Thanks for your insights about the "5.11" release number.

My OI illumos VM is now on a PCIe v3 M.2 card, the VM feels very snappy now.

Hi, is pkgsrc.smartos.org still alive? There are no commits since February on github

unixery: very much so, I don't use the default branch so it doesn't get updated very often, everything happens over on the main pkgsrc repository or my feature/release branches

what do you mean by "the main pkgsrc repository"? The NetBSD repo?

yes

but aren't the packages for illumos/macos build from the pkgsrc.smartos.org repo? Or how do package updates happen?

yes but not from the default branch, they're built from the 'release/*' branches (e.g. release/trunk for illumos, release/macos for macOS)

there's a doc I need to update for post-Joyent changes but it's roughly github.com/TritonDataCenter/pkgsrc/wiki/about:workflow

thanks! I've missed the release branches while scrolling. The branch list is long ;-)

we have a lot of history! ;)

BTW: could this be fixed TritonDataCenter/pkgsrc #384 ? Missing wg-quick is the reason why I still have MacPorts on my Mac

sure, I'll take a look now

done, but as per comment it'll probably be a few days until it shows up

great, thanks!

i don't suppose there's a way to force /bin/cat to not use mmap?

ksh cat looks like it doesn't, but uses a 64k buf

jbk: LD_PRELOAD that neuters mmap? Use dd instead of cat?

jbk: For a particular reason, or?

for reasons that i really don't have time to investigate, each 8mb mmap64 is taken 1-2s to do.. which means combining ~40 2gb files into one is taking hours

the writes are basically instantaneous

What is the source FS ?!

It's SMB isn't it haha

fastest workaround is probably to substitute dd with appropriate options instead of cat

I'm using a little ed script to edit a *.json file and to my surprise, ed is adding carriage returns to all the lines when I write out the result. Anyone run into that? Any idea why that happens? How to prevent it?

Huh. Seems to have to do with a missing newline at the end of the file. If I fix that first, then "ed" works as expected.

I have ancient memories that that's one of the unix demanding a newline things (ed appending a single one, at the end). Interspersing them would be new to me

matters of taste aside, does anyone known when anonymous unions became standard?

Anonymous unions were added in C11

Anyone have advice on tuning a (large) system to avoid seg_kp allocation failures? I've not run into this before.

thanks gwr

you're running out of pageable addresses!?

I don't have that experience, no

we probably need a bit more to really support them well.. i had done some stuff with mdb, but after a lot more investigation, i think it might make more sense to modify libctf when looking up struct/union members to deal with anonymous ones... i think mdb and dtrace would then DTRT

gwr: how much swap do you have?

v. how much physical memory

v. how much you're trying to allocate?

I don't immediately see ways this would fail unless we have actually run out of resources to hand out

in which case I think the only solution is more of those resources (probably VM)

It appears we've run out. physmem is 0xbfaca94

gwr: How many LWPs do you have?

I feel like that's usually what drove us to run out in the past

about 5100

That doesn't seem like that many!

I assume your segkpsize is at the default of 524288 pages (2GB)

segkpsize = 0x80000

so yes

A pthread_create failed in user-level....

right

well it would, as you're out of something an LWP needs to exist

gwr: What do you get from "::vmem segkp ! grep segkp"

For nthread = 2799 I only have ~82MB in use there

also 0 failed allocations etc

that comes out to about 29KB/thread, which I think makes sense, as lwp_default_stksize is 20480