I am attempting to debug some C# / .NET 5 code in WSL 2 with Ubuntu on Windows. I have WSL 2 setup with Windows 10 and want to test out creating a Systemd service. Unfortunately, it appears Systemd is not enabled with WSL 2 by default, even though a standard Ubuntu install does have it enabled by default. Is there any way to get Systemd enabled in WSL 2?
Note: See footnote at bottom of this answer for background on this Community Wiki.
There are several possible paths to enabling Systemd on WSL2 (but not WSL1). These are summarized here, with more detail provided below.
Option 1: Upgrade WSL to the latest application release (if supported by your system) and opt-in to the Systemd feature
Option 2: Run a Systemd-helper script designed for WSL2
Option 3: Manually run Systemd in its own namespace
And while not part of this question, for those simply looking to run certain applications that require Systemd, there are alternatives:
On WSL1 and WSL2:
Alternative 1: SysVInit scripts (e.g. sudo service <service_name> start) where available
Alternative 2: Manually configuring and running the service
On WSL2-only:
Alternative 3: Docker
Should you enable Systemd in WSL?
First, consider whether you should or need to enable Systemd in WSL. Enabling Systemd will automatically start a number of background services and tasks that you really may not need under WSL. As a result, it will also increase WSL startup times, although the impact will be dependent on your system. Check the Alternatives section below to see if there may be a better option that fits your needs. For example, the service command may do what you need without any additional effort.
More detail on each answer:
Option 1: Upgrade WSL to the latest application release (if supported by your system) and opt-in to the Systemd feature
Microsoft has now integrated Systemd support in the WSL2 application release (as opposed to the older "Windows feature" implementation).
Starting with WSL Application Release 1.0.0, this feature is available on both Windows 10 and Windows 11. Windows 10 users do need to be on UBR (update build revision) 2311 or later. The UBR is the last 4 digits of your full Windows build number (e.g. 10.0.19045.2311 for Windows 10 22H2). 2311 is installed with KB5020030, an optional Preview update, although if you are reading this later, it will likely be a later (non-Preview) monthly servicing update.
If you are on a supported Windows release, the WSL application with Systemd support can be installed:
Through the Microsoft Store (as "Windows Subsystem for Linux").
Or from the Releases page in the Github repo. To install a release manually:
Reboot (to make sure that WSL is not in use at all). A simple wsl --shutdown may work, but often will not.
Download the 1.0.0 (or later) release from the link above.
Start an Administrator PowerShell and:
Add-AppxPackage <path.to>/Microsoft.WSL_1.0.0.0_x64_ARM64.msixbundle
wsl --version # to confirm
To enable, start your Ubuntu (or other Systemd) distribution under WSL (typically just wsl ~ will work).
sudo -e /etc/wsl.conf
Add the following:
[boot]
systemd=true
Exit Ubuntu and again:
wsl --shutdown
Then restart Ubuntu.
sudo systemctl status
... should show your Systemd services.
Option 2: Run a Systemd-helper script designed for WSL2
There are a number of Systemd-enablement scripts available from various sources. Given the complexities involved in running Systemd under WSL, it is recommended that you:
Use one that is actively maintained
Attempt to understand, as much as possible, how they operate, and how they may impact other features and applications in your distribution(s) under WSL
When asking questions here or on any other site, disclose in the question which script you are using so that others can attempt to understand and/or reproduce your issue in the proper context
Several of the more popular projects that enable Systemd under WSL2 are:
Genie: 1.8k stars, last commit September, 2022
Distrod: 1.4k stars, last commit July 2022
WSL2-Hacks: 1.1k stars, mostly instructional, with a supporting script example. Last commit January, 2022
At the core, all of them operate on the same principles covered in the next option ...
Option 3: Manually run Systemd in its own namespace
One of the main issues with running Systemd in earlier versions of WSL is that both inits need to be PID 1. To get around this, it is possible to create a new namespace or container where Systemd can run as PID 1.
To see how this is done (at a very basic level):
Run:
sudo -b unshare --pid --fork --mount-proc /lib/systemd/systemd --system-unit=basic.target
This starts Systemd in a new namespace with its own PID mapping. Inside that namespace, Systemd will be PID1 (as it must, to function) and own all other processes. However, the "real" PID mapping still exists outside that namespace.
