Data loss during server restarts in a Node.js application often occurs when in-flight requests or ongoing transactions are terminated abruptly. To avoid this, implementing a graceful shutdown is essential. A graceful shutdown allows your server to finish processing any active requests before it stops accepting new ones and eventually shuts down. You can do this by listening for system signals like `SIGTERM` or `SIGINT` and closing server connections gracefully. Another technique is to persist in-flight data, such as writing incomplete transactions to a temporary storage like Redis or a database, so they can be resumed once the server is back online. Additionally, using load balancers or clustering can help direct new requests to other instances while one instance is restarting, preventing downtime. In serverless environments, handling partial requests at the function level ensures that work is saved and resumed correctly. These techniques ensure your application can handle restarts smoothly without risking data loss or interruption to users.