I would like to run a custom Docker image with GPU on Google Compute Engine.
I have built and pushed the image to the Google Container Registry.
It seems logical to use Container-Optimized OS for a host machine in Google Cloud Engine since I don't need any extra soft on the host machine except Docker, Nvidia GPU drivers and nvidia-container-runtime.
I managed to install nvidia-drivers with this solution.
But I can't run my Docker image with GPU (using --gpu all option) without nvidia-container runtime. This step is specified in official Docker documentation.
Is there a way to install nvidia-container-runtime on Container-Optimized OS in Google Cloud VM?
You don't have to set --gpu all, because this is the default option for nvidia-container-runtime. The assumption, that you don't need anything else is wrong, because it requires libnvidia-container.
To precisely answer the question: No, because libnvidia-container needs to be installed on the OS and nvidia-container-runtime needs to be installed within the K8s container. The one exposes an interface - and the other connects it. And so the one is useless without the other.
Related
I'm currently developing a system that some private libraries. I'm developing in local mode and then when I need to process something specific I use Sagemaker Processing Jobs. The thing is that in order to speed up the process it would be nice to have the possibility of developing everything in a cloud environment.
I'm wondering if is possible to use the same Docker image that I use
for batch processing (the one that I use for Sagemaker Processing Job)
in my Sagemaker Jupyter Notebooks of my cloud environment?
The main problem here is that every time that I work in my cloud Notebooks I have to deal with dependencies conflicts and etc. Using a Docker image would avoid this, and will also allow to each member of the team use the same image to develop in the cloud without having to deal with these kind of conflicts.
You can use the same Docker image to run a processing job locally using SageMaker local mode (basically setting the instance_type parameter on the Processor to local.
However, it sounds like you'd want to use the same image as your dev environment in notebooks. In SageMaker notebook instances, the solution would be to create and maintain conda environments with the same requirements and versions (you can also use LCCs to install a set of packages at notebook start, see some samples here).
An alternative is to use SageMaker Studio, where you can create and bring your own custom image for Studio. There is a detailed tutorial here, and some sample dockerfiles for you to get started here.
I'm preparing to get in to the world of cloud computing.
My first question is:
Is it possible to programmatically create a new, or duplicate an existing VM from my server?
Project Background
I provide a file processing service, and as it's been growing I need to offer a better service.
Project Requirement
Machine specs:
HDD: Min 16gb
CPU: Min 1 core
RAM: Min 2
GB GPU: Min CUDA 10.1 compatible
What I'm thinking is the following steps:
User uploads a file
A dedicated VM is created for that specific file inside Google Cloud Compute
The file is sent to the VM
File is processed using a Anaconda environment
Results are downloaded to local server
Dedicated VM is removed
Results are served to user
How is this accomplished?
PS: I'm looking for resources and advice. Not code.
Your question is a perfect formulation of the concept of Google Cloud Run. At the highest level concept, you create a Docker image (think of it like a VM) and then register that Docker image with GCP Cloud Run. When a trigger occurs, GCP will spin up an instance of that Docker container and pass in information about the cause of that trigger (a file created in GCS or a REST request or others ...). What you do in your container is up to you. You have full power of the Linux environment (under Docker) to do as you like. When your request ends, the container is spun down. You are only billed for the compute resources you use. If your container (VM) isn't being used, you pay nothing until the next trigger.
An alternative to Cloud Run is Cloud Functions. This is a higher level abstraction where instead of providing a Docker container, you provide the body of a function (JavaScript, Java, Python or others) and the request is passed to that function when a trigger occurs. Which you use is mostly personal choice (you didn't elaborate on "File is processed").
References:
Cloud Run
Cloud Functions
I want to use Manjaro in GCP compute engine with GUI but there isn't any image available in compute engine
You need to install Manjaro in your own machine first using something like VirtualBox. Then you can upload the local boot disk to GCS and use that.
See details here: https://cloud.google.com/compute/docs/images/importing-virtual-disks
Seems Manjaro is based on Arch but GCP only supports CentOS/Debian/Red Hat/Ubuntu
(https://cloud.google.com/compute/docs/images/importing-virtual-disks#supported_operating_systems)
Don't if it will work.
For GUI, you need to install X and VNC server then use vnc client to connect.
Besides what Cloud Ace already mentions, there's another resource I believe you might find of interest.
There is actually an ArchLinux dedicated repo in the official GCP GitHub page, which contains some instructions as per how to install ArchLinux into a GCE instance by either using a preconfigured image in the available public images in GCP or building yourself your own custom image. I believe the process of building an image could be attempted with Manjaro, given the fact that it is originally intended for ArchLinux. It has potential to work.
In the end, if the custom image building does not work out with Manjaro, you can always use the Arch Linux public image mentioned in the GitHub I shared (which is the minimal base Arch Linux image), and install the desktop environment you like.
Hope this helps.
I am using the compute engine of the google cloud platform to do computations.
I am using Ubuntu as the OS and every time I create a new instance, I have to install the software I need from scratch, including the build-essential.
I am pretty sure there is a way to specify the software I would like to have in my VM but couldn´t figure out a straightforward way to do it.
You should use GCE custom images to create VM images with pre-installed software that you need.
Alternatively, you can consider using startup scripts in which you can install software during VM startup. But in contrast to custom images it will increase VM startup time, because startup script will be running during VM startup.
I'm a little bit confused, GCP has this new feature Deploying Containers on VMs and Managed Instance Groups which is currently marked as an Alpha release of Containers on Compute Engine and you actually need to request to be whitelisted for this feature.
What I'm struggling with is to understand how is it different from simply choosing Container-Optimized OS in the list of OS images when creating a new CE instance and then running your docker container on that instance? What are the benefits of the new approach?
Container-Optimized OS images have a number of benefits if all you want to do is run containers on your Compute Engine instance.
There is less configuration involved as they come pre-installed and configured with Docker which will already be running as a service when the machine starts.
There is a tick box in the Console when creating a new Container-Optimized OS instance labelled "Deploy a container image to this VM instance". Checking this provides a method of deploying containers/adding images via the Console/Gui and adding settings for commands to be issued to the container, restart policies, environmental variables, host mounts and other mount paths. This essentially allows you to bring up a container at the same time you create your VM.
In general it's more secure as it has a smaller attack surface than a standard VM, as the OS has a smaller footprint. It also includes a 'locked down' firewall and other security settings.
Due to the fact the OS is based on Chromium OS project, and not a full Linux OS, it benefits from automatic updates and comes configured to automatically download weekly updates (a reboot is necessary to install these updates).
So if you want to run containers with minimal setup on a simple operating system with high security, Container-Optimized OS may be suitable.
It should also be said that there are some use cases where these images are not suitable. For example, if you require the flexibility of a full Linux OS (for example, Container-Optimized OS doesn't include a package manager) or if your containers depends on Linux/kernel modules that may not be available in Container-Optimized OS. It would also not be suitable if you wanted your image and OS application to be supported outside of Google Cloud Platform. You would be better off considering public images other than Container-Optimized OS images in these scenarios.