Re: aufs based upgrade tests

2009-03-21 Thread John McCabe-Dansted
On Sat, Mar 21, 2009 at 2:28 AM, Alan Pope a...@popey.com wrote:
 That would fail for users who make one-time changes to their data. For
 example users who download their mail via pop. If they upgrade using
 aufs and then proceed to test the new features of their mail client,
 it might be configured to download new mail via pop and delete it from
 the server. In this case if they choose to not upgrade but throw the
 overlay away they lose their new mail and the ability to get it back.

This could also happen if we overlay /var/spool/mail. The type of user
is who stores mail in /var is likely to be more computer literate that
one who stores mail in ~. However I have already suggested not
overlaying /var/apt/cache, may we shouldn't overlay /var/spool/mail
either. This could get a bit piecemeal though, as we probably want to
overlay at least part of /var,  e.g. /var/lib/apt/lists/.



-- 
John C. McCabe-Dansted
PhD Student
University of Western Australia

-- 
Ubuntu-devel-discuss mailing list
Ubuntu-devel-discuss@lists.ubuntu.com
Modify settings or unsubscribe at: 
https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel-discuss


Re: aufs based upgrade tests

2009-03-20 Thread Lars Wirzenius
to, 2009-03-19 kello 16:20 +0100, Martin Soto kirjoitti:
 Seconded, you start your session once with the new version installed,
 and there's often no way back. Applications will surely update their
 configuration data as soon as they're started, and there's no guarantee
 that they'll work when the older system is reinstated.

In my humble opinion, such applications are broken. They will already
break when, for example, $HOME is shared over NFS, and different hosts
are not completely in sync, version-wise.

 As a possible solution, what about putting user accounts under aufs as
 well? This would have the advantage of allowing for testing how
 applications update their configuration data, which is also a good idea
 because this is a common source of bugs.

This seems like a good idea, so I second it, because applications often
are broken. It might be good to optionally allow /home to be excluded.


-- 
Ubuntu-devel-discuss mailing list
Ubuntu-devel-discuss@lists.ubuntu.com
Modify settings or unsubscribe at: 
https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel-discuss


Re: aufs based upgrade tests

2009-03-20 Thread Oliver Grawert
hi,
Am Freitag, den 20.03.2009, 09:49 +0200 schrieb Lars Wirzenius:
 to, 2009-03-19 kello 16:20 +0100, Martin Soto kirjoitti:
  Seconded, you start your session once with the new version installed,
  and there's often no way back. Applications will surely update their
  configuration data as soon as they're started, and there's no guarantee
  that they'll work when the older system is reinstated.
 
 In my humble opinion, such applications are broken. They will already
 break when, for example, $HOME is shared over NFS, and different hosts
 are not completely in sync, version-wise.
 
  As a possible solution, what about putting user accounts under aufs as
  well? This would have the advantage of allowing for testing how
  applications update their configuration data, which is also a good idea
  because this is a common source of bugs.
 
 This seems like a good idea, so I second it, because applications often
 are broken. It might be good to optionally allow /home to be excluded.
 
how about just adding a test user in the overlay that gets removed with the
aufs and have the option that application settings can be taken over from
an existing user for this account ?

ciao
oli




signature.asc
Description: This is a digitally signed message part
-- 
Ubuntu-devel-discuss mailing list
Ubuntu-devel-discuss@lists.ubuntu.com
Modify settings or unsubscribe at: 
https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel-discuss


Re: aufs based upgrade tests

2009-03-20 Thread Alexander Sack
On Wed, Mar 18, 2009 at 09:18:50AM +0100, Michael Vogt wrote:
 On Tue, Mar 17, 2009 at 03:53:23PM +0200, Marius Gedminas wrote:
  On Mon, Mar 16, 2009 at 07:00:12PM +0100, Michael Vogt wrote:
   Yes, after the upgrade the system will be jaunty until the next
   reboot, then the writable overlay is removed and the system is exactly
   in the same state as before the upgrade.
  
