Security Groups & Firewalls
NovaPod Security Groups act as virtual firewalls controlling inbound and outbound network traffic to your GPU instances at the hypervisor level.
Default Behavior
- Ingress: All ports are blocked by default except the ports explicitly defined in your instance launch configuration.
- Egress: All outbound traffic is permitted by default, allowing instances to download models from Hugging Face, pull packages via apt/pip, and push metrics to Wandb.
Creating a Rule Set
You can create named security groups and apply them across multiple instances simultaneously:
# Create a security group for an OpenAI-compatible serving cluster
np sg create --name llm-api-sg --description "Allow SSH from office and public HTTPS"
# Add rule: Allow SSH only from your static office IP
np sg add-rule llm-api-sg \
--direction ingress \
--protocol tcp \
--port 22 \
--cidr 203.0.113.50/32
# Add rule: Allow public HTTPS traffic to API port 8000
np sg add-rule llm-api-sg \
--direction ingress \
--protocol tcp \
--port 8000 \
--cidr 0.0.0.0/0
Attaching to an Instance
np launch \
--gpu H100_SXM \
--security-group llm-api-sg \
--name secure-serving-node