Bug#888183: libc6: Breaks openvz virtuals again

[James Cloos]

> "AJ" == Aurelien Jarno  writes:

AJ> No, that simply doesn't scale. Other packages than the one in src:glibc
AJ> depends on libc6 >= 2.26, and more and more are going to pick up this
AJ> dependency in the next months. They might also be unpacked before libc6
AJ> preinst is ran. We can't add this check in every package.

Then re-direct this to apt.  There needs then to be a way to mark the
deb that it *must* be installed first, before anything which depends
on it.  Not just "at the same time", but *before*.

And then add that notation to the glibc debs.

-JimC
-- 
James Cloos  OpenPGP: 0x997A9F17ED7DAEA6

Bug#888183: libc6: Breaks openvz virtuals again

[Aurelien Jarno]

On 2018-01-23 19:56, James Cloos wrote:
> AJ> Do you have a list of other packages from the glibc src that are
> AJ> upgraded before?
> 
> The following packages have unmet dependencies:
>  libc-dev-bin : Depends: libc6 (> 2.26) but 2.25-6 is installed
>  libc6-dev : Depends: libc6 (= 2.26-4) but 2.25-6 is installed
>  libc6-i386 : Depends: libc6 (= 2.26-4) but 2.25-6 is installed
>  libc6-pic : Depends: libc6 (= 2.26-4) but 2.25-6 is installed
>  libc6-x32 : Depends: libc6 (= 2.26-4) but 2.25-6 is installed
>  locales : Depends: libc-bin (> 2.26) but 2.25-6 is installed
>  nscd : Depends: libc6 (> 2.26) but 2.25-6 is installed
> 
> And on another box:
> 
> The following packages have unmet dependencies:
>  libc-dev-bin : Depends: libc6 (> 2.26) but 2.25-6 is installed
>  libc6-dev : Depends: libc6 (= 2.26-2) but 2.25-6 is installed
>  libnih1 : Depends: libc6 (> 2.26) but 2.25-6 is installed
>  locales : Depends: libc-bin (> 2.26)
> 
> And a third:
> 
> The following packages have unmet dependencies:
>  libc-dev-bin : Depends: libc6 (> 2.26) but 2.25-6 is installed
>  libc6-dev : Depends: libc6 (= 2.26-4) but 2.25-6 is installed
>  locales : Depends: libc-bin (> 2.26) but 2.25-6 is installed
> 
> Those are after one previous attempt at --fix-broken.

All the above packages have a proper dependency on libc6 >= 2.26. It's
apt which decide they can already be unpacked before libc6 preinst is
ran.

> AJ> In any case it's not something we can control, the abort is done
> AJ> inside the libc6 preinst, there is no way the libc6 package can
> AJ> do that earlier.
> 
> Of course it is.  Just make every deb from the glibc src do that abort.

No, that simply doesn't scale. Other packages than the one in src:glibc
depends on libc6 >= 2.26, and more and more are going to pick up this
dependency in the next months. They might also be unpacked before libc6
preinst is ran. We can't add this check in every package.

Aurelien

-- 
Aurelien Jarno  GPG: 4096R/1DDD8C9B
aurel...@aurel32.net http://www.aurel32.net

Bug#888183: libc6: Breaks openvz virtuals again

[James Cloos]

AJ> Do you have a list of other packages from the glibc src that are
AJ> upgraded before?

The following packages have unmet dependencies:
 libc-dev-bin : Depends: libc6 (> 2.26) but 2.25-6 is installed
 libc6-dev : Depends: libc6 (= 2.26-4) but 2.25-6 is installed
 libc6-i386 : Depends: libc6 (= 2.26-4) but 2.25-6 is installed
 libc6-pic : Depends: libc6 (= 2.26-4) but 2.25-6 is installed
 libc6-x32 : Depends: libc6 (= 2.26-4) but 2.25-6 is installed
 locales : Depends: libc-bin (> 2.26) but 2.25-6 is installed
 nscd : Depends: libc6 (> 2.26) but 2.25-6 is installed

And on another box:

The following packages have unmet dependencies:
 libc-dev-bin : Depends: libc6 (> 2.26) but 2.25-6 is installed
 libc6-dev : Depends: libc6 (= 2.26-2) but 2.25-6 is installed
 libnih1 : Depends: libc6 (> 2.26) but 2.25-6 is installed
 locales : Depends: libc-bin (> 2.26)

And a third:

The following packages have unmet dependencies:
 libc-dev-bin : Depends: libc6 (> 2.26) but 2.25-6 is installed
 libc6-dev : Depends: libc6 (= 2.26-4) but 2.25-6 is installed
 locales : Depends: libc-bin (> 2.26) but 2.25-6 is installed

Those are after one previous attempt at --fix-broken.