Note that this is a "bare minimum" command-line for starting Systemd. It will not have support for, at least:
Windows Interop (the ability to run Windows .exe)
The Windows PATH (which isn't necessary without Windows Interop anyway)
WSLg
The scripts and projects listed above do extra work to get these things working as well.
Wait a few seconds for Systemd to start up, then:
sudo -E nsenter --all -t $(pgrep -xo systemd) runuser -P -l $USER -c "exec $SHELL"
This enters the namespace, and you can now use ps -efH to see that systemd is running as PID 1 in that namespace.
At this point, you should be able to run systemctl.
And after proving to yourself that it's possible, it is recommended that you exit all WSL instances completely, then doing wsl --shutdown. Otherwise, some things will be "broken" until you do. They can likely be "fixed", but that's beyond the scope this answer. If you are interested, please refer to the projects listed above to see how they handle these situations.
Alternative 1: SysVInit scripts (e.g. sudo service <service_name> start) where available
In Ubuntu, Debian, and some other distributions on WSL, many of the common system services still have the "old" init.d scripts available to be used in place of systemctl with Systemd units. You can see these by using ls /etc/init.d/.
So, for example, you can start ssh with sudo service ssh start, and it will run the /etc/init.d/ssh script with the start argument.
Even some non-default packages such as MySql/MariaDB will install both the Systemd unit files and the old init.d scripts, so you can still use the service command for them as well.
On the hand, some packages, like Elasticsearch, only install Systemd units. And some distributions only provide Systemd units for most (if not all) packages in their repositories.
Alternative 2: Manually configuring and running the service
For those services that don't have a init-script equivalent, it can be possible to run them "manually".
For simplicity, let's assume that the ssh init.d script wasn't available.
In this case, the "answer" is to figure out what the Systemd unit files are doing and attempt to replicate that manually. This can vary widely in complexity. But I'd start with looking at the Systemd unit file that you are trying to run:
less /lib/systemd/system/ssh.service
# Trimmed
[Service]
EnvironmentFile=-/etc/default/ssh
ExecStartPre=/usr/sbin/sshd -t
ExecStart=/usr/sbin/sshd -D $SSHD_OPTS
RuntimeDirectory=sshd
RuntimeDirectoryMode=0755
Some of the less relevant lines have been trimmed to make it easier to parse, but you can man systemd.exec, man systemd.service, and others to see what most of the options do.
In this case, when you sudo systemctl start ssh, it:
Reads environment variables (the $SSHD_OPTS) from /etc/default/ssh
Tests the config, exits if there is a failure
Makes sure the RuntimeDirectory exists with the specified permissions. This translates to /run/sshd (from man systemd.exec). This also removes the runtime directory when you stop the service.
Runs /usr/sbin/sshd with options
So, if you don't have any environment-based config, you could just set up a script to:
Make sure the runtime directory exists. Note that, since it is in /run, which is a tmpfs mount, it will be deleted after every restart of the WSL instance.
Set the permissions to 0755
Start /usr/sbin/sshd as root
... And you would have done the same thing manually without Systemd.
Again, this is probably the simplest example. You might have much more to work through for more complex tasks.
Alternative 3: Docker
Many packages/services are available as Docker images. Docker typically runs very well under Ubuntu on WSL2 (specifically WSL2; it will not run on WSL1). If there's not a SysVinit "service" script for the service you are trying to start, there may very well be a Docker image available that runs in a containerized environment.
Example: Elasticsearch, as in this question.
Bonus #1: Doesn't interfere with other packages already installed (no dependency issues).
Bonus #2: The Docker images themselves pretty much never use Systemd, so you can often inspect the Dockerfile to see how the service is started without Systemd. For more information see the next option - "The manual way."
Microsoft recommends Docker Desktop for Windows for running Docker containers under WSL2.
Footnote This answer is being posted as a Community Wiki because it can apply to multiple Stack Overflow questions. It is originally based on answers to this Ask Ubuntu question. However, it is hoped that this wiki-answer can be continuously updated by the community as Systemd evolves on WSL.
This question has been chosen since:
It appears to be the most canonical, straightforward, "How do I enable Systemd on WSL?" question.
It is on-topic, as *creating Systemd services is (or at least can-be) unique to programming.
