Date: Wed, 20 Apr 2016 19:14:16 +0300 From: Roman Bogorodskiy <novel@FreeBSD.org> To: Neel Natu <neelnatu@gmail.com> Cc: Roman Bogorodskiy <bogorodskiy@gmail.com>, "freebsd-virtualization@freebsd.org" <freebsd-virtualization@freebsd.org> Subject: Re: Understanding Bhyve shutdown Message-ID: <20160420161414.GA87187@kloomba> In-Reply-To: <CAFgRE9FJrWnk5deije-tbqVj1SGH2QRvQsGGZJ2CZSY2_xQVSw@mail.gmail.com> References: <20160413105520.GB84953@dev.san.ru> <CAFgRE9FJrWnk5deije-tbqVj1SGH2QRvQsGGZJ2CZSY2_xQVSw@mail.gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
--uAKRQypu60I7Lcqm Content-Type: text/plain; charset=us-ascii Content-Disposition: inline Content-Transfer-Encoding: quoted-printable Neel Natu wrote: > Hi Roman, >=20 > On Wed, Apr 13, 2016 at 3:55 AM, Roman Bogorodskiy > <bogorodskiy@gmail.com> wrote: > > Hi, > > > > I was trying to get better understanding of how to properly shutdown VMs > > in bhyve, but unfortunately the documentation does not provide much > > details on that. > > > > Specifically, handbook [I] suggests to reboot a machine and then run > > bhyvectl --destroy on it. > > > > The bhyvectl(8) manpage mentions the '--force-reset' and > > '--force-poweroff' switches, but does not give details on those. > > > > I: https://www.freebsd.org/doc/handbook/virtualization-host-bhyve.html > > > > I tried all the options I know and wrote down the results. I also have > > some questions, hopefully you'll be able to answer some of them. > > > > 1. bhyvectl --vm=3D$name --destroy > > > > * looks like hard poweroff in the guest > > * the corresponding bhyve(8) process goes away > > * /dev/vmm/ entry goes away > > > > In my experience, it's a dangerous way to shutdown a VM because > > sometimes it appears it damages the image and VM fails to boot with > > something like this: > > > > --- > > Starting devd. > > mode =3D 0100600, inum =3D 170269, fs =3D / > > panic: ffs_valloc: dup alloc > > cpuid =3D 0 > > KDB: stack backtrace: > > #0 0xffffffff80984e30 at kdb_backtrace+0x60 > > #1 0xffffffff809489e6 at vpanic+0x126 > > #2 0xffffffff809488b3 at panic+0x43 > > #3 0xffffffff80b74a6e at ffs_valloc+0x84e > > #4 0xffffffff80bb60ad at ufs_makeinode+0x7d > > #5 0xffffffff80bb24fd at ufs_create+0x2d > > #6 0xffffffff80e71841 at VOP_CREATE_APV+0xa1 > > #7 0xffffffff809cd9e6 at uipc_bindat+0x346 > > #8 0xffffffff809c5488 at kern_bindat+0x108 > > #9 0xffffffff809c52a7 at sys_bind+0x77 > > #10 0xffffffff80d4b3f7 at amd64_syscall+0x357 > > #11 0xffffffff80d30adb at Xfast_syscall+0xfb > > Uptime: 3s > > > > Dump failed. Partition too small. > > --- > > >=20 > Yup, this is biggest hammer you could use to shutdown a virtual > machine. As you noticed, this is usually a bad thing because it does > not give the guest OS an opportunity to shutdown cleanly. >=20 > > 2. kill -SIGTERM $bhyve_pid > > > > If guest supports ACPI shutdown: > > > > * guest shuts down cleanly > > * the corresponding bhyve(8) process terminates > > * /dev/vmm entry is still here, need bhyvectl --destroy for complete > > cleanup > > > > If guest does not support ACPI shutdown (such as doing sysctl > > hw.acpi.power_button_state=3DNONE): > > > > * Nothing happens > > > > Q1: Is there a way to know if a guest reacted to power button but > > waiting for the bhyve process to terminate? >=20 > Not really, except to wait for some amount of time to give the guest a > chance to shutdown. >=20 > > Q2: Why it's not done via bhyvectl (it seems that it's easier for users > > + don't have to overload a useful SIGTERM signal) > > >=20 > It seems natural to overload SIGTERM to do this. For e.g. when the > host is shutting down it will send a SIGTERM to all running processes > and translating this into an ACPI poweroff event for the guest allows > it to shutdown cleanly. >=20 > > 3. bhyvectl --vm=3D$name --force-poweroff > > > > * looks like hard poweroff in the guest > > * the corresponding bhyve(8) process goes away > > * /dev/vmm entry is still here, need bhyvectl --destroy for complete > > cleanup > > > > Q: what's the practical difference with just doing --destroy right away? > > >=20 > 'force-poweroff' is equivalent to a hard power off on real hardware. >=20 > The only practical difference between '--force-poweroff' and > '--destroy' is that destroy will also remove the device node in > /dev/vmm and release any memory allocated for the guest back to the > host. >=20 > > 4. bhyvectl --vm=3D$name --force-reset > > > > Looks very similar to item #3 with just different exit code (reboot > > appears to be using 0, while shutdown and halt use 1 and 2). > > > > Q: what's the practical use of it? > > >=20 > The exit code can be used by scripts on top of 'bhyve' to decide > whether to restart execution of the guest (reset) or stop completely > (poweroff). >=20 > > Would greatly appreciate if somebody could provide more details on that. > > I guess we'll need to update Handbook with this information as well > > because it needs to mention SIGTERM for ACPI shutdown at least. > > >=20 > Hope that helps. >=20 > best > Neel Neel, thanks for the response, that's a kind of an answer I was looking for! One question regarding that '--force-poweroff' vs '--destroy' thing: what is the point of not releasing guest resources by doing '--force-powerof'? Initially I thought it could be useful to speed up the next boot by not requiring running 'bhyveload' (or other loader) for the guest again, but, apparently, it's still required to run '--destroy' and then bhyveload again, otherwise I get something like this: bhyve: could not activate CPU 0: Device busy I'm wondering if that's just for debugging purposes or there is something else? Sorry for being extra curious. :-) Roman Bogorodskiy --uAKRQypu60I7Lcqm Content-Type: application/pgp-signature; name="signature.asc" -----BEGIN PGP SIGNATURE----- Version: GnuPG v1 iQEcBAEBAgAGBQJXF6rWAAoJEMltX/4IwiJq1+IH/iAbACAob+uicIyhcKPfh6zv ahRnSoxYB3kNkqAxGqF05RWxJEqs8v/n/vjbbJzgIejo5zCoKm3Rs2s31xPbi1X7 /DCvQ91j2/QmqHZSLeNNvKUp4o8y0Azm5rgJ/uLZTVN2cg8GixVU+OshSmhad60T MqXzJJLUV26i9JMaa+5xnFxHgo+CorHSO0JQhvFn3DY4D4AZmxO6OfAGdcHuUGJ6 IlzuYhEG5aD4U5nigDtUoU784FJLLM9b+MLW4OPI0I2YVsKUG9jF80lsugWy8vdo j0mUfYiIB5lLA8PrLn/1lzto+f5zQBr7pajrRx6laFclcnw1hei5mCs70alzjTE= =NGYJ -----END PGP SIGNATURE----- --uAKRQypu60I7Lcqm--
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20160420161414.GA87187>