  Does this overlay cover the whole filesystem?  If people start creating
  files in /home that are going to be lost on reboot, it would be nice to
  warn them about that.
 [..]
 
 The /home dir is excluded from the overlay. There is still a certain
 risk with this however, assume hypothetical e.g. the new rhythmbox is
 run after the test upgrade and it converts its database files to a new
 format that the old (intrepid) version can no longer read. This is
 pretty rare, but it does sometimes happens.

Is there any reason why we don't overlay /home? Just so that you can
test upgrade for existing configs?

 - Alexander


-- 
Ubuntu-devel-discuss mailing list
Ubuntu-devel-discuss@lists.ubuntu.com
Modify settings or unsubscribe at: 
https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel-discuss


Re: aufs based upgrade tests

2009-03-20 Thread Alan Pope
2009/3/18 Alexander Sack a...@ubuntu.com:
 Is there any reason why we don't overlay /home? Just so that you can
 test upgrade for existing configs?


That would fail for users who make one-time changes to their data. For
example users who download their mail via pop. If they upgrade using
aufs and then proceed to test the new features of their mail client,
it might be configured to download new mail via pop and delete it from
the server. In this case if they choose to not upgrade but throw the
overlay away they lose their new mail and the ability to get it back.

Cheers,
Al.

-- 
Ubuntu-devel-discuss mailing list
Ubuntu-devel-discuss@lists.ubuntu.com
Modify settings or unsubscribe at: 
https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel-discuss


Re: aufs based upgrade tests

2009-03-19 Thread Vincenzo Ciancia
On 18/03/2009 Michael Vogt wrote:
 This is
 pretty rare, but it does sometimes happens.
 

How can this be rare if the user is supposed to test things after the 
upgrade? I didn't think about it but it's actually risky - every 
application can potentially be non-forward-compatible. Users should be 
adviced to use the guest user account to test new applications, or 
perhaps it can be done in a way that only a special account sees the new 
system?

Vincenzo


-- 
Ubuntu-devel-discuss mailing list
Ubuntu-devel-discuss@lists.ubuntu.com
Modify settings or unsubscribe at: 
https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel-discuss


Re: aufs based upgrade tests

2009-03-19 Thread Martin Soto
On Thu, 2009-03-19 at 15:56 +0100, Vincenzo Ciancia wrote:
...
 How can this be rare if the user is supposed to test things after the 
 upgrade? I didn't think about it but it's actually risky - every 
 application can potentially be non-forward-compatible. Users should be 
 adviced to use the guest user account to test new applications, or 
 perhaps it can be done in a way that only a special account sees the new 
 system?

Seconded, you start your session once with the new version installed,
and there's often no way back. Applications will surely update their
configuration data as soon as they're started, and there's no guarantee
that they'll work when the older system is reinstated.

As a possible solution, what about putting user accounts under aufs as
well? This would have the advantage of allowing for testing how
applications update their configuration data, which is also a good idea
because this is a common source of bugs.

Cheers,

M. S.


-- 
Ubuntu-devel-discuss mailing list
Ubuntu-devel-discuss@lists.ubuntu.com
Modify settings or unsubscribe at: 
https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel-discuss


Re: aufs based upgrade tests

2009-03-18 Thread Michael Vogt
On Tue, Mar 17, 2009 at 03:53:23PM +0200, Marius Gedminas wrote:
 On Mon, Mar 16, 2009 at 07:00:12PM +0100, Michael Vogt wrote:
  Yes, after the upgrade the system will be jaunty until the next
  reboot, then the writable overlay is removed and the system is exactly
  in the same state as before the upgrade.
 
 Does this overlay cover the whole filesystem?  If people start creating
 files in /home that are going to be lost on reboot, it would be nice to
 warn them about that.
[..]

The /home dir is excluded from the overlay. There is still a certain
risk with this however, assume hypothetical e.g. the new rhythmbox is
run after the test upgrade and it converts its database files to a new
format that the old (intrepid) version can no longer read. This is
pretty rare, but it does sometimes happens.

