This repository contains various configuration files and scripts.

See e.g. Ubuntu Wiki.

By default Ubuntu with LVM and encryption creates following partitions:

LVM VG vgubuntu, LV root as ext 4
LVM VG vgubuntu, LV swap_1 as swap
partition #1 of /dev/nvme0n1 as ESP

The swap partition is by default only ~1G. In order to resize the root logical volume and increase swap do the following:

1. Install Ubuntu with LVM and encryption
2. Restart to 'Try Ubuntu', i.e. 'Live CD' so that root is not actively used
3. Run the increase_swap.sh script that moves 40G space from root to swap
4. Reboot to installed Ubuntu and continue the setup

Ubuntu release upgrades can be done using the do-release-upgrade utility from the update-manager-core package.

One can verify that it's installed from the output of

dpkg -s update-manager-core

As a prerequisite make sure the system is up to date:

sudo apt update
sudo apt upgrade
sudo apt dist-upgrade
sudo apt autoremove

And finally, run the release upgrade with

sudo do-release-upgrade

or to upgrade from the latest supported release to the development release, run:

sudo do-release-upgrade -d

Note that new LTS versions become available with release xx.04.1, as explained here.

The default make target will call the $XDG_BIN_HOME/upgrade script:

make

Note that if the upgrade script has not yet been linked, then the recipe will call it from this repository.

There is a config installation make target that links config files to dedicated locations.

WARN: Note that original files will be overwritten!

make links

First check whether the system runs Wayland - e.g., by looking at XDG_SESSION_TYPE.

There is a Gnome extension alacritty-toggle, which enables toggling the visibility of the terminal window.

This extension can be installed with make alacritty-toggle which clones the repository and configures the toggle button to F1.

Unfortunately, the extension is picked up only after restarting current Gnome session so one must logout, log back in and re-run the make target.

1. Make sure GNOME Tweaks is installed
2. Follow these instructions

This configuration is actually stored as .config/zsh/p10k.zsh and automatically symlinked to $POWERLEVEL9K_CONFIG_FILE.

It is important to note that $POWERLEVEL9K_CONFIG_FILE is set in .zshrc and uses $XDG_CONFIG_HOME. Additionally, .zshrc disables the p10k configure configuration wizard.

Therefore, if one deliberately wants to run it, the wizard must be re-enabled with

POWERLEVEL9K_DISABLE_CONFIGURATION_WIZARD=false p10k configure

Also the XDG specification, respectively $POWERLEVEL9K_CONFIG_FILE, should be respected if p10k.zsh ought to be rewritten by the wizard.

Generally, there are two setups which, according to the p10k documentation, respect terminal's color palette (and hence Base16):

• Lean prompt style with 8 colors
• Rainbow prompt style which currently does not offer an option to select between 8 and 256 colors (empirically 256)

Some settings used in p10k configure when p10k.zsh was originally created include:

1. Prompt Style: Rainbow or Lean
2. Character Set: Unicode
3. Prompt Separators: Slanted (only relevant to Rainbow)
4. Prompt Heads: Slanted (only relevant to Rainbow)
5. Prompt Tails: Flat (only relevant to Rainbow)
6. Prompt Colors: 256 colors (only relevant to Lean)
7. Show current time?: 24-hour format
8. Prompt Height: Two lines
9. Prompt Connection: Disconnected
10. Prompt Frame: No frame
11. Prompt Spacing: Sparse
12. Icons: Many icons
13. Prompt Flow: Concise
14. Enable Transient Prompt?: same-dir
15. Instant Prompt Mode: quiet

Additionally one can download Meslo Nerd Fonts and configure terminals as instructed. The font download is part of make fonts.

Tinted Shell can be installed with make tinted-shell. One can then manually switch themes with base16_<theme> which creates and/or updates a link $HOME/.config/tinted-theming/base16_shell_theme.

Note that the shell hook is already added to and liked the oh-my-zsh plugin tinted-shell, so after installation with make one just needs to reload zsh and pick a theme:

omz reload
base16_gruvbox-dark-hard

The default theme is set in .zshenv to BASE16_THEME_DEFAULT=gruvbox-dark-hard.

One can also test the themes with included colortest tool.