AJ> In any case it's not something we can control, the abort is done
AJ> inside the libc6 preinst, there is no way the libc6 package can
AJ> do that earlier.

Of course it is.  Just make every deb from the glibc src do that abort.

Idealy, w/ a warning on how to hold them so that the rest of the
upgradable packages can do so.

-JimC
-- 
James Cloos  OpenPGP: 0x997A9F17ED7DAEA6

Bug#888183: libc6: Breaks openvz virtuals again

[Aurelien Jarno]

On 2018-01-23 17:57, James Cloos wrote:
> > "AJ" == Aurelien Jarno  writes:
> 
> AJ> It's exactly what the message does. It stopped the upgrade and prevented
> AJ> libc6 to be unpacked. That way your system is still usable.
> 
> But that is NOT what the message does.  The other debs from the glibc
> source are already upgraded and now everything breaks.
> 
> The dialog has to occur before *any* of the debs from the glibc src are
> upgraded.

Do you have a list of other packages from the glibc src that are
upgraded before?

In any case it's not something we can control, the abort is done inside
the libc6 preinst, there is no way the libc6 package can do that
earlier.

Aurelien

-- 
Aurelien Jarno  GPG: 4096R/1DDD8C9B
aurel...@aurel32.net http://www.aurel32.net

Bug#888183: libc6: Breaks openvz virtuals again

[James Cloos]

> "AJ" == Aurelien Jarno  writes:

AJ> It's exactly what the message does. It stopped the upgrade and prevented
AJ> libc6 to be unpacked. That way your system is still usable.

But that is NOT what the message does.  The other debs from the glibc
source are already upgraded and now everything breaks.

The dialog has to occur before *any* of the debs from the glibc src are
upgraded.

-JimC
-- 
James Cloos  OpenPGP: 0x997A9F17ED7DAEA6

Bug#888183: libc6: Breaks openvz virtuals again

[Aurelien Jarno]

On 2018-01-23 17:17, James Cloos wrote:
> > "AJ" == Aurelien Jarno  writes:
> 
> AJ> This is perfectly normal that you do no get a dialog asking wether you
> AJ> want to upgrade libc6, as your kernel is too old for that.
> 
> Not it is not normal.  It still should have given that dialog so that I
> could skip libc and its fellows and upgrade what would have worked.

It's exactly what the message does. It stopped the upgrade and prevented
libc6 to be unpacked. That way your system is still usable.

> AJ> 2.6.32 kernels are not supported anymore, you need at least a 3.2 kernel
> AJ> to run buster.
> 
> Which is a bug.

No this is not a bug. At some point we can't support ancient kernels
anymore. The bug is that OpenVZ has stopped providing newer kernels,
please complain to them instead.

It's a decision of upstream, so it's nothing specific to Debian at all.
All distributions will be affected. Ubuntu, Fedora and OpenSUSE
Tumbletweed also do not support 2.6.32 kernels anymore.

> AJ> I'll add an entry to NEWS so that people get warned *before* trying to
> AJ> upgrade.
> 
> *How* will they see it first?  I only see package news after I upgrade,
> thanks to listchanges.

Well listchanges is supposed to display you the changes before the
upgrade, so that you can cancel it.

Aurelien

-- 
Aurelien Jarno  GPG: 4096R/1DDD8C9B
aurel...@aurel32.net http://www.aurel32.net

Bug#888183: libc6: Breaks openvz virtuals again

[James Cloos]

> "AJ" == Aurelien Jarno  writes:

AJ> This is perfectly normal that you do no get a dialog asking wether you
AJ> want to upgrade libc6, as your kernel is too old for that.

Not it is not normal.  It still should have given that dialog so that I
could skip libc and its fellows and upgrade what would have worked.

AJ> 2.6.32 kernels are not supported anymore, you need at least a 3.2 kernel
AJ> to run buster.

Which is a bug.

AJ> I'll add an entry to NEWS so that people get warned *before* trying to
AJ> upgrade.

