[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

help debian-installer release by processing installation reports

The d-i team is trying to release another beta release of d-i by the end
of this month, to comply with the Release Team's timeline for the sarge
release. We would like this release to be as good as it can be, but it's
hard to know if it is, since we have 480+ uncategorised installation
reports (filed against the installation-reports pseudo-package). A few
hundred of these are for beta3 and later, and since noone has gone over
them, we could have lots of serious problems we don't know about.

So I'm asking for some help from any interested DD's or anyone else who
knows their way around the BTS. No prior knowledge of d-i is necessary.
I've appended to this email a document that explains how to process an
installation report and figure out what parts of d-i are responsible for
each of the issues in it, so bugs can be cloned off to those parts. Even
processing a few installation reports will help us, we can't keep up
with the current volume on our own.

see shy jo

Dealing with d-i installation reports

Debian-Installer has a large number of installation reports in the BTS.
These are very valuable to us, since they're our only way of knowing how
well d-i is doing on widely varied hardware, operated by users who are not
intimatly familiar with d-i. But after each beta release of the installer,
we get more installation reports than our limited manpower can easily deal

This document is aimed at getting a Debian developer who is not
familiar with d-i up to the point where you can help us process and
categorise our install reports. Along the way, you should learn a lot more
about d-i.

It would be a good idea to go check out our web site
(http://www.debian.org/devel/debian-installer), read the
INSTALLATION-HOWTO, and do a test install to a spare swap partition or
machine, to get a feel for what d-i looks like, and what a user sees before
filing an installation report. You might want to file your own installation
report summarising your experiences, too.


All of our install reports should be under the "installation-reports"
pseudo-package in the BTS, although sometimes they are miscategorised in
other places (like under "installation"). As with any report, the users
often get the severity wrong; just because d-i breaks on their machine does
not really warrent a grave severity installation report.

The more current, interesting, and easier to deal with reports are at the
end of the list of normal severity reports. As you head back in time to the
beginning of the list, the versions of the installer become progressively
more broken, and our memories of the old bugs fainter.

The process of categorising an installation report is mainly one of reading
over the report, and identifying problems, and working out what part of the
installer is responsible for the problem, and cloning off a bug report to
be reassigned to that installer component. The goal is to make sure the
right people see the report, and make sure that no useful information is
disregarded or lost.

Processing a sample report

Let's look at a sample installation report, bug #230396. This walkthrough
is provided as an example of how someone knowledgeable about the parts of
d-i and how they interact would process this report. Later sections of this
document will try to fill in the gaps you'll need to be able to do the

The first thing to take note of is the version of the installer, and the
media used to install and basic description of the machine. Without this
info, many install reports will be useless, so if you find an install
report without that basic info, or that is too vague about it, you may need
to write the reporter to get more info, and tag it moreinfo in the

The summary of it is a little way down:

  Base System Installation Checklist:

  Initial boot worked:    [O]
  Configure network HW:   [E]
  Config network:         [O]
  Detect CD:              [ ]
  Load installer modules: [E]
  Detect hard drives:     [ ]
  Partition hard drives:  [ ]
  Create file systems:    [ ]
  Mount partitions:       [ ]
  Install base system:    [ ]
  Install boot loader:    [ ]
  Reboot:                 [ ]
  [O] = OK, [E] = Error (please elaborate below), [ ] = didn't try it

Well this install didn't go very well, they had problems and failed to
install. Looking in the "Comments/Problems" section, we see:

   I have a D-Link DE-220 ISA Card. The address/irq is 0x300, 11.
   The detection failed (its not a PNP card).
   Choosing the module 'ne' also failed, because it did not prompt me to
   enter the io port and irq.

   I was able to get the card to work by using insmod with the correct
   io/irq from the command line.

That explains the first "E" in the list. The part of the installer that is
responsible for configuring network hardware is the ethdetect package. The
problem is that apparently it did not make it easy enough for this user to
manually congfigure his ISA ethernet card (it autodetects only PCI cards).
So, look up its bug list and see if it has a bug for this issue. It does
not, so let's give it one:

	clone 230396 -1
	reassign -1 ethdetect
	retitle -1 failed to configure a D-Link DE-220 ISA Card
	tags -1 d-i

See the BTS documentation for help with the clone command if you're not
familiar with it. Notice that the new bug is retitled, to include as much
information about the hardware that caused the problem as possible. And
a d-i tag is added. We use these tags to be able to find all bugs in d-i,
across the set of packages that compose it.

