On Sun, Jul 31, 2016, 2:57 PM Felix Miata <mrmazda@earthlink.net> wrote:
Chris Murphy composed on 2016-07-31 12:43 (UTC-0400):

> Felix Miata wrote:

>> https://fedoraproject.org/wiki/DNF_system_upgrade quote: "This will
>> reboot your machine immediately. The system should boot again into
>> Fedora using the same kernel, but this time, the upgrade process appears
>> on the boot screen. "

>> NAICT, that last part did not take place, but the term "the boot screen"
>> is ambiguous. No bootloader is installed on the subject/target
>> installation, one of many on a multiboot machine on which everything
>> boots from a master bootloader under exclusive control of myself.

> dnf system-upgrade leverages systemd offline update mechanism. There's no
> bootloader changes. What should be try is 'dnf system-upgrade reboot' sets
> a symlink /system-update and as long as that exists, systemd picks it up
> and starts the offline update target.

> https://www.freedesktop.org/software/systemd/man/systemd.offline-updates.html

Quoting from it:
"Very early in the new boot systemd-update-generator(8) checks whether
/system-update exists. If so, it (temporarily and for this boot only)
redirects (i.e. symlinks) default.target to system-update.target, a special
target that is pulls in the base system (i.e. sysinit.target, so that all
file systems are mounted but little else) and the system update units."

NAICT, the 3 in the default boot stanza hijacked the redirect back to
multi-user.target. I restarted with the 3 removed after finding 700+ rpms in
the symlink /system-update/. Now vtty1 has stopped with these at screen bottom:

...
[OK] Started System Upgrade.
[OK] Reached target System Update.
[OK] Listening on D-Bus System Message Bus Socket.

The machine is very quiet, but I do see a flicker from the HD LED.

Later in that URL, again quoting:
"If applicable and possible, it should create a file system snapshot, then
install all packages."

What kind of snapshot does it want to create? / filesystem is EXT3, and
freespace prior to update rpms download was about 40%, 26% after. Only /home
and /usr/local are separate filesystems "required" by the F24 installation.


I'm not sure what kind of snapshot this refers to. I have done two dnf system upgrades from 23 to 24 on Btrfs, I don't recall seeing any Btrfs snapshots, but that's a weak statement.



>> Tail of .bash_history on freshly updated F24 installation cloned to a
>> new partition with boot stanza and fstab appropriately adjusted: dnf
>> remove kernel-PAE-core-4.1.0 kernel-PAE-core-4.2.0
>> kernel-PAE-core-4.3.0 df / dnf repolist all dnf install
>> dnf-plugin-system-upgrade df / dnf system-upgrade download --refresh
>> --releasever=25 dnf config-manager --set-disabled updates dnf
>> system-upgrade download --refresh --releasever=25 dnf system-upgrade
>> download --refresh --releasever=25 --allowerasing dnf system-upgrade
>> reboot

>> I had tried several times just to use DNF to upgrade, but always got
>> error messages about updates not refreshing, which is why I tried the
>> officially recommended upgrade route, and why in history you can see I
>> disabled updates. The system-upgrade reported 700+ packages successfully
>> downloaded. I saw nothing that looked like any error message.

>> All I saw after booting was a normal login screen on vtty1. On login, I
>> don't see anything happening in ps -A that looks like any kind of
>> updating, nor do I hear any HD activity. dnf repolist all shows only
>> Fedora 24 is enabled. What now?

> You'll need to figure out wher dnf system-upgrade download put all of the
> packages it downloaded. Make sure the /system-update symlink points to
> that location

It did/does.

>- since I think that gets set with the 'dnf system-upgrade
> reboot' command and then immediately reboots, you'd have to figure out a
> way to intercept the update process - e.g. boot another OS, mount the
> Fedora 24 root, and see if the symlink is there and points to the correct
> location.

I didn't shut it down, left it alone overnight.

> If it's not there or points to the wrong location, it's a bug.
> If it's there and points to the correct location, reboot and edit the
> Fedora 24 boot entry and remove 'rhgb quiet' so you can see the boot

I don't ever use any boot stanzas with quiet or rhgb in them.

> process. It's really obvious either way when offline updates are
> happening. With graphical boot, there's a status in percentage on the
> upper left of the display,

Plymouth is not installed. There is no graphical boot. GUI only appears after
boot and me doing something to start it, or booting with 5 on cmdline instead
of 3.

> and with text boot you'll see the
> systemd-offline-updates process installing packages, listing each package,
> along with a percent complete. It's not possible to miss it.

Not possible to _see_ any such thing here apparently.

> So if that's not happen, check again if the symlink is still there and
> pointed to the right location.

There are no vttys available to login on to check anything.

> Either way, you've got a bug it's just a
> matter of how it gets written up, but you'll probably need more verbose
> systemd logging to attach to the bug. So redo 'dnf system-upgrade reboot'
> and edit the Fedora 24 boot entry to add systemd.log_level=debug, boot
> that modified entry, and once you're in Fedora 24 do something like
> 'journalctl -b -o short-monotonic > journal_dnfupgradesystemdebug.log' and
> attach that log to the bug report. Hopefully it contains a hint. I'm
> pretty sure this is all done with just systemd, and doesn't involve any
> dracut stuff so you probably don't need to to also do rd.debug.

I left it alone a while. When I came back into the room, memtest was running,
indicated it rebooted itself. Logging into F25 reports "Fedora 25 (Rawhide)",
but on 4.6.4 fc24 kernel. 'rpm -qa | grep fc24 | wc -l' reports 297. Same
except s/fc24/fc25/g reports 776. Freespace is 35%. Oh, right, boot stanza
points specifically to the prior current kernel, as the upgrade process
presented me no opportunity to update kernel & initrd symlinks. 4.8.0...fc25
is installed. Reboot to 4.8 succeeds. :-)

'dnf repolist all' fails with "Error: failed to synchronize cache for repo
'updates-testing'".


Apparently normal at the current state of branch in guessing. I get the same thing on clean installs with an 20160730 compose, as well as a pre branch rawhide to F25 distro-sync.


Journals:
http://fm.no-ip.com/Tmp/Linux/F/
-b: http://fm.no-ip.com/Tmp/Linux/F/gx27c-jrnl-0.txt
-b -1: http://fm.no-ip.com/Tmp/Linux/F/gx27c-jrnl.txt
-b -2: http://fm.no-ip.com/Tmp/Linux/F/gx27c-jrnl-1.txt
-b -3: http://fm.no-ip.com/Tmp/Linux/F/gx27c-jrnl-2.txt
-b -4: http://fm.no-ip.com/Tmp/Linux/F/gx27c-jrnl-3.txt

Any suggestions where to go/what to do from here? Does 27% of total installed
packages =f24 sound about right for now?

Rawhide to F25 distrosync download was about 1/3 the size of a live ISO, which may be a completely different way of comparing this.


Chris Murphy