Cheers,
 Michael


-- 
Ubuntu-devel-discuss mailing list
Ubuntu-devel-discuss@lists.ubuntu.com
Modify settings or unsubscribe at: 
https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel-discuss


Re: aufs based upgrade tests

2009-03-17 Thread Marius Gedminas
On Mon, Mar 16, 2009 at 07:00:12PM +0100, Michael Vogt wrote:
 Yes, after the upgrade the system will be jaunty until the next
 reboot, then the writable overlay is removed and the system is exactly
 in the same state as before the upgrade.

Does this overlay cover the whole filesystem?  If people start creating
files in /home that are going to be lost on reboot, it would be nice to
warn them about that.

 I created as a stub wiki page:
 https://wiki.ubuntu.com/AufsBasedUpgrades

Marius Gedminas
-- 
Some of the more environmentally aware dinosaurs were worried about the
consequences of an accident with the new Iridium enriched fusion reactor.
If it goes off only the cockroaches and mammals will survive... they said.


signature.asc
Description: Digital signature
-- 
Ubuntu-devel-discuss mailing list
Ubuntu-devel-discuss@lists.ubuntu.com
Modify settings or unsubscribe at: 
https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel-discuss


Re: aufs based upgrade tests

2009-03-16 Thread Michael Vogt
On Sat, Mar 14, 2009 at 02:13:24PM +, (``-_-´´) -- BUGabundo wrote:
 Olá Michael e a todos.
 
 On Friday 13 March 2009 18:19:28 Michael Vogt wrote:
  during the last UDS we talked informally about using the aufs
  overlay filesystem layer for release upgrade testing. I build a
  prototype implementation of this now that should be ok for public
  testing. 
  
  The idea discussed with Evan Dandrea (and others) was to create a
  writable overlay into /tmp on top the systemdirs in / and then run
  the release upgrade. This way we can test easily if the system would
  upgrade cleanly (if no dpkg errors/maintainer script failures
  happen). All writes go into /tmp so after the upgrade and on the next
  reboot the system is back to its pre-upgraded state again (modulo
  /home, that is not overlayed). It also means the next boot takes a
  *long* time to clean /tmp - when I did test it on one of my production
  machines that wait made me *really* nervous :) But its ok, it just takes
  long (up to ~20 minutes or so).
  
  Feedback is welcome
 

 This idea seems like a really nice idea, and one that in some other
 form is requested by users/testers.  I would like to add to points:
 * if all tests go OK, and we end up with this on koala (to late for
 FFe on jaunty, right?), a checkbox when using update-manager -d /
 cli question on do-release-upgrade to use Sandbox would be much
 nicer then running all that code.  * to save the system state prior
 to upgrade, so that a user can restore the system if even after
 successful package upgrade, some application/kernel/driver upgrade
 doesnt go as good.
 
Thanks for this feedback. The longer term goal is provide the two
improvements you suggested :) The current version is a first step to
build up experience with the system. This is why its limited to
testing currently, but in the longer run we hope to make it more
capable. 

Cheers,
 Michael

-- 
Ubuntu-devel-discuss mailing list
Ubuntu-devel-discuss@lists.ubuntu.com
Modify settings or unsubscribe at: 
https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel-discuss


Re: aufs based upgrade tests

2009-03-16 Thread Michael Vogt
On Sun, Mar 15, 2009 at 02:27:16AM -0400, John Vivirito wrote:
 On 03/14/2009 10:13 AM, (``-_-´´) -- BUGabundo wrote:
  Olá Michael e a todos.
[..]
  This idea seems like a really nice idea, and one that in some other form is 
  requested by users/testers.
  I would like to add to points:
  * if all tests go OK, and we end up with this on koala (to late for FFe on 
  jaunty, right?), a checkbox when using update-manager -d / cli question on 
  do-release-upgrade to use Sandbox would be much nicer then running all that 
  code.
  * to save the system state prior to upgrade, so that a user can restore the 
  system if even after successful package upgrade, some 
  application/kernel/driver upgrade doesnt go as good.
  