Moving on the the next "E", we find this in the report:

  Load installer modules:

  Some of the mirrors listed (I tried 2 in canada) don't have the
  installer files.  (At least thats whats returned in the error message)

  While downloading files from the mirror, suddenly the installer quits to
  a console screen with the message "Terminated" repeating over and again
  once every few seconds.

The part of d-i that's responsible for picking a mirror to download debian
from is called "choose-mirror". It would be acceptible to reassign this
bug to it, as follows:

	clone 230396 -2
	reassign -2 choose-mirror
	retitle -2 failed to load all installer modules from Canadian mirrors
	tags -2 d-i

However, the second paragraph, about the installer crashing and repeating
an error message is really more interesting. This is a common symptom of
something going badly wrong, and if we look up at the top of the
installation report, where it describes the system, we find it has only 24
MB of memory. Note that beta 2 of d-i is documented to not work with less
than 32 MB. So the installer ran out of memory. Rather than discard the
report because of that, let's clone off a bug report, because it should
surely deal with low memory better than going into a crash loop:

	clone 230396 -3
	reassign -3 debian-installer
	retitle -3 goes into crash loop loading installer with 24 MB of ram
	tags -3 d-i

If it seems to be a general problem or it's not clear what part of the
installer is really at fault, it's acceptable to assign bugs to the
debian-installer pseudo package. The d-i team can always make better
reassignments later.

There is a bit more to this report that I left out. The user commented

    The root floppy has what I consider a vague name.
    Also, the rawrite2.exe tool wouldn't read the image files from the hard
    drive because the filename was too long.  I had to rename and shorten
    the image file names before I could create them under windows.  Maybe
    this is more of a problem with rawrite, but I digress.

This could also stand to be cloned off and reassigned to the
debian-installer pseudo package. It's a valuable observation.

	clone 230396 -4
	reassign -4 debian-installer
	retitle -4 floppy images have bad names that play badly with rawrite2
	tags -4 d-i

Finally, after sending off the four clone commands to
control@bugs.debian.org, you can close the installation report. Be sure to
thank the reporter for his report, suggest things he might try to get a
successfull install (upgrade his memory in this case..), and mention that
his issues have been brought to the attention of the debian-installer team.
You may also want to mail the maintainers of the packages that the report
was cloned to (if the packages are not part of d-i), to make sure they
notice the problem and understand it.

Note that some installation reports will only have one bug-worthy issue in
them. It's probably easier to not clone these, and just reassign the whole
installation report to the appropriatre package.

Background information

The example above showed that you need to know a certain amount of
information about the internals of d-i to properly categorise installation

A good place to start is by reading the d-i TODO list, in d-i's CVS.
This command will check out the whole d-i tree, which will be useful in
other ways too:

  cvs -d:pserver:anonymous@cvs.alioth.debian.org:/cvsroot/d-i co debian-installer

Then look in doc/TODO to see some of our most pressing and largest
problems. More known problems with the beta releases are documented on the
errata page, <http://www.debian.org/devel/debian-installer/errata>. If you
become familiar with these well-known problems, you can save a lot of time
dealing with the parts of install reports that repeat them, and focus
on the more interesting stuff.

You should also be aware that after a successful install, d-i writes all of
its logs to /var/log/debian-installer/ on the installed system. These logs
can be very useful.

Let's go through the stages of the install, and see what parts of the
installer are responsible for them.

  Initial boot worked:    [ ]

The parts of the installer responsible for the inital boot include the
linux kernel (if the boot error looks like a kernel bug). Debian-Installer
uses the stock Debian kernel image, At the time of this writing, it is
taken from the kernel-image-2.4.24-1-386 package for i386.

After the kernel, it's most likely a bug in the installer's initrd. If it
gets to init, and does not get to a prompt for a language, that's a good
bet. Such bugs should be assigned to the debian-installer pseudo-package.

If they're booting from a CDROM, the bug could be in the debian-cd package,
or in isolinux.

If they're netbooting, the problem could be anywhere..

If booting from floppies, a likely candidate is a bad floppy.

If booting from a USB memory stick, the most common cause of failure to
boot at all is an old BIOS.

If it booted up past init, but never got around to interacting with the
user, then other candidates include problems in main-menu and cdebconf.
main-menu is what drives the whole installation, and cdebconf is of course
what is used for interation. If these fail, the install won't get far.

