checkpointing

Checkpointing Examples

This directory contains examples demonstrating how to use Checkpointing in LangGraphGo to persist and manage graph state.

1. Background

In complex, long-running, or critical applications, keeping state solely in memory is risky and limiting. Checkpointing solves several key problems:

• Fault Tolerance: If the application crashes, you can resume from the last saved state.
• Human-in-the-loop: You can pause execution, wait for days for human input, and then resume.
• Time Travel: You can inspect past states ("what happened at step 3?") or even fork execution from a previous point.

2. Key Concepts

• CheckpointSaver: An interface for saving and loading graph state. LangGraphGo provides implementations for:
  • Memory: Ephemeral, good for testing.
  • PostgreSQL: Robust, production-grade persistence.
  • SQLite: Lightweight, file-based persistence.
  • Redis: Fast, in-memory persistence.
• ThreadID: A unique identifier for a conversation or execution thread. Checkpoints are isolated by ThreadID.
• CheckpointConfig: Configuration options like AutoSave, SaveInterval, and MaxCheckpoints.

3. Examples

In-Memory (main.go)

Demonstrates the basic API using an in-memory store. Good for understanding the concepts without setting up a database.

PostgreSQL (postgres/)

Shows how to use a PostgreSQL database for durable state storage. Requires a running Postgres instance.

SQLite (sqlite/)

Shows how to use a local SQLite file. Ideal for desktop apps or simple deployments.

Redis (redis/)

Shows how to use Redis for high-performance state storage.

4. How to Use

To run the PostgreSQL example:

export POSTGRES_CONN_STRING="postgres://user:password@localhost:5432/dbname"
cd postgres
go run main.go

To run the SQLite example:

cd sqlite
go run main.go