Cheers, can the illumos "loader" (as seen in OI for example) be used to boot FreeBSD from ZFS? I suppose it should not have any trouble booting OmniOS, FWIW?

For context: migrating some NUT CI VMs from FossHost to DigitalOcean, and an OI cloud image is something I could use as a "custom template" there - so if I can just zfs send/recv the existing VMs' rootfs'es into the image that started out as an OI VM, that would simplify a lot :)

the loader is imported from freebsd, actually.

tsoome would answer your question with 100% certainity.

jimklimov chainload should work -- assuming the disk layout is done in supported way. direct load may, but it has never been the goal there, so it is not tested in any way.

jimklimov for chainload, you want to point to [p]mbr or loader.efi from freebsd.

yupee, OI from cloud image which was sucking in the zfs send/recv pool contents from fosshost actually booted, usable console, network and all!

DigitalOcean web-console does suck by disconnection during VM reboot, so early messages and sometimes loader menu are lost :\

hm, `init 6` seems to have got stuck after `syncing file systems... done` - I think it did not reboot actually

FWIW, also says "WARNING: Fast reboot is not supported on this platform due to presence of boot-time modules" - can this be related to not rebooting a line later?

Also, is github.com/illumos/metadata-agent packaged for OI? :)

Hm, after a few minutes, a "rebooting..." line appeared

And that dying uptime's console is still up...

So at least 15 min since `init 6`

You may be hitting illumos.org/issues/14526

→ BUG 14526: illumos guest hangs on reboot under QEMU 6.0.0 (New)

you're the quickest shot in the West! :)

at least by symptoms - looks like it

I've seen exactly that behaviour on Digital Ocean a few times :-(

did anyone come up with some solution or workaround? it would be cumbersome to have to go to the hosting UI to power-cycle the VM whenever we want a reboot (more so if it were unintentional - some kernel dump, zfs deadman, etc.)

for intentional cases, maybe might add some late-shutdown restapi call to "powercycle me!" in the same style NUT drivers tell the UPS to cut the juice

just need some way to enable running userland-provided logic after svc.startd has killed everyone

Are there watchdogs on DO for tighter integration (like that metadata-agent)? Can that take care of eventual unattended reboots?

Does current illumos (OI, OO) include some qemu guest software that could 1) probe if the QEMU host provides a virtual watchdog device, and 2) actually use one? I remember there was bmc-watchdog, but was it only for certain physical hardware?.. :)

at least there was nothing evident in `prtconf`...

eh, guess I'll leave it trying to reboot after an OS update now; waiting for DNS to refresh before productizing this VM anyway...

if "rebooting..." appeared after 10-15 mins, who knows what another hour can bring? :)

[illumos-gate] 5583 support QLogic 16Gb FC-HBA -- Hans Rosenfeld <hans.rosenfeld⊙nc>

The world is safe: I came back from a walk, and that VM booted after all, 3h:10m ago

So it was roughly hanging for 2 hours, but did reboot. Not AWOL indefinitely. Better than nothing =D