I'm using the following partman settings for automated installs of VMs for some years now:
d-i partman-auto/disk string /dev/sda
d-i partman-auto/method string regular
d-i partman-lvm/device_remove_lvm boolean true
d-i partman-md/device_remove_md boolean true
d-i partman-lvm/confirm boolean true
d-i partman/alignment string "optimal"
d-i partman-auto/expert_recipe string \
boot-root :: \
64 512 300% linux-swap \
$primary{ } \
method{ swap } format{ } \
. \
500 10000 1000000000 ext4 \
$primary{ } $bootable{ } \
method{ format } format{ } \
use_filesystem{ } filesystem{ ext4 } \
mountpoint{ / } \
.
d-i partman/confirm_write_new_label boolean true
d-i partman/choose_partition select finish
d-i partman/confirm boolean true
d-i partman/confirm_nooverwrite boolean true
This creates a small swap partition at the beginning and uses the rest of the disk for the root partition. This works well and allows easy partition extension when I need to increase the size of the virtual disk.
Now I'm trying to adapt this recipe to install a number of identical bare metal servers. To do that I just switched the partitions and set the sizes to more reasonable values for a machine with 256GB RAM and 460GB system disk (SSDs in a hardware RAID1, but that shouldn't matter):
boot-root :: \
32768 65536 1000000000 ext4 \
$primary{ } $bootable{ } \
method{ format } format{ } \
use_filesystem{ } filesystem{ ext4 } \
mountpoint{ / } \
. \
16384 16384 65536 linux-swap \
$primary{ } \
method{ swap } format{ } \
.
The rest of the partman* directives are identical.
As far as I understand the documentation (and additional posts like this) this should create a large root partition that spans nearly the entire disk, with a swap partition between 16 and 64 GB at the end.
Well, it doesn't. It creates a 450MB partition, followed by a 460GB swap partition.
When I install the same machine with the preseed for VMs, the partitions are created properly as defined in the preseed file:
So, what am I doing wrong with the recipe for the bare metal machine?
If it matters, the install iso is based on the Ubuntu 16.04.5 server iso.
Output of fdisk /dev/sda and parted /dev/sda print:
Some more variants I tried:
# this belongs to tha last block, as suggested by @Peter
#d-i partman-basicfilesystems/choose_label string gpt
#d-i partman-basicfilesystems/default_label string gpt
#d-i partman-partitioning/choose_label string gpt
#d-i partman-partitioning/default_label string gpt
#d-i partman/choose_label string gpt
#d-i partman/default_label string gpt
d-i partman-auto/expert_recipe string \
boot-root :: \
##########################
65536 1 -1 ext4 \
$primary{ } $bootable{ } \
method{ format } format{ } \
use_filesystem{ } filesystem{ ext4 } \
mountpoint{ / } \
. \
65536 65536 65536 linux-swap \
$primary{ } \
method{ swap } format{ } \
.
##########################
# 1 1 -1 ext4 \
# $primary{ } $bootable{ } \
# method{ format } format{ } \
# use_filesystem{ } filesystem{ ext4 } \
# mountpoint{ / } \
# . \
# 65536 65536 65536 linux-swap \
# $primary{ } \
# method{ swap } format{ } \
# .
##########################
# 32768 50 5242880 ext4 \
# $primary{ } $bootable{ } \
# method{ format } format{ } \
# use_filesystem{ } filesystem{ ext4 } \
# mountpoint{ / } \
# . \
# 16384 100 65536 linux-swap \
# $primary{ } \
# method{ swap } format{ } \
# .
##########################
# use along with the annoted partman-* directives above
# 538 538 1075 free \
# $iflabel{ gpt } \
# $reusemethod{ } \
# method{ efi } \
# format{ } \
# . \
# 1 1 -1 ext4 \
# $primary{ } $bootable{ } \
# method{ format } format{ } \
# use_filesystem{ } filesystem{ ext4 } \
# mountpoint{ / } \
# . \
# 65536 65536 65536 linux-swap \
# $primary{ } \
# method{ swap } format { } \ .
# .
It doesn't make a difference, the resulting root partition always has only 453MB.
Special notes
It wasn't mentioned whether there is bios_grub, ESP, GPT or MBR, or which method the firmware uses. When using GPT, you need one or the other (or both). So what will happen is the installer will choose for you, based on some unknown criteria, likely related to disk size. To force using GPT, you can set things like this (which I remember not necessarily always working, despite using all 6 methods I found here):
For example, for bios_grub, you would set:
(I didn't put an
$iflabel{ gpt }in there, and that can be useful because if anything goes wrong and you later convert it from MBR to GPT, you want the reserved space)And for EFI (shamelessly copying and pasting from Preseeding debian install - EFI and didn't remove the iflabel):
And to answer the question
To use the remainder, you use
-1, not an arbitrarily hugely big number, and then just set the other numbers to bogus values like 1. And I wouldn't trust the ranges either... I just use a different preseed file for different hardware, not debian's automagic stuff. So I set it as only 65536, not a range.And make sure it's really using your new preseed file. I like to put it in the netboot initrd image because then it has no dependencies on network, etc. and find that works great. If you change something like the above, and it doesn't work, check it again, and again. If it's http/tftp, check the http/tftp logs or tcpdump. Or make a change that you will surely notice, and see if it did it.
If it's initrd, look in the initrd and see if it's really there. For example:
TL;DR:
The wrong image was mounted. Thanks to @Peter for pointing me that way.
The long story:
After Peters comment I rechecked my whole iso build pipeline, the .seed files, the isolinux txt.cfg, the custom bootlogo, the bash script that builds the .iso, finding nothing wrong. I changed the .preseed file again, this time adding a third partition that should be mounted as
/var, rebuilt the image, clicked on "Unmount" and "Mount" in the BMC interface of the server, rebooted, ran the installation, and guess what, it was the same layout as before, no additional partition. Becoming suspicious I checked/target/etc/issue:That should have been 16.04.5. During my installation tests I recreated the Ubuntu .iso completely, basing it on 16.04.5 instead of the previous 16.04.1 (resulting in a different iso name, because the .iso file contains the version number).
I have two ways of mounting ISO images on those SuperMicro boards. I can connect to a single BMC and mount the ISO there, or I can use the SuperMicro Server Manager to mount an ISO on multiple machines at once. I had a number of machines to install, so naturally I used the SSM method and mounted the ISO everywhere.
After changing to 16.04.5 I only changed the filename directly in the BMC of the host I was working on and remounted it. I only get
There is an iso file mounted.as a confirmation, no more details.Clicking on the
Savebutton on the mount image page results in this error message:I did not enter the path this way, this was set by the Server Manager, so I assumed it was correct. Apparently it isn't. After this I remounted the .iso file with the Server Manager instead on the BMC directly, and voila ...
Exactly what I had configured to test this.
The moral of the story: I wasted about a week of my time (and some of the valuable time of you helpful people) just because of the stupidity of a cheap BMC interface. Keep that in mind and recheck EVERYTHING when you run into unexplainable problems.