Nebulous

A cross-cloud container orchestrator

Think of it as a Kubernetes that can span clouds with a focus on accelerated compute and AI workloads. Performant and lightweight via Rust.

Installation

curl -fsSL -H "Cache-Control: no-cache" https://raw.githubusercontent.com/agentsea/nebulous/main/remote_install.sh | bash

Usage

nebu login

Create a container on runpod with 4 A100 GPUs

kind: Container
metadata:
  name: pytorch-test
  namespace: foo
image: pytorch/pytorch:latest
command: nvidia-smi
platform: runpod
env_vars:
  - key: HELLO
    value: world
volumes:
  - source: s3://foo/bar
    dest: /quz/baz
    bidirectional: true
    continuous: true
accelerators:
  - "4:A100"

nebu create container -f examples/basic.yaml

Create a container on EC2 with 1 L40s GPU

nebu create container \
    --name foo \
    --image tensorflow/tensorflow:latest \
    --cmd "echo hello" \
    --platform ec2 \
    --accelerators "1:L40s"

List all containers

nebu get containers

Get one container

nebu get containers foo

Delete a container

nebu delete containers foo

List available accelerators

nebu get accelerators

List available platforms

nebu get platforms

Get the IP address of a container [in progress]

nebu get containers foo --ip

SSH into a container [in progress]

nebu ssh foo

Volumes

Volumes provide a means to persist data accross clouds. Nebulous uses rclone to sync data between clouds backed by an object storage provider.

volumes:
  - source: s3://foo/bar
    dest: /quz/baz
    bidirectional: true
    continuous: true

Organizations

Nebulous is multi-tenant from the ground up. Here is an example of creating a container under the Agentsea organization.

nebu create container \
    --name "Agentsea/foo" \
    --image tensorflow/tensorflow:latest \
    --cmd "echo hello" \
    --platform ec2 \
    --accelerators "1:L40s"

Meters

Metered billing is supported through OpenMeter using the meters field.

meters:
  - cost: 0.1
    unit: second
    currency: USD
    metric: runtime

Clusters [in progress]

Clusters provide a means of multi-node training and inference.

kind: Cluster
metadata:
  name: pytorch-test
  namespace: foo
container:
  image: pytorch/pytorch:latest
  command: "echo $NODES && torchrun ..."
  platform: runpod
  env_vars:
    - key: HELLO
      value: world
  volumes:
    - source: s3://foo/bar
      dest: /quz/baz
      bidirectional: true
      continuous: true
  accelerators:
    - "8:A100"
num_nodes: 4

nebu create cluster -f examples/cluster.yaml

Each container will get a $NODES env var which contains the IP addresses of the nodes in the cluster.

Clusters always aim to schedule nodes as close to each other as possible, with as fast of networking as available.

Services [in progress]

Services provide a means to expose containers on a stable IP address.

Namespaces [in progress]

Namespaces provide a means to segregate groups of resources across clouds. Resources within a given namespace are network isolated using Tailnet, and can be accessed by simply using thier name as the hostname e.g. http://foo:8080.

Contributing

Please open an issue or submit a PR.

nebulous 0.1.21