Also, before the next item in the checklist, the installer will prompt
the user for their language (via the languagechooser package), and keyboard
(via kbd-chooser).

  Configure network HW:   [ ]

The frontend for network hardware configuration is the ethdetect package.
It in turn calls hw-detect to scan for PCI and PCMCIA hardware. Don't worry
which to assign bugs to, as they have the same source package.

If the user has PCI hardware that is not detected, then the bug should be
assigned to the discover-data package, since hw-detect uses discover to do
its job. When assigning a bug to discover-data, be sure that it includes
the module that should be loaded, and the PCI ID that discover should
associate with this module.

If the user has hardware that is detected, but the module fails to load,
it could be a bug in the kernel, and if so should be reassigned to the
appropriate kernel package, such as kernel-image-2.4.24-1-386.

Similarly, hangs during hardware detection are often kernel bugs.

  Config network:         [ ]

This is taken care of by the netcfg package. Users sometimes put an E
here when it belonged in "Configure network HW" instead.

If the problem relates to entering IP address, gateway, netmask, hostname,
etc, then netcfg is the place to assign it.

Netcfg runs a DHCP client, and currently dhcp-client is the one used.
Problems in configuring DHCP can be reassigned to that package.

  Detect CD:              [ ]

This is done by cdrom-detect. Of course, the kernel has to be able to see
the CD drive, and the CD has to be a valid CD. discover also takes care of
probing for SCSI and IDE disks, so if the installer cannot find their CD
drive at all, that's a place to look too.

  Load installer modules: [ ]

Depending on the type of install, the actual retrieval of the d-i udebs
will be done by one of cdrom-retriever, net-retriever, or floppy-retriever.
They are all controlled by anna (from the package by that name).

It's more likely that problems in this area have to do with bad media, or
networking issues, or bad mirrors.

  Detect hard drives:     [ ]

This is taken care of at the kernel level by discover again. If it fails to
find the drive, it's important to know what module should be loaded, and
of course the specifics of the hardware.

  Partition hard drives:  [ ]

d-i actually contains three different modules that can do this.

partman is our new system, which is now the default on most architectures.
It uses parted exclusively, and does file system creation too.

The old system is the "partitioner" module. This uses libparted to list the
existing partitions, and then cfdisk to do the actual partitioning.

If they say they used the automatic partitioner, that is "autopartkit".
It uses parted exclusively, and also takes care of file system creation in
the same step.

  Create file systems:    [ ]
  Mount partitions:       [ ]

Generally this is also partman's job. If the user used "partitioner" above,
then they will use partconf for these two steps. partconf uses libparted to
list the partitions. It uses standard mkfs.fstype programs to create the
various file systems.

  Install base system:    [ ]

This step is handled by base-installer, which uses debootstrap. If there is
a problem, it will 90% of the time be a problem with debootstrap, and the
debootstrap log file will be essential to working it out.

  Install boot loader:    [ ]

The default boot loader used to be lilo (for betas 1 and 2), and is now
grub (on i386 anyway). The various bootloader installation programs are
lilo-installer, grub-installer, yaboot-installer, and so on ad naseum.

If they had a boot loader install problem, it's important to know how their
partitions were set up.

  Reboot:                 [ ]

d-i does some things between boot loader install and reboot, that might
cause an error here. The prebaseconfig package is responsible for that.

Most often, a user will put an E here if their installed system failed to
boot. The boot loader is a good possibility, and if so see above. Other
possibilities include a kernel problem, and problems in the debian base

base-config also enters the picture here, as do tasksel, aptitude, all
the debian packages that could possibly be installed. If the user finds
problems in this part of the install, clone them off to the appropriate
debian package.

Various parts of d-i are responsible for setting up parts of the installed
system. Problems with /etc/network/interfaces, /etc/hosts, and similar are
in the purview of netcfg, while /etc/fstab is set up by partconf (or
autopartkit, or partman). hw-detect is responsible for ensuring that the
right modules are put in /etc/modules, and that packages like discover,
pcmcia-cs, and hotplug are installed onto the base system to deal with the

More information

The above is only an overview, and if you need more detail on a part of the
installer, you should post to debian-boot@lists.debian.org, or ask on the
#debian-boot irc channel on irc.debian.org. Or see the source.

Attachment: signature.asc
Description: Digital signature

Reply to: