3 more megs if we get rid of redundant copy of kernel?

classic Classic list List threaded Threaded
7 messages Options
Reply | Threaded
Open this post in threaded view
|

3 more megs if we get rid of redundant copy of kernel?

Douglas McClendon-2
It may well be that this has been discussed on this list before, and
worse yet, that I may have even participated, but if so I've long
forgotten, and it doesn't seem easy enough to search for.

In my livecd generation tools I go to a bit of extra hassle to play
symlink games with /boot/vmlinuz-$(uname -r), such that it doesn't get
put into the squashfs, since a copy of it is in the container iso/usb
filesystem.

Maybe we want to do that for f12?  It is a bit messy, but I think it's
worth it for 3/700M.

peace...

-dmc

--
Fedora-livecd-list mailing list
[hidden email]
https://www.redhat.com/mailman/listinfo/fedora-livecd-list
Reply | Threaded
Open this post in threaded view
|

Re: 3 more megs if we get rid of redundant copy of kernel?

Jeremy Katz
On Wednesday, July 15 2009, Douglas McClendon said:
> In my livecd generation tools I go to a bit of extra hassle to play  
> symlink games with /boot/vmlinuz-$(uname -r), such that it doesn't get  
> put into the squashfs, since a copy of it is in the container iso/usb  
> filesystem.
>
> Maybe we want to do that for f12?  It is a bit messy, but I think it's  
> worth it for 3/700M.

Nope, it breaks the ability to easily install by copying over the
filesystem.  And when you consider that someone may be booting via pxe
and then trying to install (see livecd-iso-to-pxeboot), there's no
guarantee that we can get that file back.

Jeremy

--
Fedora-livecd-list mailing list
[hidden email]
https://www.redhat.com/mailman/listinfo/fedora-livecd-list
Reply | Threaded
Open this post in threaded view
|

Re: 3 more megs if we get rid of redundant copy of kernel?

Douglas McClendon-2
Jeremy Katz wrote:

> On Wednesday, July 15 2009, Douglas McClendon said:
>> In my livecd generation tools I go to a bit of extra hassle to play  
>> symlink games with /boot/vmlinuz-$(uname -r), such that it doesn't get  
>> put into the squashfs, since a copy of it is in the container iso/usb  
>> filesystem.
>>
>> Maybe we want to do that for f12?  It is a bit messy, but I think it's  
>> worth it for 3/700M.
>
> Nope, it breaks the ability to easily install by copying over the
> filesystem.

I grant the 'easily' part.  But adding a bit of complexity to the
installer to dereference a symlink and get the file from /mnt/live isn't
that tough.

  And when you consider that someone may be booting via pxe
> and then trying to install (see livecd-iso-to-pxeboot), there's no
> guarantee that we can get that file back.

In this case, isn't the kernel in the massive initrd that gets
downloaded and booted?  Could you have some logic in the initrd that
detects PXE boot, and makes sure to save a copy of the kernel?

I guess I'm just of the mindset of looking at the 700M livecd and seeing
two copies of a 3M file, and thinking that somehow there is a way to
bypass that redundancy (even for all corner cases, but I admit, I'm
probably not familiar with all of them).

-dmc

--
Fedora-livecd-list mailing list
[hidden email]
https://www.redhat.com/mailman/listinfo/fedora-livecd-list
Reply | Threaded
Open this post in threaded view
|

Re: 3 more megs if we get rid of redundant copy of kernel?

Jeremy Katz
On Thursday, July 16 2009, Douglas McClendon said:

> Jeremy Katz wrote:
>> On Wednesday, July 15 2009, Douglas McClendon said:
>>> In my livecd generation tools I go to a bit of extra hassle to play  
>>> symlink games with /boot/vmlinuz-$(uname -r), such that it doesn't
>>> get  put into the squashfs, since a copy of it is in the container
>>> iso/usb  filesystem.
>>>
>>> Maybe we want to do that for f12?  It is a bit messy, but I think
>>> it's  worth it for 3/700M.
>>
>> Nope, it breaks the ability to easily install by copying over the
>> filesystem.
>
> I grant the 'easily' part.  But adding a bit of complexity to the  
> installer to dereference a symlink and get the file from /mnt/live isn't  
> that tough.

And yet, it's the sort of stuff that regularly leads to weird installer
bugs that are undiagnosable and turn people off of the distro.  Sorry,
the cost isn't worth the benefit here.

>  And when you consider that someone may be booting via pxe
>> and then trying to install (see livecd-iso-to-pxeboot), there's no
>> guarantee that we can get that file back.
>
> In this case, isn't the kernel in the massive initrd that gets  
> downloaded and booted?  Could you have some logic in the initrd that  
> detects PXE boot, and makes sure to save a copy of the kernel?

No.  The kernel is just loaded into memory.  You don't have any way to
just "save a copy of the kernel" aside from really disgusting things
involving parsing memory

Jeremy

--
Fedora-livecd-list mailing list
[hidden email]
https://www.redhat.com/mailman/listinfo/fedora-livecd-list
Reply | Threaded
Open this post in threaded view
|

Re: 3 more megs if we get rid of redundant copy of kernel?

Douglas McClendon-2
Jeremy Katz wrote:

> On Thursday, July 16 2009, Douglas McClendon said:
>> Jeremy Katz wrote:
>>> On Wednesday, July 15 2009, Douglas McClendon said:
>>>> In my livecd generation tools I go to a bit of extra hassle to play  
>>>> symlink games with /boot/vmlinuz-$(uname -r), such that it doesn't
>>>> get  put into the squashfs, since a copy of it is in the container
>>>> iso/usb  filesystem.
>>>>
>>>> Maybe we want to do that for f12?  It is a bit messy, but I think
>>>> it's  worth it for 3/700M.
>>> Nope, it breaks the ability to easily install by copying over the
>>> filesystem.
>> I grant the 'easily' part.  But adding a bit of complexity to the  
>> installer to dereference a symlink and get the file from /mnt/live isn't  
>> that tough.
>
> And yet, it's the sort of stuff that regularly leads to weird installer
> bugs that are undiagnosable and turn people off of the distro.  Sorry,
> the cost isn't worth the benefit here.

That's a valid concern.

>
>>  And when you consider that someone may be booting via pxe
>>> and then trying to install (see livecd-iso-to-pxeboot), there's no
>>> guarantee that we can get that file back.
>> In this case, isn't the kernel in the massive initrd that gets  
>> downloaded and booted?  Could you have some logic in the initrd that  
>> detects PXE boot, and makes sure to save a copy of the kernel?
>
> No.  The kernel is just loaded into memory.  You don't have any way to
> just "save a copy of the kernel" aside from really disgusting things
> involving parsing memory

Skimming the file, it looked to me like the iso was appended to the
initrd.  The iso having the kernel, suggests to me that the initrd
scripts would have access to the kernel as a regular file.

-dmc

--
Fedora-livecd-list mailing list
[hidden email]
https://www.redhat.com/mailman/listinfo/fedora-livecd-list
Reply | Threaded
Open this post in threaded view
|

Re: 3 more megs if we get rid of redundant copy of kernel?

Jeroen van Meeuwen-2
In reply to this post by Jeremy Katz
On 07/16/2009 04:07 PM, Jeremy Katz wrote:

> On Wednesday, July 15 2009, Douglas McClendon said:
>> In my livecd generation tools I go to a bit of extra hassle to play
>> symlink games with /boot/vmlinuz-$(uname -r), such that it doesn't get
>> put into the squashfs, since a copy of it is in the container iso/usb
>> filesystem.
>>
>> Maybe we want to do that for f12?  It is a bit messy, but I think it's
>> worth it for 3/700M.
>
> Nope, it breaks the ability to easily install by copying over the
> filesystem.  And when you consider that someone may be booting via pxe
> and then trying to install (see livecd-iso-to-pxeboot), there's no
> guarantee that we can get that file back.
>

OTOH, someone may opt-in to remove the duplicate files or foo or bar
basically breaking this feature which is not essential to the concept of
Live media (but is a Fedora feature and so must be on all Fedora
released Live media).

An 'rm' in %post would be too early, and so the "user" is out of
alternatives.

I'd suggest if someone really wants to remove the kernel from the
filesystem before it is squashed, but after it has been copied, they
write a plugin for Revisor and send me the patch ;-)

-- Jeroen

--
Fedora-livecd-list mailing list
[hidden email]
https://www.redhat.com/mailman/listinfo/fedora-livecd-list
Reply | Threaded
Open this post in threaded view
|

Re: 3 more megs if we get rid of redundant copy of kernel?

Douglas McClendon-2
Jeroen van Meeuwen wrote:

> On 07/16/2009 04:07 PM, Jeremy Katz wrote:
>> On Wednesday, July 15 2009, Douglas McClendon said:
>>> In my livecd generation tools I go to a bit of extra hassle to play
>>> symlink games with /boot/vmlinuz-$(uname -r), such that it doesn't get
>>> put into the squashfs, since a copy of it is in the container iso/usb
>>> filesystem.
>>>
>>> Maybe we want to do that for f12?  It is a bit messy, but I think it's
>>> worth it for 3/700M.
>>
>> Nope, it breaks the ability to easily install by copying over the
>> filesystem.  And when you consider that someone may be booting via pxe
>> and then trying to install (see livecd-iso-to-pxeboot), there's no
>> guarantee that we can get that file back.
>>
>
> OTOH, someone may opt-in to remove the duplicate files or foo or bar
> basically breaking this feature which is not essential to the concept of
> Live media (but is a Fedora feature and so must be on all Fedora
> released Live media).

Well, maybe it might break something else, but I corrected Jeremy on the
livecd-is-to-pxeboot case.

>
> An 'rm' in %post would be too early, and so the "user" is out of
> alternatives.
>
> I'd suggest if someone really wants to remove the kernel from the
> filesystem before it is squashed, but after it has been copied, they
> write a plugin for Revisor and send me the patch ;-)

If only there were a thousand hours in a day...

peace/out...

-dmc

--
Fedora-livecd-list mailing list
[hidden email]
https://www.redhat.com/mailman/listinfo/fedora-livecd-list