Note that alacritty.yml automatically imports set color schemes:

import = ["~/.config/alacritty/colors.toml"]

The colors.toml is a symlink created by make alacritty which points to a tinted-alacritty theme. Note that currently the theme is selected just once before the installation and further changes won't be automatically picked up.

Bat supports base16 color theme by default via --theme=base16 option or BAT_THEME environment variable.

Current setting is BAT_THEME=base16-256 which, according to the documentation:

is designed for base16-shell

Custom themes are supported if appropriate file exists in BAT_CONFIG_DIR/themes When new themes are downloaded and linked, one must update the cache by running:

bat cache --build

and then check the theme is available via:

bat --list-themes

Unless the Base16 extension is properly synced, install it and follow the instructions.

Finish Poetry setup by manually configuring auth tokens.

Nvim uses its built-in LSP support and leverages nvim-lspconfig to configure individual LSPs. In order to install LSP binaries for various languages, refer to corresponding sub-section below.

Rust Analyzer is installed together with other rustup components as part of

make rust

The Python LSP Server together with all its dependencies can be installed via:

make python-language-server

The LuaLS can be installed via:

make lua-language-server

The installation of the Haskell Language Server (HLS) is done as part of ghcup installation in

make haskell

Settings are automatically synchronized via build-in mechanism so one can just login to GitHub.

The nvidia-settings application does not currently support XDG specification. The workaround is to set custom config path using the --config option.

.envrc defines an alias for this so when running nvidia-settings from shell it will automatically use $XDG_CONFIG_HOME/nvidia-settings/nvidia-settings-rc as the configuration file.

However, the desktop entry (i.e. when running the app from Gnome menu) must be edited manually. One can do this by executing the following make target:

make nvidia-settings-rc-xdg-path

The default user script location (under $HOME) is compiled with the shell, so the only option how to change it is modifying the system-wide configs /etc/bash.bashrc (resp. /etc/bash.bash.logout).

There is a handy make bash target which does so automatically and links custom files from this repository.

Since the shell in use is zsh, bash configs are not commited in this repo. So to change the location of the ~/.bash_history file one can manually run the following:

mkdir -p "$XDG_STATE_HOME"/bash && \
mv ~/.bash_history $XDG_STATE_HOME/bash/history

and make sure this line is in the linked bashrc:

export HISTFILE="${XDG_STATE_HOME:-$HOME/.local/state}"/bash/history

NSS data in ~/.pki should be safe to move to XDG directories, see: bugzilla issue.

Migration commands:

mkdir -p $XDG_CONFIG_HOME/pki/nssdb $XDG_DATA_HOME/pki/nssdb && \
  mv $HOME/.pki/nssdb/pkcs11.txt $XDG_CONFIG_HOME/pki/nssdb && \
  mv $HOME/.pki/nssdb/*.db $XDG_DATA_HOME/pki/nssdb && \
  rm -r $HOME/.pki

Note that ~/.pki might be created by other tools:

• Visual Studio Code
• Google Chrome (should support the XDG_CONFIG_HOME path if it exists)

ipython >= 8.x supports XDG. So it should be sufficient to just run (it's not quite clear which XDG directories one should use):

mv ~/.ipython $XDG_CONFIG_HOME/ipython

Resources:

• Simple installtion tutorial
• Older but comprehensive guide
• Ubuntu wiki

Whether it's even possible to setup KVM on certain system or not can be checked with

kvm-ok

If the CPU supports virtualization one can install all required packages and then run following validation check

virt-host-validate

Typically one can see

• WARN for IOMMU which is handled automatically by the Makefile target. However, this step (and the kvm target in general) needs a reboot.
• Another QUEMU warning which according to Ubintu wiki should be fine

Start Minikube on KVM by running

minikube start --cpus 2 --memory 2048 --vm-driver kvm2

or for larger machnines

minikube start --cpus 4 --memory 8192 --vm-driver kvm2

Note that the test-k8s make target sometimes fails due to timeout after waiting on minikube. It may also leave some resources when it does not succeed.

Ubuntu LTS releases become available with release xx.04.1, as explained here.

