17.10 Install Fails to Start

Bug #1694531 reported by Joshua Powers
10
This bug affects 1 person
Affects Status Importance Assigned to Milestone
linux (Ubuntu)
Invalid
Critical
Unassigned
Artful
Invalid
Critical
Unassigned
main-menu (Ubuntu)
Fix Released
Critical
Unassigned
Artful
Fix Released
Critical
Unassigned
ubiquity (Ubuntu)
Artful
Invalid
Undecided
Unassigned

Bug Description

As of May 25, 2017 it appears the Ubuntu 17.10 server ISOs have failed to start an install. Trying other options like 'rescue a broken system' or 'Install MAAS [Region|Rack] Controller' have the same issue.

Expected results:
The Ubuntu Server installer menu system should have started asking for a language or keyboard mapping.

Steps to reproduce:
1. Download an ISO dated after May 24 (server or desktop) [1]
2. Boot system to ISO
3. Choose language
4. Select 'Install Ubuntu Server'
5. Black screen will flicker with blinking cursor in bottom left

Investigation:
Running 'Check the disk for defects' finds nothing wrong.

The last few syslog entries do not give any glaring issues:
acpiphp: Slot [30] registered
acpiphp: Slot [31] registered
PCI host brdige to bus 0000:00
random: crng init done: 0xffffffffffffffff max_cycles: 0x39c5

[1] http://cdimage.ubuntu.com/ubuntu-server/daily/

Revision history for this message
Joshua Powers (powersj) wrote :

Here is a list of packages that were updated or changed between the May 24 and May 30 ISO:
https://paste.ubuntu.com/24718548/

Revision history for this message
Joshua Powers (powersj) wrote :

Here is a list of udeb differences between the versions:
https://paste.ubuntu.com/24718580/

Because this is occurring before the install, the udeb and kernel on the system are what I suspect to be the root cause. I'll add the kernel to this defect.

Revision history for this message
Dean Henrichsmeyer (dean) wrote :

In my case, the desktop ISO doesn't even boot. It gets to the ISOLINUX splash and just hangs indefinitely.

Revision history for this message
Brad Figg (brad-figg) wrote : Missing required logs.

This bug is missing log files that will aid in diagnosing the problem. From a terminal window please run:

apport-collect 1694531

and then change the status of the bug to 'Confirmed'.

If, due to the nature of the issue you have encountered, you are unable to run this command, please add a comment stating that fact and change the bug status to 'Confirmed'.

This change has been made by an automated script, maintained by the Ubuntu Kernel Team.

Changed in linux (Ubuntu):
status: New → Incomplete
Revision history for this message
Joshua Powers (powersj) wrote :

This is during an install and unable to run apport-collect.

description: updated
Changed in linux (Ubuntu):
status: Incomplete → Confirmed
tags: added: artful
Revision history for this message
Dean Henrichsmeyer (dean) wrote :

My case is different than what Josh reports. Mine turned out to be a failing USB key. Using a different one with the current daily booted and installed fine.

Revision history for this message
Brad Figg (brad-figg) wrote :

@josh,

i have just successfully booted today's daily live image on an older macbook air. i have also installed an artful MAAS image on a server in the DC

Revision history for this message
Joshua Powers (powersj) wrote :

@brad-figg,

Can you give the md5sum of the file you used sucessfully? This is what I am using:
46eb85873d18b45badf86c3e8618c2ff dated May 30, 2017 from
http://cdimage.ubuntu.com/ubuntu-server/daily/pending/

Note I originally thought this was desktop and server due to Dean's comment's, but like him I verified Desktop is working as expected and I am only seeing this on Server.

Revision history for this message
Brad Figg (brad-figg) wrote :

@josh,

It's a MAAS daily image, installed via MAAS.

Revision history for this message
Brad Figg (brad-figg) wrote :

@powersj

I downloaded the ISO and tried booting it on an Intel NUC. No luck. The monitor I have attached just flashes after the language screen. Something bad has happened here.

Changed in linux (Ubuntu):
importance: Undecided → Critical
tags: added: kernel-key
Changed in linux (Ubuntu):
status: Confirmed → Triaged
Revision history for this message
Joseph Salisbury (jsalisbury) wrote :

I also downloaded the ISO and tried installing into a VM. I can select a language, but as soon as I select "Install Ubuntu Sever" the screen starts flashing.

I'm trying all the other options to see which work and don't work. I'll also try on Hyper-V just as another data point.

Revision history for this message
Joseph Salisbury (jsalisbury) wrote :

Options I tried from the initial installation menu:

Install Ubuntu Server: FAIL
Install MAAS Region Controller: FAIL
Install MAAS Rack Controller: FAIL
Check disk for defects: PASS
Test memory: PASS
Boot from first hard disk: N/A
Rescue a broken system: FAIL

Revision history for this message
Joseph Salisbury (jsalisbury) wrote :

A VM on Hyper-V fails in exact same manner.

Revision history for this message
Joseph Salisbury (jsalisbury) wrote :

Same issue using the 32bit and 64bit ISOs.

Revision history for this message
Joseph Salisbury (jsalisbury) wrote :

It seems if you press CTRL ALT F2 or another tty during the flashing, you can get to a busy box prompt.

