Scroll Proving Agent

Introduction

The Scroll proving agent is a high-performance service client for efficient block proving for layer-2 rollups based on the Scroll SDK. With this integration, Scroll SDK chains gain access to robust GPU-proving capabilities, allowing for the fast and effective proving of blockspaces.

Snarkify specializes in ZK GPU acceleration and is a multi-time winner of the ZPRIZE ZK acceleration competition. Snarkify is the only provider that has developed an in-house GPU prover that is 1.25x more efficient than the official GPU provers. With the in-house GPU proving capability, Snarkify offers best-in-class GPU proving services for both Scroll Mainnet and SDK chains. On a daily basis, Snarkify serves 25% to 50% of the Scroll Mainnet proving jobs.

How It Works

The Scroll Proving agent integrates with the Scroll Proving SDK to handle block proving in three main steps:

1. Task Retrieval: The agent connects to the coordinator endpoint to retrieve proving tasks and to the Geth endpoint for block trace data required for chunk tasks.
2. Proof Generation: Each task is dispatched to the Snarkify platform, where proofs are generated.
3. Task Completion: The agent periodically checks the task’s status on the Snarkify platform and, once complete, submits the proof back to the coordinator.

Getting Started

Install Helm Chart

To deploy the agent, you’ll need Helm. Install Helm with:

curl https://raw.githubusercontent.com/helm/helm/main/scripts/get-helm-3 | bash

Create an API Key

You’ll need an API key to authenticate. Follow the steps here to obtain one.

Set Up Scroll Proving Service

Set up a Scroll Proving service and obtain its service ID. Please contact us on Telegram if you want to set up a service.

Prepare the Configuration

There are 3 types of block proving in Scroll SDK: Chunk, Batch and Bundle, so you need to create 3 different configuration files chunk-config.yaml, batch-config.yaml and bundle-config.yaml. Here is the example of chunk-config.yaml, replacing placeholders as needed:

scrollConfig: |
  {
    "prover_name_prefix": "<Your-Prover-Name-Prefix>",
    "keys_dir": "/keys",
    "coordinator": {
      "base_url": "<Your-Coordinator-Endpoint>",
      "retry_count": 3,
      "retry_wait_time_sec": 5,
      "connection_timeout_sec": 60
    },
    "l2geth": {
      "endpoint": "<Your-Geth-Endpoint>"
    },
    "prover": {
      "circuit_type": 1,
      "circuit_version": "v0.13.1",
      "n_workers": <Your-Number-Of-Workers>,
      "cloud": {
        "base_url": "https://api.snarkify.io",
        "api_key": "<Your-API-Key>",
        "retry_count": 3,
        "retry_wait_time_sec": 5,
        "connection_timeout_sec": 60
      }
    }
  }

env:
  serviceId: "<Your-Service-ID>"

• prover_name_prefix: A prefix for your prover name, should end with an underscore.
• coordinator.base_url: Your coordinator endpoint.
• l2geth.endpoint: Your Geth endpoint.
• prover.circuit_type: The circuit type, can be 1 (Chunk), 2 (Batch) or 3 (Bundle).
• prover.n_workers: The number of workers to run in parallel.
• cloud.api_key: Your Snarkify API key.
• serviceId: Your Scroll Proving service ID in Snarkify platform.

Deploy Scroll Proving Agent

Deploy the agent using Helm with your configuration files.

export HELM_EXPERIMENTAL_OCI=1
helm install scroll-proving-agent-chunk oci://ghcr.io/snarkify/scroll-proving-agent/helm/scroll-proving-agent --version 0.0.1 -f chunk-config.yaml

helm install scroll-proving-agent-batch oci://ghcr.io/snarkify/scroll-proving-agent/helm/scroll-proving-agent --version 0.0.1 -f batch-config.yaml

helm install scroll-proving-agent-bundle oci://ghcr.io/snarkify/scroll-proving-agent/helm/scroll-proving-agent --version 0.0.1 -f bundle-config.yaml

Verify the Deployment

You can verify the deployment by checking the pods and the logs.

kubectl get pods
kubectl logs -l app.kubernetes.io/name=scroll-proving-agent

Uninstall the Helm Chart

helm uninstall scroll-proving-agent-chunk
helm uninstall scroll-proving-agent-batch
helm uninstall scroll-proving-agent-bundle

Community

Please connect with the team on Telegram to learn more.