On Fri, Oct 24, 2014 at 12:58 AM, Roman Bogorodskiy <bogorodskiy@gmail.com> wrote:
Conrad Meyer wrote: Hi Conrad.
Thanks for the patch. As for the /tmp, probably storing it somewhere in BHYVE_STATE_DIR would be more clean.
Sure.
However, I'm concerned about libvirt dealing with the grub-bhyve specifics as it involves some assumptions about its behaviour ahd has limitations, like the one you pointed with installing from CD.
Sure. However, the user can always override the assumptions. And I think the CD case is actually quite easy to fix, if that's a blocking issue.
My idea was to add the following behaviour:
* bootloader specified and bootloader_args specified -- use that as is * bootloader not specified, bootloader_args specified -- call bhyveload $bootloader_args $default_bhyveloader_args * bootloader and bootloader_args not set -- use defaults (current behaviour) * bootloader specified, bootloader_args missing -- fail here
Ok. I think the grub-bhyve stuff is just a special case of the 4th bullet here; but otherwise my patch matches up with this behavior.
Actually, I've implemented that already, but didn't send because it needs some more testing.
Ok. I am happy to guinea-pig bhyve improvements for libvirt, if you need it.
The reason of my concern about adding a special handling of grub-bhyve case that, as I've mentioned, we have to make assumptions and there are also caveats you mentioned. Probably it'd worth to wait for bhyve to provide single-step VM boot, i.e. UEFI support, that should appear soon anyway.
Any idea how soon? Months? A year? 5 years? I'm not comfortable postponing improvements indefinitely for vaporware. In the wonderful bhyve-UEFI future, we can ignore/warn about <bootloader>.
There are some comments inline.
Thanks!
- /* Image path */ - virCommandAddArg(cmd, "-d"); - virCommandAddArg(cmd, virDomainDiskGetSource(disk)); + /* XXX cleanup this file. */ + fd = mkstemp(tmpmapfile);
This yields an error in 'syntax-check'. Running 'make syntax-check' allows to avoid some of the general code and formatting issues.
I did run syntax-check, but did not see any error around there. syntax-check has lots of console output; does it log somewhere else too? Thanks, Conrad