For BZ1206936 [1] we started adding resume= kernel command line by default [2] in fedora installs. This is causing issues for ostree based systems, which I haven't fully investigated yet, but figured I would ask the question:
Is resume=/path/to/swap something we really want on server installs like Fedora Server and Fedora Atomic Host?
Dusty
[1] https://bugzilla.redhat.com/show_bug.cgi?id=1206936 [2] https://github.com/poncovka/anaconda/commit/07c8591b107c34f3825dfa2d2b20c5c9...
On Wed, Oct 17, 2018 at 11:27 PM Dusty Mabe dusty@dustymabe.com wrote:
For BZ1206936 [1] we started adding resume= kernel command line by default [2] in fedora installs. This is causing issues for ostree based systems, which I haven't fully investigated yet, but figured I would ask the question:
Is resume=/path/to/swap something we really want on server installs like Fedora Server and Fedora Atomic Host?
It seems *unnecessary*, but I'm not sure I understand what problems it causes. Could you got into more detail there?
I can't see anyone putting a Server or Atomic Host into hibernation, though. Does this just direct the kernel where that should go if hibernation is requested, or does it reserve space or something for it?
On 18/10/2018 12:58, Stephen Gallagher wrote:
On Wed, Oct 17, 2018 at 11:27 PM Dusty Mabe dusty@dustymabe.com wrote:
For BZ1206936 [1] we started adding resume= kernel command line by default [2] in fedora installs. This is causing issues for ostree based systems, which I haven't fully investigated yet, but figured I would ask the question:
Is resume=/path/to/swap something we really want on server installs like Fedora Server and Fedora Atomic Host?
It seems *unnecessary*, but I'm not sure I understand what problems it causes. Could you got into more detail there?
I can't see anyone putting a Server or Atomic Host into hibernation, though. Does this just direct the kernel where that should go if hibernation is requested, or does it reserve space or something for it?
AFAIK it tells the kernel where to look for a hibernation image to restore from - so it normally points at one of the swap partitions.
If the kernel finds a hibernation image on the specified device then it will restore that and switch to it otherwise it will just continue booting as normal.
Tom
On Thu, Oct 18, 2018 at 07:58:33AM -0400, Stephen Gallagher wrote:
I can't see anyone putting a Server or Atomic Host into hibernation, though. Does this just direct the kernel where that should go if hibernation is requested, or does it reserve space or something for it?
There might be some use for it for VMs....
On Thu, Oct 18, 2018 at 8:29 AM Matthew Miller mattdm@fedoraproject.org wrote:
On Thu, Oct 18, 2018 at 07:58:33AM -0400, Stephen Gallagher wrote:
I can't see anyone putting a Server or Atomic Host into hibernation, though. Does this just direct the kernel where that should go if hibernation is requested, or does it reserve space or something for it?
There might be some use for it for VMs....
Doesn't the hypervisor do that itself for VMs?
On Thu, Oct 18, 2018 at 08:32:56AM -0400, Stephen Gallagher wrote:
On Thu, Oct 18, 2018 at 8:29 AM Matthew Miller mattdm@fedoraproject.org wrote:
On Thu, Oct 18, 2018 at 07:58:33AM -0400, Stephen Gallagher wrote:
I can't see anyone putting a Server or Atomic Host into hibernation, though. Does this just direct the kernel where that should go if hibernation is requested, or does it reserve space or something for it?
There might be some use for it for VMs....
Doesn't the hypervisor do that itself for VMs?
You have a choice, but most people do use hypervisor save/restore. It is rare to see guest S3/S5 save/restore used in VMs.
The hypervisor initiated save/restore VM state to a file will work with any OS, as it is transparent to the guest. The saved state can also be explicitly managed by the host mgmt tools, letting you do snapshotting and rollback of state, if you combine memory state with file snapshots.
The key downside is that it is transparent to the guest, so when you restore the guest, its clock will be wrong and needs fixing / catching up.
With guest initiated save/restore you're not using any additional host resources and its all confined in the guest execution environment. The guest knows to update its clock upon restore since it is aware of what is happening. The downside is that not all guest OS are reliable at doing S3/S5, and for that matter not all hypervisors will actually advertize S3/S5 support in their BIOS. IIRC, RHEL KVM explicitly disables this.
Regards, Daniel
On Thu, Oct 18, 2018 at 01:41:21PM +0100, Daniel P. Berrangé wrote:
There might be some use for it for VMs....
Doesn't the hypervisor do that itself for VMs?
You have a choice, but most people do use hypervisor save/restore. It is rare to see guest S3/S5 save/restore used in VMs.
Okay then. I was remembering this from https://fedoraproject.org/wiki/Features/Virt_Guest_Suspend_Hibernate
On 10/18/2018 07:58 AM, Stephen Gallagher wrote:
On Wed, Oct 17, 2018 at 11:27 PM Dusty Mabe dusty@dustymabe.com wrote:
For BZ1206936 [1] we started adding resume= kernel command line by default [2] in fedora installs. This is causing issues for ostree based systems, which I haven't fully investigated yet, but figured I would ask the question:
Is resume=/path/to/swap something we really want on server installs like Fedora Server and Fedora Atomic Host?
It seems *unnecessary*, but I'm not sure I understand what problems it causes. Could you got into more detail there?
Yeah. https://pagure.io/atomic-wg/issue/513#comment-536736
Basically for some reason on ostree based systems the LV isn't available before the systemd unit for hibernation runs and boot has to wait for that unit to timeout before continuing. I'm still investigating why that is.
I can't see anyone putting a Server or Atomic Host into hibernation, though. Does this just direct the kernel where that should go if hibernation is requested, or does it reserve space or something for it?
On 10/18/18 6:02 AM, Dusty Mabe wrote:
On 10/18/2018 07:58 AM, Stephen Gallagher wrote:
On Wed, Oct 17, 2018 at 11:27 PM Dusty Mabe dusty@dustymabe.com wrote:
For BZ1206936 [1] we started adding resume= kernel command line by default [2] in fedora installs. This is causing issues for ostree based systems, which I haven't fully investigated yet, but figured I would ask the question:
Is resume=/path/to/swap something we really want on server installs like Fedora Server and Fedora Atomic Host?
It seems *unnecessary*, but I'm not sure I understand what problems it causes. Could you got into more detail there?
Yeah. https://pagure.io/atomic-wg/issue/513#comment-536736
Basically for some reason on ostree based systems the LV isn't available before the systemd unit for hibernation runs and boot has to wait for that unit to timeout before continuing. I'm still investigating why that is.
It does sound like anaconda could be more picky here for when it adds this. Additional to the atomic/ostree case, if you have secure boot enabled it's useless to add, if your swap is smaller than your memory it's useless to add, etc.
kevin
-
Daniel P. Berrangé -
Dusty Mabe -
Kevin Fenzi -
Matthew Miller -
Stephen Gallagher -
Tom Hughes