPE since that determines (among other things) what is printable? I think LC_ALL was just the more conservative choice at the time

jbk: Yeah, I guess if it's not printable we wouldn't want to. So maybe that's the least bad option.

tozhu left, but their NVMeoF vs NVMe/TCP vs iSCSI results made me think they hadn't tuned any TCP/IP settings under the Linux kernel

is it possible to run a command with memory-limit? e.g. ctrun -o max-virtuelmem=1G thisprogram ?

How about ulimit -v ?

(The snag there being that it applies to the current shell, so you can't set it too low or that gets killed off.)

we don't support rctl's to -o no

but supporting rctls like that somewhere sounds great

for contracts

I don't think we have virtual memory limits at contract/task/project scope - only zone and process scope if I'm reading correctly

I used to run firefox with a memory limit (ulimit) but then it went multiprocess and that became ineffective...

hmm... dmake: Warning: Command failed for target `pics/ctf_create.o'

I wonder if I missed something from the building docs

trying again

antranigv: the useful message is usually the one before the first instance of "*** Error code"

`less '+/Error code' log/latest/nightly.log` and then scroll up a bit :)

thank you all! At the moment I test smth like "unlimit -v 100000"

something like the ctrun with the usual resource controles from prctl/projects/... would be very neat. Doesn't need the workaround of defining projects and such. Just put in your script or crontab ctrun -o mem=1g thisprogram

it could be wrapper that creates a random project on the fly and spawns the badly behaving program just into that context

[illumos-gate] 16246 ttymon: tty-mode line is not read when terminal name is set -- Toomas Soome <tsoome⊙mc>

Turn out here's the issue... dmake: Warning: Command failed for target `smatch.o'

but I have no idea where that target comes from

here's the log file if anyone's interested: build0.illumos.am/log/log.2024-02-08.20:54/nightly.log

this is the reason: sh: /usr/gcc/10/bin/gcc: not found [No such file or directory]

huh.....

I thought illumos-tools from OmniOS would download that automatically

if you are building in omnios, it is in /opt/gcc-10

ahaaaaa

your env file is to blame

you need omnios-illumos

I just copied the default one from illumos-gate... I'll go over the docs again, maybe I missed that step :))

for now I'll modify the illumos.sh file manually as much as I can, just to learn more about it.

Yes, you need the file in /opt/onbld/env/omnios-illumos-gate or one of the ones from downloads.omnios.org/env

The one in gate is pretty much set up for OpenIndiana as far as I remember

It's close but not quite right for OpenIndiana

I have to set PKGVERS_BRANCH and several PERL_* variables.

I should probably dust off illumos.org/rb/r/2360

BTW, stumbled across a bit of an oddity in our terminal code which trips up Python. github.com/illumos/illumos-gate/bla…/src/uts/common/io/ptem.c#L936-L943

you can't get the window size of a freshly created pty until you set it once.

Ah, that rings a bell

and python refuses to set the window size until it gets it once.

I think the linked lines should just be purged.

(to be more precise, you get an EINVAL if all four dimensions are zero)

I fixed it with github.com/Bill-Sommerfeld/oi-userl…hes/50-termios-set-before-get.patch

I prefer yours, and yes, those lines do look rodd.

and odd too

how long does it take to make a full build... usually? I have a pretty modern system (24 cores, 46GB of RAM, AMD EPYC 77something)

53 minutes was the last one I did - that's both non-DEBUG and DEBUG, and using both shadow compilers.

I have a system with 4 cores/8 hyperthreads and 24G and it takes 2+ hours with shadow compilation and both DEBUG & non-DEBUG

Looking at the build logs (mail_msg) that people submit along with changes to the illumos.topicbox.com/groups/advocates is quite useful

RTI builds are supposed to be debug+non-debug, on my system its about 50 minutes

illumos.topicbox.com/groups/advocat…m-readiness-time-reporting-decoding was 24 minutes

I'm overdue for an upgrade :-)

==== Total build time ==== real 0:35:00

Not sure if this was DEBUG or not... have to check the default env file...

The mail_msg file under log/latest/ will have the details

and show you which shadow compilers were used too

I should get a package dir... no? I don't have it

unless there was a build error, yes, mail_msg will tell you

I see

I also see errors

ah... perl :))

i really need to pick back up my 'theperlthe' branch

[illumos-gate] 16226 yacc: remove WSFMT macro -- Toomas Soome <tsoome⊙mc>

Now this is new for me. After running nightly.sh, I see that a large number of files are untracked (according to git status), is that normal or am I doing something wrong?

ah, I see, untracked... not modified

yes, that's normal

they are the generated files

most people set status.showUntrackedFiles=no

So... it's done... can I just run software now or do I always need to use `onu` to install it?

according to onu, nope, I need to install in a BE. okay, when I guess I can mount the BE and... run it as a zone?

In terms of running the software, depends what you modified.

In general I don't recommend a zone based on a disjoint BE.

Most things need to be in sync (e.g. libc / kernel and other private interfaces), but depends a lot on what you modified.

If I just modified a single binary to fix something or library, I often cheat and lofs mount or similar.

But it all depends on the specifics.