 I am a bit on the short end of this topic due to trouble with having
 this set to digest mode. What exactly is this going to do. It sounds
 very interesting. is this similar to system restore in windows?
 The following quote makes it sound like after reboot it is going to
 restore itself to before the latest upgrade:
   All writes go into /tmp so after the upgrade and on the next
  reboot the system is back to its pre-upgraded state again 

Right now its a tool to help test if your version of ubuntu can
upgrade to the next version of ubuntu without errors. It does a
full regular upgrade from 8.10 to 9.04 but instead of writing it to
the system disk it writes all changes to a directory in /tmp
 
 Doe the above always write to /tmp? If so does it clear upon restart
 automatically?

Yes, after the upgrade the system will be jaunty until the next
reboot, then the writable overlay is removed and the system is exactly
in the same state as before the upgrade.

 Is there somewhere where i can get more information on it, a wiki or,
 blueprint or something?

Unfortunately not right now. I created as a stub wiki page:
https://wiki.ubuntu.com/AufsBasedUpgrades

and we will probably talk about it at the next UDS and create a more
formal plan. The currently version is build to get experience with the
system and fint bugs and limitations with the aufs based approach,
this is why its relatively complicated to enable it.

Cheers,
 Michael

-- 
Ubuntu-devel-discuss mailing list
Ubuntu-devel-discuss@lists.ubuntu.com
Modify settings or unsubscribe at: 
https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel-discuss


Re: aufs based upgrade tests

2009-03-16 Thread Michael Vogt
On Sun, Mar 15, 2009 at 11:51:52AM +0100, Vincenzo Ciancia wrote:
 When doing something like this one should be careful because here you 
 have a copy of all files that are modified during the upgrade. 
 Applications keeping these files open will write to the old copies, and 
 applications which reopen the file after the upgrade will not see this 
 data. This may be dangerous and lead to unexpected behaviour.

Thanks for the feedback. This is a problem we are aware of. The best
ways to fix it will probably we discussed at UDS. One approach (that
is available in the code as well) is to just create the overlay for
the dpkg child, this means that the regular desktop stuff (including
firefox) keeps working during the upgrade.

But in my tests it has been less of a practical problem than I
anticiapted, i.e. I have not had any issues because of that in my
tests (but of course this is all young and has not been tested that
much yet).
 
 Apart from that, as I ranted in the past, let me say that this is a very 
 important change and I am really happy that ubuntu developers are making 
 it happen.

Thanks, if it works out in the way we hope it will mean more
robustness to upgrades and that is certainly a good thing.

Cheers,
 Michael

-- 
Ubuntu-devel-discuss mailing list
Ubuntu-devel-discuss@lists.ubuntu.com
Modify settings or unsubscribe at: 
https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel-discuss


Re: aufs based upgrade tests

2009-03-15 Thread John Vivirito
On 03/14/2009 10:13 AM, (``-_-´´) -- BUGabundo wrote:
 Olá Michael e a todos.
 
 On Friday 13 March 2009 18:19:28 Michael Vogt wrote:
 during the last UDS we talked informally about using the aufs
 overlay filesystem layer for release upgrade testing. I build a
 prototype implementation of this now that should be ok for public
 testing. 

 The idea discussed with Evan Dandrea (and others) was to create a
 writable overlay into /tmp on top the systemdirs in / and then run
 the release upgrade. This way we can test easily if the system would
 upgrade cleanly (if no dpkg errors/maintainer script failures
 happen). All writes go into /tmp so after the upgrade and on the next
 reboot the system is back to its pre-upgraded state again (modulo
 /home, that is not overlayed). It also means the next boot takes a
 *long* time to clean /tmp - when I did test it on one of my production
 machines that wait made me *really* nervous :) But its ok, it just takes
 long (up to ~20 minutes or so).

 Feedback is welcome
 
 This idea seems like a really nice idea, and one that in some other form is 
 requested by users/testers.
 I would like to add to points:
 * if all tests go OK, and we end up with this on koala (to late for FFe on 
 jaunty, right?), a checkbox when using update-manager -d / cli question on 
 do-release-upgrade to use Sandbox would be much nicer then running all that 
 code.
 * to save the system state prior to upgrade, so that a user can restore the 
 system if even after successful package upgrade, some 
 application/kernel/driver upgrade doesnt go as good.
 
