Fork

When you hibernate a sandbox, we keep a memory snapshot of the underlying Firecracker VM. When you start that sandbox, or if any network request is made to the sandbox, we'll restore the memory snapshot and continue from where you left off. Because we can snapshot a sandbox, we can also fork it! For example, you can fork a sandbox to create a new sandbox running the same server.

import { CodeSandbox } from '@codesandbox/sdk'
const sdk = new CodeSandbox();
 
const sandbox = await sdk.sandbox.create();
 
// Run anything on the sandbox
await sandbox.shells.run('echo test > test.txt');
 
const sandbox2 = await sandbox.fork();
 
// Now we have two sandboxes that have the same fs & memory state!

You can use this to add support for checkpoint/restore functionality, or A/B test different agent iterations. At CodeSandbox we use this to enable users to quickly fork shared Sandboxes to their own account.

Manually Creating a Memory Snapshot

You can manually create a memory snapshot by calling sandbox.hibernate():

import { CodeSandbox } from '@codesandbox/sdk'
const sdk = new CodeSandbox();
 
const sandbox = await sdk.sandbox.create();
 
// Do work
 
await sandbox.hibernate();

Creating a memory snapshot can take between 3-10 seconds. Resuming from a memory snapshot takes between 0.5-2 seconds.

Live snapshots

If a Sandbox is already running we can still fork its exact current state. This has a small overhead of about 0.5 seconds.

Learn More

We have written a couple blog posts about how memory snapshots work under the hood:

Persistence FAQ