I run a centos 8 distro on docker and I would like to have bash TAB completion with dnf package manager. According to other posts, I did the following once my docker container is started:
dnf clean all && rm -r /var/cache/dnf && dnf upgrade -y && dnf update -y
and then
dnf install bash-completion sqlite -y
After doing that I restart the container but there is still no bash completion. I also tried to source directly the bash completion file by doing:
source /etc/profile.d/bash_completion.sh
but without any better effect.
Would you know what I am doing wrong ?
You shouldn't need BASH Completion in a Docker container. The only time you should be manually connecting to a shell inside a Linux container is to troubleshoot why the process running in the container is behaving abnormally. In fact, some container design advice might even go as far as suggesting you not include a shell inside your base OS at all!
The reason this isn't working for you is due to the way in which Linux containers operate. A Container is simply a namespaced process that is managed by the kernel installed on the Host OS. This process cannot be modified or interrupted or the container will be destroyed since the process will be sent a SIGTERM. When you attempt to source the bash_completion.sh script, you are attempting to pass new configuration arguments to your existing namespaced process managed by Docker.
If you really wanted to do this the best way to do it would be to create a new Docker Container Image based on the original CentOS 8 Base Image. And then from there install the bash completion package and add an echo command to add the source line to your user's .bashrc file.
EDIT:
With regards to the additional question asked OP in the comments of this answer I have added additional information below.
Why should not I need bash completion in a container
The reason you do not need bash completion in a container is because containers are not meant to be attached to with a shell. A is simply supposed to be a single instance of a process running under specific configured criteria. Containers aren't meant to be used to create dev environments for you to connect to, they're meant to run processes and applications in software infrastructure.
Manually updating & installing packages
You mention that one of the first things you do when you spin up a container is install packages. This is also alarming to me because you are not supposed to be manually interacting with a container at all. This includes package installation. Instead, you should generate a new Container Image from the older Base Image and add additional RUN statements to the Dockerfile to update the system and install these desired packages.
Cannot believe it is not possible
It is possible if you create a new Dockerfile that purposely installs it on a new layer of the base image and produces a new container image for you to use. BUT the point is that you shouldn't be connecting to Docker containers in the first place to even get to a point where you could need something like bash completion!
Here is a great summary on the difference between a container and a virtual machine that might help clarify some of this for you. In a nutshell, containers are supposed to run, and only run, processes.
Using Ubuntu 18.04.
I am trying to install a kubernetes cluster on my local machine (localhost) using this guide (LXD + conjure-up kubernetes):
https://kubernetes.io/docs/getting-started-guides/ubuntu/local/#before-you-begin
When I run:
conjure-up kubernetes
I select the following installation:
and select localhost for "Choose a cloud" and use the defaults for the rest of the install wizard. It then starts to install and after 30-40 minutes it completes with this error:
Here is the log:
https://pastebin.com/raw/re1UvrUU
Where one error says:
2018-07-25 20:09:38,125 [ERROR] conjure-up/canonical-kubernetes - events.py:161 - Unhandled exception in <Task finished coro=<BaseBootstrapController.run() done, defined at /snap/conjure-up/1015/lib/python3.6/site-packages/conjureup/controllers/juju/bootstrap/common.py:15> exception=BootstrapError('Unable to bootstrap (cloud type: localhost)',)>
but that does not really help much.
Any suggestion to why the install wizard/conjure-up fails?
Also based on this post:
https://github.com/conjure-up/conjure-up/issues/1308
I have tried to first disable firewall:
sudo ufw disable
and then re-run installation/conjure install wizard. But I get the same error.
Some more details on how I installed and configured LXD/conjure-up below:
$ snap install lxd
lxd 3.2 from 'canonical' installed
$ /snap/bin/lxd init
Would you like to use LXD clustering? (yes/no) [default=no]:
Do you want to configure a new storage pool? (yes/no) [default=yes]:
Name of the new storage pool [default=default]:
Name of the storage backend to use (btrfs, ceph, dir, lvm) [default=btrfs]:
Create a new BTRFS pool? (yes/no) [default=yes]:
Would you like to use an existing block device? (yes/no) [default=no]:
Size in GB of the new loop device (1GB minimum) [default=26GB]:
Would you like to connect to a MAAS server? (yes/no) [default=no]:
Would you like to create a new local network bridge? (yes/no) [default=yes]:
What should the new bridge be called? [default=lxdbr0]:
What IPv4 address should be used? (CIDR subnet notation, “auto” or “none”) [default=auto]:
What IPv6 address should be used? (CIDR subnet notation, “auto” or “none”) [default=auto]:
Would you like LXD to be available over the network? (yes/no) [default=no]:
Would you like stale cached images to be updated automatically? (yes/no) [default=yes]
Would you like a YAML "lxd init" preseed to be printed? (yes/no) [default=no]:
Configured group membership:
sudo usermod -a -G lxd $USER
newgrp lxd
Next installed:
sudo snap install conjure-up --classic
And then ran installation:
conjure-up kubernetes
I wasn't able to reproduce your exact problem but i got conjure-up + lxd installed and in the end Kubernetes on my newly installed VirtualBox Ubuntu 18.04 (Desktop) VM. Hopefully this answer could help you somehow!