 
 
I am a bit on the short end of this topic due to trouble with having
this set to digest mode. What exactly is this going to do. It sounds
very interesting. is this similar to system restore in windows?
The following quote makes it sound like after reboot it is going to
restore itself to before the latest upgrade:
  All writes go into /tmp so after the upgrade and on the next
 reboot the system is back to its pre-upgraded state again 

Doe the above always write to /tmp? If so does it clear upon restart
automatically?
Is there somewhere where i can get more information on it, a wiki or,
blueprint or something?

-- 
Sincerely Yours,
John Vivirito

https://launchpad.net/~gnomefreak
https://wiki.ubuntu.com/JohnVivirito
Linux User# 414246

How can i get lost, if i have no where to go
-- Metallica from Unforgiven III



signature.asc
Description: OpenPGP digital signature
-- 
Ubuntu-devel-discuss mailing list
Ubuntu-devel-discuss@lists.ubuntu.com
Modify settings or unsubscribe at: 
https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel-discuss


Re: aufs based upgrade tests

2009-03-15 Thread Vincenzo Ciancia
When doing something like this one should be careful because here you 
have a copy of all files that are modified during the upgrade. 
Applications keeping these files open will write to the old copies, and 
applications which reopen the file after the upgrade will not see this 
data. This may be dangerous and lead to unexpected behaviour.

Apart from that, as I ranted in the past, let me say that this is a very 
important change and I am really happy that ubuntu developers are making 
it happen.

Vincenzo


-- 
Ubuntu-devel-discuss mailing list
Ubuntu-devel-discuss@lists.ubuntu.com
Modify settings or unsubscribe at: 
https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel-discuss


Re: aufs based upgrade tests

2009-03-14 Thread (``-_-´´) -- BUGabundo
Olá Michael e a todos.

On Friday 13 March 2009 18:19:28 Michael Vogt wrote:
 during the last UDS we talked informally about using the aufs
 overlay filesystem layer for release upgrade testing. I build a
 prototype implementation of this now that should be ok for public
 testing. 
 
 The idea discussed with Evan Dandrea (and others) was to create a
 writable overlay into /tmp on top the systemdirs in / and then run
 the release upgrade. This way we can test easily if the system would
 upgrade cleanly (if no dpkg errors/maintainer script failures
 happen). All writes go into /tmp so after the upgrade and on the next
 reboot the system is back to its pre-upgraded state again (modulo
 /home, that is not overlayed). It also means the next boot takes a
 *long* time to clean /tmp - when I did test it on one of my production
 machines that wait made me *really* nervous :) But its ok, it just takes
 long (up to ~20 minutes or so).
 
 Feedback is welcome

This idea seems like a really nice idea, and one that in some other form is 
requested by users/testers.
I would like to add to points:
* if all tests go OK, and we end up with this on koala (to late for FFe on 
jaunty, right?), a checkbox when using update-manager -d / cli question on 
do-release-upgrade to use Sandbox would be much nicer then running all that 
code.
* to save the system state prior to upgrade, so that a user can restore the 
system if even after successful package upgrade, some application/kernel/driver 
upgrade doesnt go as good.


-- 
Hi, I'm BUGabundo, and I am Ubuntu (whyubuntu.com)
(``-_-´´)   http://LinuxNoDEI.BUGabundo.net
Linux user #443786GPG key 1024D/A1784EBB
http://BUGabundo.net
ps. My emails tend to sound authority and aggressive. I'm sorry in advance. 
I'll try to be more assertive as time goes by...


signature.asc
Description: This is a digitally signed message part.
-- 
Ubuntu-devel-discuss mailing list
Ubuntu-devel-discuss@lists.ubuntu.com
Modify settings or unsubscribe at: 
https://lists.ubuntu.com/mailman/listinfo/ubuntu-devel-discuss