Revision history for this message
Joseph Salisbury (jsalisbury) wrote :

Pressing CTRL ALT F4 will show the following scrolling:

init: starting pid NNN, tty '': '/sbin/reopen-console /sbin/debian-installer'
debconf: Setting debconf/language to en
init: process '/sbin/reopen-console /sbin/debina-installer' (pid NNN) exited. Scheduling for restart.

Revision history for this message
Joseph Salisbury (jsalisbury) wrote :

I was able to boot the 4.10.0-21.23 kernel on another VM successfully, which is the kernel included in this ISO.

Revision history for this message
Joseph Salisbury (jsalisbury) wrote :

There is a trace in syslog,but it may or may not be related. See screen shot: Screenshot_ArtfulServer_2017-05-30_20:45:16TRACE.png

Revision history for this message
Joseph Salisbury (jsalisbury) wrote :

There is one change to mm in 4.10.0-21 and it is:
ddfc93a33110 x86/mm/KASLR: Exclude EFI region from KASLR VA space randomization

Again, it may not be related, just investigating the trace posted in comment #18

Revision history for this message
Joseph Salisbury (jsalisbury) wrote :

And I don't see that call trace with Hyper-V, so it is probably something different.

Revision history for this message
Joseph Salisbury (jsalisbury) wrote :

Here's a screenshot of the scrolling messages after "Install Ubuntu Server" selection:

Revision history for this message
Joshua Powers (powersj) wrote :

Confirmed that with the amd64 ISO dated June 2nd this is no longer an issue.

The i386 and ppc64el ISOs passed testing this morning, amd64 did not, but it appears to have ran with the June 1 ISO still. I have kicked off a run to download the ISO again (June 2) and try again.

Revision history for this message
Mathieu Trudel-Lapierre (cyphermox) wrote :
Download full text (3.6 KiB)

This had already been identified and "fixed" on Wednesday, probably around the time the bug was filed (I uploaded the main-menu update around 11h30 EST). For posterity, I've captured relevant information at http://people.canonical.com/~mtrudel/lp1694531/)

Unfortunately, it's just the kind of bug that is genuinely hard to recognize and debug unless you have a good understanding of how d-i works; so I'll explain for the future:

We uploaded a libdebian-installer change on May 24th, which did some changes to the symbols provided by that library. It looked tame to me, so I sponsored the changes and did not insist on rebuilding the reverse-dependencies immediately as non of them relied on the exact symbols changed (SHA256 support). Namely, main-menu and netcfg were identified as needing a rebuild eventually, and needing a merge. Someone else had started doing d-i merges.

In preparing the netcfg merge and other changes, while testing I noticed that d-i would not start correctly. You'd reach the "d-i screen", but you would see the screen flicker, restarting repeatedly, forever. A little bit of debugging (starting the installer in expert mode, command-line, using the mini.iso image) and with the knowledge that the screen flickers after loading what looks like the d-i background, it's likely the main menu or some part of the early steps (language/keyboard selection, etc) that fail, as well as the messages already shown here in screenshots, I identified main-menu as the culprit. /sbin/debian-installer is a short script that just runs other pieces, some of which can be quickly dismissed as unlikely to apply to a default install, which helps narrowing things down.

Running main-menu from the console (ctrl-alt-f2) yields an error (see screenshot included in my capture of relevant files). This makes it apparent that the issue is related to libdebian-installer, which was changed with a date greater than main-menu's last upload. Rebuilding main-menu would be a reasonable attempt to fix the bug, but given that we were lagging a bit behind Debian, it was merged. main-menu was directly uploaded to the archive since the merge changes were straightforward and correct (and the merge obviously needed to be done anyway).

The important part when changing d-i components, however, is that some of them are so important that they are also used during d-i's build process, so as to be included in main.iso. They are listed under build/pkg-lists in the d-i source tree. main-menu is listed in build/pkg-lists/base. When listed, the packages are included in mini.iso, and thus require a debian-installer rebuild when changed. debian-installer was also uploaded in a PPA for testing with main-menu.

After verifying that the mini.iso image booted fine, I contacted Adam Conrad to synchronize on the necessary debian-installer upload, since he routinely does d-i uploads for new kernels. d-i was uploaded to artful and published in -proposed on 2017-05-29, and finally migrated to -release at 2017-06-01 16:54:47 EDT; the following ubuntu-server (20170602) includes all fixes.

Once thing to notice is that 20170602 is still listed under ubuntu-server/daily/pending rather than ubuntu-serv...

Read more...

Revision history for this message
Mathieu Trudel-Lapierre (cyphermox) wrote :

Moving to main-menu and closing the ubiquity task: ubiquity was unaffected (it was an issue with a bad USB key, and ubiquity does not use main-menu in any way). It's obviously also not a kernel issue. Thanks for Joseph for the great debugging work!

no longer affects: ubiquity (Ubuntu)
Changed in linux (Ubuntu Artful):
status: Triaged → Invalid
Changed in main-menu (Ubuntu Artful):
status: New → Fix Released
importance: Undecided → Critical
Changed in ubiquity (Ubuntu Artful):
status: New → Invalid
To post a comment you must log in.
This report contains Public information  
Everyone can see this information.

Other bug subscribers

Remote bug watches

Bug watches keep track of this bug in other bug trackers.