I looked through the kubernetes.io documentation page and that one lacked tiny bits of information, it does mention lxd but not the part with lxd init which i assume you picked up in the conjure-up user manual.
So with that said, i followed the conjure-up user manual with some minor changes on the way. I'm assuming that it's OK for you to use the edge version of conjure-up, i started off with the stable one but changed to edge when testing different combinations.
Also please ensure that you have the recommended resources available stated by the user manual, conjure-up and the Canoncial Distribution of Kubernetes launches a number of containers for you. You might not need 3 x etcd, 3 x worker nodes and 2 x Master, and if you don't just tune the number of containers down in the conjure-up wizard.
These are the steps i performed (as my local user):
Make sure your Ubuntu box are updated: sudo apt update && sudo apt upgrade
Install conjure-up by running: sudo snap install conjure-up --classic --edge
Install lxd by running: sudo snap install lxd
With lxd comes the client part which is lxc, if you run e.g. lxc list you should get an empty table (no containers started yet). I got an permission error at this time, i ran the following: sudo chown -R lxd:lxd /var/snap/lxd/ to change owner and group of the lxd directory containing the socket you'll be communicating with using lxc.
Add your user to the lxdgroup: sudo usermod -a -G lxd $USER && newgrp lxd, log off and on to make this permanent and not only active in your current shell.
Now create a lxd bridge manually with the following command: lxc network create lxdbr1 ipv4.address=auto ipv4.nat=true ipv6.address=none ipv6.nat=false
Now let's run the init part of lxd with lxd init. Remember to answer no when being asked to create a new local network bridge?, in the next prompt provide your newly created network bridge instead (lxdbr1). The rest of the answers to the questions can be left as default.
Now continue with running conjure-up kubernetes and choose localhost as your type. For me the localhost choice was greyed out from the beginning, it worked when i created the network bridge manually and not via the lxd init step.
Skip the additional components you can install like Rancher, Prometheus etc.
Choose your new network bridge and the default storage pool, proceed to the next step.
In the next step customize your Kubernetes cluster if needed and then hit Deploy. And now you wait!
You can always troubleshoot and list all containers created with the lxc tool. If you've ever used Docker the lxc tool feels a lot like the docker client.
And finally some thoughts and observations, there's a lot of moving parts to conjure-up as you might have seen. It's actually described as: conjure-up is a thin layer spanning a few different underlying technologies - Juju, MAAS and LXD.
For reference, i ended up having the following versions installed:
lxd version 3.3
conjure-up version 2.6.1
from this example. DockerOperator has the docker_url parameter which is "URL of the host running the docker daemon.".
But when i run in Kubernetes engine on Google Cloud Platform, how can i find this docker_url on Kubernetes?
You can run the following command to find out the docker url:
$docker-machine url [docker_machine_name]
Docker machine is not installed on the container images by default. You will have to install docker-machine manually by following these steps.
You will also have to use the Ubuntu image if you would like this functionality. I tried to install docker machine using a cos image, and it does not work since the image does not have the necessary dependencies.
There is a datastorage, an mysql container, a php and a nginx. Is it possible to let these processes run on different oses?
So one is on debian, the other on centos and so on?
Example
this one is debian
docker run --name sql -d buildsql
this one is centos
docker run --name php --linked sql:db -d buildphp
Containers talk to each other over the network, so they are normally unaware of the OS being used by other containers, in exactly the same way that your browser doesn't really care about the OS of the webservers it talks to.
Most of the official images are based on Debian, so you quite often find your containers are all running Debian, but there's no need for this to be true. Some containers don't have an OS at all and just contain a binary that gets run when the container starts.
In short, there is no problem in using different OSs, unless you have some funky application specific problem with networking.