Skip to content

Latest commit

 

History

History

snark_worker

README.md

Snark Worker Library

This directory contains the Mina SNARK worker library implementation. The SNARK worker is responsible for generating zero-knowledge proofs (SNARKs) required by the Mina protocol.

Overview

The SNARK worker operates as either:

  1. An integrated service within the Mina daemon
  2. A standalone process (see the standalone/ directory)

SNARK workers generate proofs for transactions and receive fees for their work, creating an economic incentive for proof generation in the Mina network.

High-Level Design

The SNARK worker follows a simple polling loop:

  1. Request work – The worker calls Get_work RPC on the daemon to obtain a Spec.Partitioned.t. The daemon's work partitioner partitions pending scan-state jobs into individually-provable chunks before handing them out.

  2. Generate proof – Prod.Impl.perform_partitioned drives the actual proof generation. Depending on the configured proof level it either:

    • Runs the full SNARK prover (Transaction_snark) for a Single spec or a Sub_zkapp_command segment/merge spec, or
    • Returns a dummy proof when the proof level is Check or No_check.
  3. Submit result – The worker calls Submit_work RPC on the daemon with the completed Result.Partitioned.t. The daemon's work partitioner combines partial sub-zkapp results and, once a full work unit is assembled, adds it to the snark pool.

  4. Report failures – If proof generation fails, the worker calls Failed_to_generate_snark RPC so the daemon can reassign the work to another worker.

The worker retries all RPC calls with jitter-based back-off. When no work is available it naps for a randomised interval to avoid busy-polling.

Worker State

Prod.Impl.Worker_state.t is created once at startup and holds:

  • The proof-level variant (Full, Check, or No_check) together with the instantiated Transaction_snark module.
  • A proof-cache database (Proof_cache_tag.cache_db) for storing intermediate ZK-app segment proofs to disk, avoiding out-of-memory issues for large ZK-app commands.
  • Logger and signature-kind configuration.

Library Structure

  • snark_worker.ml – Top-level module re-exporting RPCs, entry point, and the production implementation.
  • intf.ml – Legacy interface signatures (Work_S / Rpcs_versioned_S) from the original RPC design. The current RPC protocol is defined in the individual rpc_*.ml modules using Spec.Partitioned / Result.Partitioned types.
  • prod.ml – Production implementation: Worker_state creation and the perform_partitioned / perform_single proof-generation functions.
  • entry.ml – The main polling loop and the command_from_rpcs CLI command used by the integrated worker.
  • events.ml – Structured-log event definitions.
  • rpc_get_work.ml – Versioned RPC definition for requesting work from the daemon.
  • rpc_submit_work.ml – Versioned RPC definition for submitting completed proofs to the daemon.
  • rpc_failed_to_generate_snark.ml – Versioned RPC definition for reporting proof-generation failures.

Standalone Worker

For information about the standalone SNARK worker executable, see the standalone README.

Usage

The library is primarily used:

  1. Internally by the Mina daemon (entry.ml / command_from_rpcs) to run an integrated SNARK worker process.
  2. By external SNARK workers that connect to a Mina daemon via the versioned RPC protocol to perform proof generation independently.

SNARK workers communicate with the Mina daemon via versioned RPCs:

RPC Direction Purpose
Get_work Worker → Daemon Poll for a partitioned work spec
Submit_work Worker → Daemon Submit a completed proof
Failed_to_generate_snark Worker → Daemon Report a proof-generation error

For a high-level description of how the SNARK worker interacts with the SNARK pool and the rest of the daemon, see docs/snark-worker-and-pool.md.