Pods

Manage GPU containers using CLI commands.

Create Pod

Interactive mode is supported:

podstack pod create

The wizard guides you through project selection, image/template, compute resources, GPU options, and billing period.

Basic Creation

podstack pod create \
  --name my-pod \
  --image pytorch/pytorch:latest \
  --gpu-type A100

Full Options

podstack pod create \
  --name training-pod \
  --project my-project \
  --image pytorch/pytorch:2.0.0-cuda11.7-cudnn8-runtime \
  --gpu-type A100 \
  --gpu-count 2 \
  --cpu 8 \
  --memory 64 \
  --disk 200 \
  --port 8888 \
  --port 6006 \
  --env WANDB_API_KEY=xxx \
  --env HF_TOKEN=xxx \
  --volume vol-123:/data \
  --command "jupyter lab --ip=0.0.0.0" \
  --label team=ml \
  --label experiment=gpt \
  --wait

Create Options

Flag	Description
`--name`	Pod name (required)
`--image`	Container image (required)
`--gpu-type`	GPU type: A100, H100, V100, L40S, T4
`--gpu-count`	Number of GPUs (default: 1)
`--cpu`	CPU cores (default: 4)
`--memory`	Memory in GB (default: 16)
`--disk`	Disk size in GB (default: 50)
`--port`	Expose port (can repeat)
`--env`	Environment variable KEY=VALUE (can repeat)
`--volume`	Mount volume ID:PATH (can repeat)
`--command`	Startup command
`--label`	Label KEY=VALUE (can repeat)
`--project`	Project ID or name
`--wait`	Wait for pod to be running
`--timeout`	Wait timeout in seconds

List Pods

# All pods
podstack pod list

# Filter by status
podstack pod list --status running
podstack pod list --status stopped

# Filter by project
podstack pod list --project my-project

# Filter by label
podstack pod list --label team=ml

# Output formats
podstack pod list --output table  # default
podstack pod list --output json
podstack pod list --output yaml
podstack pod list --output wide   # more columns

# IDs only (for scripting)
podstack pod list --quiet

Get Pod Details

# Basic info
podstack pod get my-pod

# JSON output
podstack pod get my-pod --output json

# Watch for changes
podstack pod get my-pod --watch

Pod Lifecycle

Start Pod

podstack pod start my-pod

# Wait for running
podstack pod start my-pod --wait

Stop Pod

# Graceful stop
podstack pod stop my-pod

# Force stop
podstack pod stop my-pod --force

Restart Pod

podstack pod restart my-pod

Delete Pod

# Delete stopped pod
podstack pod delete my-pod

# Force delete running pod
podstack pod delete my-pod --force

# Delete without confirmation
podstack pod delete my-pod --yes

Connect to Pod

SSH

# Interactive SSH
podstack pod ssh my-pod

# SSH as different user
podstack pod ssh my-pod --user ubuntu

# SSH with specific key
podstack pod ssh my-pod --identity ~/.ssh/my_key

Execute Commands

# Single command
podstack pod exec my-pod -- nvidia-smi

# Multiple commands
podstack pod exec my-pod -- bash -c "cd /workspace && python train.py"

# Interactive
podstack pod exec my-pod -it -- /bin/bash

# With timeout
podstack pod exec my-pod --timeout 3600 -- python long_training.py

Logs

# Recent logs
podstack pod logs my-pod

# Last N lines
podstack pod logs my-pod --tail 100

# Follow logs
podstack pod logs my-pod --follow

# Since timestamp
podstack pod logs my-pod --since 2024-01-15T10:00:00Z

# Since duration
podstack pod logs my-pod --since 1h

Stats / Metrics

Retrieve real-time resource utilization statistics for a pod:

podstack pod stats my-pod

# JSON format
podstack pod stats my-pod -o json

Scaling

Scale the number of replicas dynamically:

# Scale to 3 replicas
podstack pod scale my-pod 3

# Scale down to 1
podstack pod scale my-pod 1

Batch Operations

Stop All Running Pods

podstack pod list --status running --quiet | xargs -I {} podstack pod stop {}

Delete All Stopped Pods

podstack pod list --status stopped --quiet | xargs -I {} podstack pod delete {} --yes

Examples

Quick Training Job

# Create, train, cleanup
podstack pod create --name train --image pytorch/pytorch:latest --gpu-type A100 --wait
podstack pod exec train -- python /workspace/train.py
podstack pod delete train --force

Jupyter Development

podstack pod create \
  --name jupyter-dev \
  --image jupyter/pytorch-notebook \
  --gpu-type A100 \
  --port 8888 \
  --command "jupyter lab --ip=0.0.0.0 --allow-root --NotebookApp.token=''" \
  --wait

# Get Jupyter URL
podstack pod get jupyter-dev --output json | jq -r '.jupyter_url'

Next Steps

Storage - Bucket and volume commands
GPUs & Templates - Check compute availability
Configuration - CLI settings