Ephemeral Containers
Why containers are temporary, and why that's a feature.
The Core Idea
Containers = temporary compute sessions Workspaces = permanent storage
Like shutting down your laptop when done - your files are safe on the disk.
The DS01 Workflow
# Need GPU? Spin up a container
project launch my-thesis --open
# Do your compute-intensive work
# (files saved to /workspace/)
# Done? Clean up
exit
container retire my-thesis # Frees GPU for others
# Need GPU again later? Spin up another container
project launch my-thesis --open
# All your files are exactly where you left them
Container is new. Workspace files persist.
What's Temporary vs Permanent
Temporary (lost on retire):
- Container instance, running processes, pip packages, files outside
/workspace/, GPU allocation
Permanent (always safe):
- Files in
/workspace/, Dockerfile, Docker images, Git history on host
Quick Commands
# Deploy (create + start)
project launch my-project
# Retire (stop + remove + free GPU)
container retire my-project
# Check what you have running
container-list
Mental Model
Think of it like a laptop:
| Laptop | DS01 |
|---|---|
| Shut down when done | container retire |
| Files on SSD | Files in /workspace/ |
| Boot takes 30 sec | Deploy takes 30 sec |
Common Questions
"Will I lose work?"
No, if you save to
/workspace/. That's the whole point of the workspace.
"How long does recreating take?"
About 30 seconds. Image is already built.
"What if I need a GPU for 3 days?"
Keep the container running. Retire when actually done.
"What if container stops unexpectedly?"
Load your checkpoint and resume. Save progress frequently.
Want Deeper Understanding?
This is a brief introduction to the ephemeral model. For comprehensive explanation of:
- Why this design (fairness, reproducibility, industry standards)
- How to handle long-running jobs and checkpointing
- When to use different strategies
- Industry context (AWS, Kubernetes, HPC)
See Ephemeral Philosophy in Educational Computing Context (20 min read).
Next Steps
- Workspaces and Persistence - Where to save files
- Containers and Images - How images work
- Long-Running Jobs - Multi-day training