.png%3Falt%3Dmedia%26token%3Db9e6d2df-4021-410f-851f-17dc1cc69e2f)
.png%3Falt%3Dmedia%26token%3Db9e6d2df-4021-410f-851f-17dc1cc69e2f)
CoreWeave
Search…
CoreWeave Kubernetes
Virtual Servers
Workflows
Solutions
SSH Server with CUDA
SSH server with root disk persistence useful for development and testing
Introduction
This example deploys a Kubernetes Deployment with a Linux container including the CUDA Runtime and development tools, including CUDNN as well as running an SSH server. The Deployment allocates 6 GPUs. The Pod includes an
initContainer that copies the entire root system to a persistent volume on boot, and then mounts all directories on subsequent boots. This gives a VM-like container where manually installed applications will survive a restart.The Kubenetes Control-Plane in the CoreWeave Cloud will ensure that the Deployment is continuously running. The Control-Plane will reserve GPU, CPU and RAM on CoreWeaves compute nodes. The example Deployment does showcase some node affinity rules. These are purely for demonstration purposes, and the entire affinity section can be removed without breaking the example.
Service
A Service is included to show how to publish a Pod to the public Internet. The Service publishes the SSH server to the Internet.
Persistent Storage
A Persistent Volume is allocated to the Pods root directories. The allocation is done via a Persistent Volume Claim requesting a 200GB storage size and SSD backing. This volule claim is then mounted to the the different root directories in the Deployment definition. A separate, HDD backed 500GB Volume Claim is mounted under
/mnt/data for data storage. Utilizing a persistent volume ensures that files persist even if the node currently running the Pod fails.Getting Started
Clone the repository. After installing
kubectl and adding your CoreWeave Cloud access credentials, the following steps will deploy the components.- 1.Apply the resources. This can be used to both create and update existing manifests1$ kubectl apply -f sshd-root-pvc.yaml2persistentvolumeclaim/sshd-root-pv-claim configured3$ kubectl apply -f sshd-data-pvc.yaml4persistentvolumeclaim/sshd-data-pv-claim configured5$ kubectl apply -f sshd-service.yaml6service/sshd configured7$ kubectl apply -f sshd-deployment.yaml8deployment.apps/sshd configuredCopied!
- 2.List pods to see the Deployment working to instantiate all our requested instances1$ kubectl get pods2NAME READY STATUS RESTARTS AGE3sshd-7b5f48f555-5h5g6 0/1 ContainerCreating 0 2sCopied!
- 3.After a little while, all pods should transition to the
Runningstate1$ kubectl get pods2NAME READY STATUS RESTARTS AGE3sshd-7b5f48f555-5h5g6 0/1 Running 0 6sCopied!
Show the log for the init container to grab the auto-generated root password
1
$ kubectl logs -f sshd-7b5f48f555-5h5g6 init
2
Root password is: BdxbmsVsiONQY Initialization complete
Copied!
To get the public IP assigned to the service, simply list all services
1
$ kubetl get service git:(master↓3|…
2
NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE
3
sshd LoadBalancer 10.134.100.93 64.79.105.198 22:30877/TCP 63m
Copied!
The external IP is allocated for the life of the service, and will not change unless the service is deleted. You can now SSH to the SSH server running inside the Pod from the Internet.
1
$ ssh [email protected]
2
[email protected]'s password:
3
[email protected]:~# ls
4
lost+found
5
[email protected]:~# nvidia-smi
6
Wed Oct 30 21:02:14 2019
7
+-----------------------------------------------------------------------------+
8
| NVIDIA-SMI 430.50 Driver Version: 430.50 CUDA Version: 10.1 |
9
|-------------------------------+----------------------+----------------------+
10
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
11
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
12
|===============================+======================+======================|
13
| 0 NVIDIA Graphics... Off | 00000000:01:00.0 Off | N/A |
14
| 0% 28C P8 5W / 180W | 15MiB / 8119MiB | 0% Default |
15
+-------------------------------+----------------------+----------------------+
16
| 1 NVIDIA Graphics... Off | 00000000:02:00.0 Off | N/A |
17
| 0% 30C P8 5W / 180W | 9MiB / 8119MiB | 0% Default |
18
+-------------------------------+----------------------+----------------------+
19
| 2 NVIDIA Graphics... Off | 00000000:03:00.0 Off | N/A |
20
| 0% 33C P8 5W / 180W | 9MiB / 8119MiB | 0% Default |
21
+-------------------------------+----------------------+----------------------+
22
| 3 NVIDIA Graphics... Off | 00000000:04:00.0 Off | N/A |
23
| 0% 34C P8 5W / 180W | 9MiB / 8119MiB | 0% Default |
24
+-------------------------------+----------------------+----------------------+
25
| 4 NVIDIA Graphics... Off | 00000000:08:00.0 Off | N/A |
26
| 0% 28C P8 5W / 180W | 9MiB / 8119MiB | 0% Default |
27
+-------------------------------+----------------------+----------------------+
28
| 5 NVIDIA Graphics... Off | 00000000:09:00.0 Off | N/A |
29
| 0% 31C P8 5W / 180W | 9MiB / 8119MiB | 0% Default |
30
+-------------------------------+----------------------+----------------------+
31
​
32
+-----------------------------------------------------------------------------+
33
| Processes: GPU Memory |
34
| GPU PID Type Process name Usage |
35
|=============================================================================|
36
+-----------------------------------------------------------------------------+
37
Copied!
The SSH Pod can be used as a simple Linux machine. You can install applications and libraries that will be persisted on restart. You will however not be able to do the following
- 1.Run Docker images. Docker images need to be run as their own Deployments, Pods or Jobs. If you need to run Docker images for development from inside an SSH container, please contact support.
- 2.Run systemd services. There is no systemd running in the container.
Last modified 1yr ago
Copy link