hibernation support - lack of distro-wide coordination between systemd, dracut, anaconda, pm-utils and maybe more?

[Chris Murphy]

On Wed, Apr 15, 2015 at 9:02 AM, Bastien Nocera <bnocera at redhat.com> wrote:
>
>
> ----- Original Message -----
>> On Wed, Apr 15, 2015 at 10:26:14AM -0400, Bastien Nocera wrote:
>> >
>> >
>> > ----- Original Message -----
>> > > On Tue, Apr 14, 2015 at 09:30:31AM -0600, Chris Murphy wrote:
>> > > > On Tue, Apr 14, 2015 at 3:07 AM, Bastien Nocera <bnocera at redhat.com>
>> > > > wrote:
>> > > > >
>> > > > >
>> > > > > ----- Original Message -----
>> > > > >> OK not everyone is on the same page, apparently. This bug was just
>> > > > >> closed by Anaconda as WONTFIX.
>> > > > >>
>> > > > >> suggested swap for laptop seems low
>> > > > >> https://bugzilla.redhat.com/show_bug.cgi?id=1037472
>> > > > >>
>> > > > >> I don't see how hibernation works reliably with such a low default
>> > > > >> swap
>> > > > >> size.
>> > > > >
>> > > > > This isn't the way to fix it. The hibernation file/partition should
>> > > > > really be independent
>> > > > > of swap, because 1) you can't be sure how much swap will actually be
>> > > > > used
>> > > > > by the applications
>> > > > > so you can't be sure you'll ever have enough swap to save the RAM 2)
>> > > > > Too
>> > > > > much swap and the
>> > > > > (lack of) interactivity will make you want to advocate physical
>> > > > > violence
>> > > > > when your machine
>> > > > > is unusable for an hour because of a hungry Javascript in your 50th
>> > > > > Firefox tab.
>> > > >
>> > > > Windows and OS X both use swapfiles rather than swap partition, and a
>> > > > sleep image file rather than a partition. OS X's swapfiles are
>> > > > dynamically created on demand in variable size increments.
>> > > I think the problem is in the ways filesystems are implemented.  The
>> > > fs has to be mounted to access the swap file, and this can change the
>> > > fs, even with a read-only mount. Because we don't have
>> > > really-read-only fs mounting, we need to support swap-as-partition, so
>> > > we might just as well use it by default.
>> > >
>> > > > Both OS's have a feature that I find invaluable on a laptop which is
>> > > > the automatic switch from suspend-to-RAM to suspend-to-disk.
>> > > Yes, integrating with firmware would be great. So far this hasn't been
>> > > hapenning...
>> > > What we can do instead is use hybrid sleep. It's not smart at all,
>> > > and doesn't prevent your battery from draining completely, but it does
>> > > protect
>> > > your data.
>> > >
>> > > Systemd supports hybrid-sleep as another option analogous to suspend
>> > > and hibernation, so for anything using systemd to suspend swithing to
>> > > hybrid should be trivial. Maybe we should make this an F23 goal:
>> > > - use hybrid-sleep from Gnome and other DE by default
>> >
>> > Hybrid sleep as offered in systemd still is just suspend + hibernation, and
>> > the way we do hibernation is broken.
>> Can you be more specific? Do you consider hibernate-to-swap-partition
>> unacceptable?
>
> I think that conflating "memory-to-disk swap space" with "I can hibernate my machine"
> is unacceptable. We need a new partition type that Anaconda would setup, or
> a whitelist of laptops with firmwares that support rapid start (and again, Anaconda
> to set it up), or use a temporary file of any sort to store the hibernation data.

I'm willing to bet there's an incongruence between IRST and multiboot.

>
> If my machine has 8 gigs of memory, I don't want to need 8 gigs plus of swap to be
> able to hibernate it, when run away processes can make my machine unusable for hours
> if they start hitting that swap.

Googling brings up the swapspace project for dynamically creating
swapfiles. However, for Btrfs and LVM thinp installs, I think the base
swap code needs revisiting to defer rw through the fs rather than
assuming direct rw is workable.

-- 
Chris Murphy