PERSISTENT_SANDBOX_SETUP.txt

PERSISTENT SANDBOX SETUP GUIDE
================================

Date: 2024-11-30
Purpose: Configure Manim renderer to reuse existing sandbox instead of creating new ones

OVERVIEW
--------
The renderer now connects to a single persistent Blaxel sandbox instead of creating
temporary sandboxes for each render. This eliminates:
- Sandbox creation overhead (2-5 seconds per render)
- 503 Service Unavailable errors when API is busy
- Resource quota issues from creating too many sandboxes
- Cleanup complexity

CONFIGURATION
-------------

Required Environment Variables (.env file):

1. MANIM_SANDBOX_NAME=manim-sandbox
   - Name of your deployed persistent sandbox
   - This sandbox stays alive between renders
   - Default: "manim-sandbox"

2. MANIM_SANDBOX_IMAGE=sandbox/manim-sandbox:hpsksd5b50u7
   - Image used by your sandbox
   - Only used for verification, not creation
   - Your deployed sandbox already uses this image

3. BLAXEL_API_KEY=bl_aaabzou85ml4qgfryp6qfvne1b5u8s5o
   - Your Blaxel API key for authentication

4. BLAXEL_SANDBOX_URL=https://sbx-python-app-c8wjh6.us-pdx-1.bl.run
   - URL of your deployed sandbox (optional)

DEPLOYMENT
----------

Your sandbox "manim-sandbox" is already deployed with:
✓ Manim 0.18.1+
✓ FFmpeg with full codec support
✓ LaTeX (texlive packages)
✓ Python 3.12 + all dependencies
✓ Status: DEPLOYED in us-pdx-1 region

To check your deployed sandboxes:
  bl get sandboxes

To redeploy if needed:
  ./deploy_sandbox.sh

HOW IT WORKS
------------

Old Behavior (REMOVED):
1. Create new sandbox for each render
2. Wait for sandbox initialization (2-5 seconds)
3. Install dependencies (if not in image)
4. Run render
5. Download output
6. Delete sandbox

New Behavior (CURRENT):
1. Connect to existing "manim-sandbox" (instant)
2. Verify Manim/FFmpeg available (~5 seconds)
3. Run render
4. Download output
5. Keep sandbox alive for next render

BENEFITS
--------
✓ No creation overhead - instant connection
✓ No 503 errors from sandbox creation API
✓ No deletion/cleanup needed
✓ Consistent environment across renders
✓ Faster turnaround time
✓ Lower API quota usage

RENDER FLOW
-----------

Step 1: Connect to Sandbox
  - SandboxInstance.get("manim-sandbox")
  - Verifies sandbox exists and is accessible
  - Logs: "Successfully connected to sandbox: manim-sandbox"

Step 2: Quick Verification (30 seconds timeout)
  - Checks: python3 -c "import manim; print('Manim version:', manim.__version__)"
  - Checks: ffmpeg -version | head -n 1
  - Logs: "✓ Custom sandbox verified: Manim and FFmpeg are available"

Step 3: Write Manim Code
  - Creates /tmp/{scene_name}.py in sandbox
  - Contains your animation scene class

Step 4: Render Animation
  - Runs: manim {quality_flag} --fps {frame_rate} -o {output}.mp4 {file}.py {scene}
  - Timeout: 600 seconds (10 minutes)
  - Outputs to: /tmp/media/videos/{scene_name}/

Step 5: Download Output
  - Reads rendered file from sandbox
  - Saves to local outputs/ directory

Step 6: Cleanup
  - Keeps sandbox alive (no deletion)
  - Files remain in /tmp for next render (no conflict)

PERFORMANCE
-----------

Typical render times:
- Sandbox connection: 0.5-1 second
- Dependency verification: 5-10 seconds
- Simple animation: 30-60 seconds
- Complex animation: 2-10 minutes

Total overhead: ~5-10 seconds (vs 2-5 minutes with installation)

TROUBLESHOOTING
---------------

Error: "Sandbox 'manim-sandbox' not found"
Solution:
  - Check deployed sandboxes: bl get sandboxes
  - Deploy if missing: ./deploy_sandbox.sh
  - Verify MANIM_SANDBOX_NAME in .env matches deployed name

Error: "Custom sandbox verification failed"
Solution:
  - Sandbox might be in bad state
  - Restart sandbox: bl restart sandbox manim-sandbox
  - Or redeploy: ./deploy_sandbox.sh

Error: "Authentication required"
Solution:
  - Check BLAXEL_API_KEY is correct in .env
  - Login to Blaxel CLI: bl login
  - Verify key: echo $BLAXEL_API_KEY

Error: Render hangs or times out
Solution:
  - Check sandbox status: bl get sandboxes manim-sandbox
  - View sandbox logs: bl logs manim-sandbox
  - Restart if needed: bl restart sandbox manim-sandbox

SANDBOX MAINTENANCE
-------------------

Restart sandbox (if it becomes unresponsive):
  bl restart sandbox manim-sandbox

View sandbox logs:
  bl logs manim-sandbox

Check sandbox status:
  bl get sandboxes manim-sandbox

Redeploy sandbox (if corrupted):
  ./deploy_sandbox.sh
  # Update MANIM_SANDBOX_NAME in .env if name changes

Delete and recreate (nuclear option):
  bl delete sandbox manim-sandbox
  ./deploy_sandbox.sh

VERIFICATION
------------

Run the verification script:
  python3 verify_sandbox_setup.py

Expected output:
  ✓ Custom sandbox image configured: sandbox/manim-sandbox:hpsksd5b50u7
  ✓ Persistent sandbox name configured: manim-sandbox
  ✓ Blaxel API key configured: bl_aaabz...
  ✓ All checks passed! Your setup is ready.

Then test a render:
  python3 app.py
  # Or: python3 main.py

Monitor logs for:
  - "Connecting to persistent sandbox: manim-sandbox"
  - "Successfully connected to sandbox: manim-sandbox"
  - "✓ Custom sandbox verified: Manim and FFmpeg are available"
  - No creation or installation messages

FILE STATE
----------

Files in /tmp/ persist between renders on the same sandbox.
This is fine because:
- Each render uses unique scene names as filenames
- Manim overwrites existing files with same name
- /tmp/ is large enough for multiple outputs
- No cleanup needed between renders

If you want to clear /tmp/:
  # Connect to sandbox and run cleanup
  bl exec manim-sandbox "rm -rf /tmp/*.py /tmp/media/*"

MIGRATION NOTES
---------------

Changed from:
  - SandboxInstance.create() → SandboxInstance.get()
  - New sandbox per render → Reuse one sandbox
  - Delete after render → Keep alive

This means:
  - Faster renders (no creation time)
  - More reliable (no 503 errors)
  - Simpler code (no creation/deletion logic)
  - Persistent state (files remain in /tmp)

IMPORTANT
---------
Keep your sandbox running! Don't manually delete "manim-sandbox" unless you're
redeploying. The renderer expects it to be always available.

If you accidentally delete it:
  ./deploy_sandbox.sh
  # Verify: bl get sandboxes
  # Update .env if name changed