The issue manifests when the check indicates there's new LTS release:

❯ sudo do-release-upgrade -cd
Checking for a new Ubuntu release
New release '22.04 LTS' available.
Run 'do-release-upgrade' to upgrade to it.

but it fails to do the release upgrade:

❯ sudo do-release-upgrade
Checking for a new Ubuntu release
There is no development version of an LTS available.
To upgrade to the latest non-LTS development release
set Prompt=normal in /etc/update-manager/release-upgrades.

There are two options to solve this problem:

1. Wait for the xx.04.1 release (~ 3 months from the LTS release)
2. Upgrade to the latest development release, as proposed in the error message. This can be done by sudo do-release-upgrade -d.

With Ubuntu 22.04 a notification might appear reporting "Pending Update of Snap Store".

The issue is that neither from terminal nor from Ubuntu Software application the update works. This is because snap-store starts automatically on login.

So first kill the process (either from command line or using System Monitor), and then run

sudo snap refresh

Note: see all pending updates by running snap refresh --list.

Missing nvidia-smi after some updates (which was working after installation):

❯ nvidi-smi
zsh: command not found: nvidi-smi

Fixed by running following and then rebooting.

sudo ubuntu-drivers autoinstall

Advised here and empirically might work even without restart.

Other option which might work (although one should be careful with this one) is to run dist upgrade:

sudo apt dist-upgrade

The issue with "missing" drivers is that the driver packages are "kept back" (shown in the snippet below). This is due to an incompatibility between driver libraries and the kernel which apt cannot resolve on its own.

❯ sudo apt upgrade
Reading package lists... Done
Building dependency tree
Reading state information... Done
Calculating upgrade... Done
The following packages have been kept back:
  libnvidia-cfg1-460 libnvidia-compute-460 [...]
  libnvidia-gl-460 libnvidia-gl-460:i386 libnvidia-ifr1-460 [...]
  nvidia-utils-460 xserver-xorg-video-nvidia-460
0 upgraded, 0 newly installed, 0 to remove and 21 not upgraded.

As a consequence nvidia-smi is unable to communicate to the driver and fails (see this post).

The solution is as described above (possibly combination of both approaches). Also, reboot might be necessary for the fix to take effect.

Ubuntu 22.04 (LTS) will fallback to and start with X11 on a machine with Nvidia. It's possible, however, to override this rule and run it with Wayland just fine.

Enabling Wayland requires a manual change as described here. Additionally, also modify /etc/gdm3/custom.conf as follows:

PreferredDisplayServer=wayland
WaylandEnable=true
XorgEnable=false

Note that on Ubunut, one can tell the session type from the XDG_SESSION_TYPE environment variable.

If minikube continously fails to boot, and as a last resort, one can clean up the whole minikube by running:

minikube delete

or even more excessive cleanup which also deletes the ~/.minikube dir

minikube delete --purge

The installation of cargo-audit might fail if installed manually and the system does not have libssl-dev installed (note that if one uses make rust-tools this should not happen).

In such case, and possibly in other situations, the advisory database might get into an invalid state.

Fortunately, according to this issue, the fix is simply to drop the database:

rm -rf $CARGO_HOME/advisory-db/

Next invocation of cargo audit will fetch it again and fix the issue.

According to zsh-sdkman's usage notes:

if you do new installations on your sdkman candidate or just play around with new versions, uninstallations or updates, don't forget to refresh the completion script files with the following command:

sdk-refresh-completion-files

According to this issue comment once should be able to fix pipx installations as follows:

pipx reinstall-all

Alternatively, one can reinstall pipx and the tools manually (mentioned in the same issue that was referenced above):

1. Uninstall python tools: pipx uninstall-all
2. Uninstall pipx and remove $XDG_BIN_HOME/pipx and ~/.local/pipx
3. Re-install pipx and tools using make python-tools

An indication for issues with pipx might be an error log mentioning

No module named pip

Some profiling tools (e.g. samply or flamegraph) require access to the performance events system.

These settings can help:

sudo sysctl kernel.perf_event_paranoid=1
sudo sysctl kernel.perf_event_mlock_kb=2048

Some tools (e.g. heaptrack) need adjustment of the ptrace scope:

echo 0 | sudo tee /proc/sys/kernel/yama/ptrace_scope

Also, for Rust binaries it's necessary not to strip debug symbols from the binary. This might be the case when building/running in the --release profile. One possibility is to add a custom profile like the following one:

[profile.profiling]
inherits = "release"
debug = true