*How* will they see it first?  I only see package news after I upgrade,
thanks to listchanges.

AJ> You can use http://snapshot.debian.org/ to properly downgrade to the
AJ> version prior to migration of glibc 2.26 to testing, which happened on
AJ> 2018-01-12. Taking a bit of margin, that gives you:

AJ>  deb http://snapshot.debian.org/archive/debian/20180110T00Z/ buster main

After running:

echo 'Acquire::Check-Valid-Until "false";' >/etc/apt/apt.conf.d/98expires

my update script almost worked with that.  I still needed --fix-broken install.
And a subsequent switch to stretch, even with dist-upgrade, didn't
switch anything over to the stretch packages.

Except of course on the 3 machines which report:

apt: /lib/x86_64-linux-gnu/libc.so.6: version `GLIBC_2.25' not found (required 
by /lib/x86_64-linux-gnu/libudev.so.1)

I had to grab this and install it with dpkg to get apt working again on
those (after a *lot* of earlier attempts):

  
http://snapshot.debian.org/archive/debian/20180109T213416Z/pool/main/s/systemd/libudev1_232-25%2Bdeb9u1_amd64.deb

-JimC
-- 
James Cloos  OpenPGP: 0x997A9F17ED7DAEA6

Bug#888183: libc6: Breaks openvz virtuals again

[Aurelien Jarno]

On 2018-01-23 20:04, cl...@jhcloos.com wrote:
> Package: libc6
> Version: 2.26-4
> Severity: important
> 
> [Reporting this from a different box.]
> 
> The latest libc upgrade again breaks openvz virtuals.
> 
> On kvm/xen/bare you get a dialog asking whether you want to upgrade libc.
> 
> On vz you get a notice about the (un-upgradable) kernel version but
> not the do you want dialog.

This is perfectly normal that you do no get a dialog asking wether you
want to upgrade libc6, as your kernel is too old for that. 2.6.32
kernels are not supported anymore, you need at least a 3.2 kernel to run
buster.

> Things which depend on libc6 get upgraded but the libc6 upgrade aborts.
> 
> This leaves blocks apt from installing or upgrading *any* packages.
> 
> apt --fix-broken install does not help.

I'll add an entry to NEWS so that people get warned *before* trying to
upgrade.

> Even trying to switch to stretch fails.
> 
> And the earlier 2.26 debs are gone from the mirrors, so downgrading to
> a working version is not possible either.
> 
> I tried using dpkg to downgrade to the current stretch libc version
> but that also fails because it is only 2.24 and
> /lib/x86_64-linux-gnu/libudev.so.1 requires GLIBC_2.25.  And that
> breaks all kinds of things, even some which have nothings to do with
> udev.

You can use http://snapshot.debian.org/ to properly downgrade to the
version prior to migration of glibc 2.26 to testing, which happened on
2018-01-12. Taking a bit of margin, that gives you:

 deb http://snapshot.debian.org/archive/debian/20180110T00Z/ buster main

Aurelien

-- 
Aurelien Jarno  GPG: 4096R/1DDD8C9B
aurel...@aurel32.net http://www.aurel32.net

Bug#888183: libc6: Breaks openvz virtuals again

[cloos]

Package: libc6
Version: 2.26-4
Severity: important

[Reporting this from a different box.]

The latest libc upgrade again breaks openvz virtuals.

On kvm/xen/bare you get a dialog asking whether you want to upgrade libc.

On vz you get a notice about the (un-upgradable) kernel version but
not the do you want dialog.

Things which depend on libc6 get upgraded but the libc6 upgrade aborts.

This leaves blocks apt from installing or upgrading *any* packages.

apt --fix-broken install does not help.

Even trying to switch to stretch fails.

And the earlier 2.26 debs are gone from the mirrors, so downgrading to
a working version is not possible either.

I tried using dpkg to downgrade to the current stretch libc version
but that also fails because it is only 2.24 and
/lib/x86_64-linux-gnu/libudev.so.1 requires GLIBC_2.25.  And that
breaks all kinds of things, even some which have nothings to do with
udev.

-- System Information:
Debian Release: buster/sid
  APT prefers unstable
  APT policy: (500, 'unstable')
Architecture: amd64 (x86_64)

Locale: LANG=en_US.UTF-8, LC_CTYPE=en_US.UTF-8 (charmap=UTF-8), 
LANGUAGE=en_US.UTF-8 (charmap=UTF-8)