jbk: I only use SNMP for things without a functional alternative haha

i.e., in the current lab, it's just the APC PDUs and the Cisco switches

I feel like the prometheus metrics format is looking to achieve a new SNMP status at this point haha

viz., not especially good, but well-defined and widely used

yeah, i think it made more sense at the time (2007/2008 i think) why i was surprised about the desire for it now since it seems like everyone uses some other format

[illumos-gate] 16339 sockfs: variable may be used uninitialized -- Toomas Soome <tsoome⊙mc>

Hye, I see very slow perfromance for ramdisk (ramdisk -a test 2g) -> read 422 MB/s.

Is it buggy or It could be improved somehow by tuning something ?

command: dd if=/dev/ramdisk/test of=/dev/null bs=1M

vetal how much throughput do you get reading from cached file?

tsoome_: "dd if=/dev/zvol/rdsk/rpool/swap bs=1M count=2048 of=/dev/null" gets 655 MB/s

tsoome_: But it is ZFS with its locks, etc.

tsoome_: ramdisk should be fast like in Linux 5-10 GB/s

tsoome_: even on non-debug kernel (OI hipster) it produces reading from ramdisk (874 MiB/sec)

I meant something simple like reading iso or usb image few times to see the memory throughput: dd if=OpenIndiana_Text_X86.usb of=/dev/null bs=1M

886046720 bytes (845 MiB) transferred in 0,161758 secs (5,1 GiB/sec)

regarding to ramdisk, I have no comments about its implementation, probably worth grabbing flamegraph or like and see what is going on.

we do not use ramdisk too much (I think oxide is probably using it;)

tmpfs (used by /tmp, /var/run and such, is not ramdisk)

hm, it may be the installer (in some distribution) may use ramdisk with miniroot from install media...

[illumos-gate] 16340 sockfs: label defined but not used -- Toomas Soome <tsoome⊙mc>

tsoome_: Good, Thanks, I will look into it and will try to improve.

[illumos-gate] 16192 Retire libc's thr_debug.h in favor of <sys/debug.h> -- Bill Sommerfeld <sommerfeld⊙ho>

[illumos-gate] 16480 krtld doesn't properly handle tail called sdt probes on x86 -- Luqman Aden <luqman⊙oc>

tsoome_: unless you're going to get 16498 really soon, I probably can.

but I have literal bytes/second of bandwidth, so it's... exciting to use ssh over a vpn right now :)

alanc's not theh only one typing blind and hoping the right window is underneath it once the characters catch up :)

good evening, does anyone here has a recommendation for a S3 bucket server on illumos? MinIO doesn't build because of some file method abstractions missing, garage fails because one dependency complains it has no idea what "illumos" is...

[illumos-gate] 16446 dtrace consumers should not be isaexec'd -- Richard Lowe <richlowe⊙rn>

richlowe :) I still have small queue of my active bits (and long queue of inactive ones:D)

Hello (echo)

[illumos-gate] 16455 want TCP_MD5SIG socket option -- Andy Fiddaman <illumos⊙fn>

github.com/illumos/illumos-gate/blo…common/sys/sata/impl/sata.h#L63-L66 whoops :)

jbk: ha, I wonder how hard it is to train smatch to catch that

someone wanted to typedef the struct definition, saw that it was done differently, but didn't completely convert it...

rzezeski: you need something like lint's global crosschecks to notice that nobody references it as a global..

ouch...

they did typedef it, just with more vowels :)

unless you wanted to add some sort of "identifiers must be unique even if vowels are missing" rule.

which will really mess up everyone's favorite for loop variable.

or maybe an opinionated "things ending in _t must be types" rule.

vowel-insensitive names, one step beyond case-insensitive names.

sommerfeld: I don't know exactly how smatch works, but I thought it had some notion of the AST. So I was thinking more along the lines of "we never define a struct with a declarator". But probably not worth the effort given how unlikely this is to happen.

its always about cost versus profit:) but then again, it can also be interesting challenge:)

rzezeski: I've tried a few times to add things to smatch, and not been happy.

if one of you do figure out how to, I'd love a check that getopt's return is not a character, and rmustacc would like a check that id_t is not printed with a literal %d or %ld

'cos they're non-portable but impossible to teach gcc about

I did contribute to smatch to fix a bug (that helps with perf), but I didn't really look at how you write rules for it.

but if I ever do look, I'll keep those in mind

if this is _after_ perf fixes, I'd hate to see before

have you ever watched it think over nvme.c?

I thought I'd broke something and went running to rm :)

haha, pretty sure we haven't updated to the version that includes my fix, it was pretty recent