[illumos-gate] 16182 Convert exec(2) to mdoc -- Andy Fiddaman <illumos⊙fn>

[illumos-gate] 16194 s10 zones broken after #14128 -- Andy Fiddaman <illumos⊙fn>

I just changed a prop using svccfg `svccfg -s pkgsrc/bind setprop config-files/entities = astring: "file://localhost/opt/local/etc/named/named.conf"` . where are these changes written to?

They go into the SMF database. It's usually in `/etc/svc/` but that's an internal detail.

After setting a property, you should generally refresh the service to it to take effect in the current snapshot - `svcadm refresh pkgsrc/bind`

and you might need to restart it depending on how the service manifest is written

andyf_ thank you!

I love XML!

okay that sounded like sarcasm, but I really meant it :))

I feel like there should be a hier(7)

It is not a view I hear often!

filesystem(7)

andyf_ danke! <3

another silly question, say I want to log some data, say it's bind/named. should I put it in /var/log/named or is there an illumos-y/SMF-y way to do it?

I've seen different strategies, but I think /var/log is a reasonable option. I've seen /var/log/opt/local/bind before, and even /var/opt/llocal/log/bind

[illumos-gate] 14444 Convert dladm(8) to mdoc -- Andy Fiddaman <illumos⊙fn>

xml isn't that bad... and has a number of things that can in circumstances make it nicer than json

It's no YAML, that's for sure

yeah, no hanging a ruler off your screen :P

but having a schema specification that you can validate things against can be nice for some uses...

i don't think anything's really emerged like that for JSON as something common (though ISTR many attempts that never seemed to gain traction)

While reviewing Carlos's LX tcp_info mods for SmartOS, I realized that our TCP keep tcp_rto in MILLISECONDS which might fall-down-go-boom in a world where ping-times are measured in 10s-even-100s of MICROSECONDS.

Moving it out of clock_t's probably would be reasonable.

Especially given that we have other data that is used to calculate it in terms of hrtime_t's.

Exactly.

That may be a large knit item we unravel if we go down that path, however.

@sommerfeld *might* have more of that cached in his head...

I'm sure there's a lot. But if you're going to address it, I'd probably try to eliminate the reliance on ticks as much as plausible.

(Hence my comment about the large knit item being unravelled.)

[illumos-gate] 15989 Contention in kidmap_cache_lookup*() on global mutexes -- Gordon Ross <gwr⊙rc>

Is there a guide or man page on how to take an NVMe drive and configure it such that you're using, say, 33% of its capacity so the other 66% can be sacrificed to the drive's wear-balancers?

(ah shit, I hit return too soon... I'm AFK for the rest of today, and I wanted to paste that into another window for tomorrow)

danmcd: Not really with what we have at this moment.

But there are two different pieces here. If the device supports namespace management, you can carve that up to be different at some point.

On my SATA SSDs I just use format(1) and partition off chunks for not being used.

However, some vendors offer a resize bit.

I've not yet plugged it in to my HDC so can't tell what I have just yet. Too much else going on for HDC HW maintenance right now.

So absent namespace management you can still do that same thing.

You have a partition table on the device and can also just not partition the whole namespace (which is kind of not sort of like a LUN)

Thank you. I've a 250GB "data center" M.2 SSD. I'd like to use 80-90 of it for slog. Eventually i'll pair it with a same-size 250GB M.2 SSD and mirror them.

THis will allow me to stop overloading my boot SSDs (which are 500GB and only have 150GB in active partitions).

danmcd I just fixed the cstyle nits, thank you for looking into this