richlowe/copec/jclulow: see zfs(8)'s description of volsize. By default a zvol is created with a refreservation but does not have blocks assigned to it. If swap did the DKIOCFREE thing, it would free the blocks to the pool but it would still have a commitment from the pool that it could have equivalent blocks back in the future if it needed them.

sommerfeld: Yeah, I get it

I am in favour of it haha

At the moment our strategy is to delete and recreate the zvol (with a new, random encryption key) at boot

on Oxide systems

We don't persist the key anywhere

But for classical UNIX systems I expect trimming would be valuable

(sorry, was away from IRC for a bit and saw some misconceptions floating in..). ephemerally encrypted swap is a good add-on.

sommerfeld, yeah, even though the blocks don't get de-allocated it hasn't been an issue because I've been running smartos for so long and the swap zvol is tiny compared to everything I have on "zones"

I did have a lot of OI boxes with 40GB Intel SSDs that it would have been really helpful with like 2010

rmustacc: thanks for your help. will try investigate vmware

hrm..

github.com/illumos/illumos-gate/blo…c/lib/libsysevent/libsysevent.h#L52 -- is there any reason to not include the function signature in the definition?

granted, i'm guessing syseventd plugins aren't at the top of people's lists for things to write...

but as i hack on zfs_mod I noticed that

jbk: yes, that definitely should get the full function prototype there.

are there any examples offhand of a service that runs on initial boot and disables itself?

(mostly wondering if there's any gotchas with just doing a `svcadm disable $SMF_FMRI` at the end of such a service

There is indeed a gotcha

I'll dig up a reference.

that's why I asked :)

and also why I want to get illumos 15320 (fenix) done some day.

FEATURE 15320: Add support for SMF_EXIT_TEMP_DISABLE (In Progress)

↳ illumos.org/issues/15320

it seemed like there could be, but I couldn't find anything that decribed it

I know some of this comment exists in gate, I just knew where to find this

yeah, the trick is always knowing where to look

Here's an example in gate, but without the comment.. github.com/illumos/illumos-gate/blo…sr/src/cmd/hotplugd/svc-hotplug#L40

ahh ok

5 seems to be favourite :)

tsoome: what is the eeprom syntax to set the default os console? I thought I had it, but can't find it.. ISTR that the key/type parameters you have to specify weren't immediately obvious

(at least to me)

[illumos-gate] 16348 e1000g I219 V17 does not attach -- Robert Mustacchi <rm⊙fo>

jbk with eeprom the simplest variant is eeprom console=value -- that one will be stored in /boot/solaris/bootenv.rc

jbk the -b bits really is an attempt to make it possible to create non-volatile store, saved in bootenv area of zfs label and making it possible to have "persistent" variables one can change from loader (bootenv.rc can only be changed from OS). Probably not the best UI;)

yeah.. this was to default the os_console value for a specific system

you can override bootenvv.rc with -B (see boot(8)), but that's even worse UI usually.

if you're stuck with the console in the OS image, and need to whack it

-B or set variable=value but thats not permanent:)

right, yeah, I just meant if jbk was needing to whack it temporarily, and the image on the OS was disagreeing with him :)

i mean, you set 'set os_console=foo' from the loader prompt

which works fine

but for an unattended boot

system X might need ttya, system Y might need ttyb, systemz might be text

and there's a way you can change the default, i didn't think it was just 'eeprom os_console=XXXX'

but something with the nvstore

but ISTR the the values you actually provide had some non-obvious bit about them that i can't recall anymore

and I can't find my notes on it now since this was like 2 years ago or so

it isn't ringing any bells here

it needs nvlist with name illumos:nvstore and pool name (if pool is not rpool)

probably should document it but perhaps need to improve it a bit first...

[illumos-gate] 16368 pkg:/driver/network/platform should just be pkg:/driver/network/dnet -- Richard Lowe <richlowe⊙rn>