Run EthProofs as a durable Fleet-managed local worker

Cargo.toml

@@ -50,6 +50,7 @@ provider = { path = "./crates/provider" }
 zkm-build = { git = "https://github.com/ProjectZKM/Ziren" }
 zkm-core-executor = { git = "https://github.com/ProjectZKM/Ziren" }
 zkm-prover = { git = "https://github.com/ProjectZKM/Ziren" }
+zkm-cuda = { git = "https://github.com/ProjectZKM/Ziren" }
 zkm-sdk = { git = "https://github.com/ProjectZKM/Ziren" }
 
 # reth

bin/eth-proofs/Cargo.toml

@@ -4,7 +4,7 @@ name = "eth-proofs"
 edition = "2021"
 
 [dependencies]
-tokio.workspace = true
+tokio = { workspace = true, features = ["signal"] }
 url.workspace = true
 tracing.workspace = true
 serde_json.workspace = true
@@ -13,6 +13,11 @@ tracing-subscriber = "0.3.18"
 dotenv = "0.15.0"
 clap = { version = "4.5.7", features = ["derive", "env"] }
 eyre = "0.6.12"
+bincode = "1.3.3"
+sha2 = "0.10.8"
+hex = "0.4.3"
+fs2 = "0.4.3"
+libc = "0.2"
 reqwest = { workspace = true, features = ["json"] }
 reqwest-middleware = { version = "0.3.3", features = ["json"] }
 reqwest-retry = "0.6.1"
@@ -28,6 +33,8 @@ provider.workspace = true
 
 # Ziren
 zkm-sdk.workspace = true
+zkm-cuda.workspace = true
+zkm-prover.workspace = true
 
 # alloy
 alloy-chains.workspace = true
@@ -41,3 +48,6 @@ zkm-build.workspace = true
 
 [features]
 test = []
+
+[dev-dependencies]
+tempfile = "3"

bin/eth-proofs/build.rs

@@ -1,5 +1,12 @@
-use zkm_build::build_program;
+use zkm_build::{build_program_with_args, BuildArgs};
 
 fn main() {
-    build_program("../guest");
+    println!("cargo:rerun-if-env-changed=ETH_PROOFS_BUILD_ELF");
+    if let Ok(path) = std::env::var("ETH_PROOFS_BUILD_ELF") {
+        let path = std::fs::canonicalize(path).expect("ETH_PROOFS_BUILD_ELF must exist");
+        println!("cargo:rerun-if-changed={}", path.display());
+        println!("cargo:rustc-env=ZKM_ELF_reth={}", path.display());
+    } else {
+        build_program_with_args("../guest", BuildArgs { locked: true, ..Default::default() });
+    }
 }

bin/eth-proofs/src/cli.rs

@@ -14,7 +14,7 @@ pub struct Args {
 
     /// The WS rpc url used to fetch data about the block.
     #[clap(long, env)]
-    pub ws_rpc_url: Url,
+    pub ws_rpc_url: Option<Url>,
 
     /// The debug HTTP rpc url used to fetch data about the block trace.
     #[clap(long, env)]
@@ -28,6 +28,24 @@ pub struct Args {
     #[clap(long, default_value_t = 100)]
     pub block_interval: u64,
 
+    /// Which residue class of `block_interval` this instance proves: blocks
+    /// with `number % block_interval == block_residue`.  Two instances with
+    /// `--block-interval 2` and residues 0 and 1 split the chain between two
+    /// provers and prove two blocks at a time.
+    #[clap(long, default_value_t = 0)]
+    pub block_residue: u64,
+
+    /// How many prepared blocks (fetched + natively executed) may wait for the
+    /// prover.  The fetch of block N+1 always overlaps the proof of block N;
+    /// this bounds how far ahead the fetcher runs.
+    #[clap(long, default_value_t = 1)]
+    pub prefetch_depth: usize,
+
+    /// Skip a block when it is more than this many blocks behind the newest
+    /// header seen (0 = never skip: prove every block the stream delivers).
+    #[clap(long, default_value_t = 0)]
+    pub max_lag: u64,
+
     /// ETH proofs endpoint.
     #[clap(long, env)]
     pub eth_proofs_endpoint: String,
@@ -44,6 +62,34 @@ pub struct Args {
     #[clap(long, env)]
     pub pager_duty_integration_key: Option<String>,
 
+    /// Direct local GPU server, set by Fleet for an integrated app worker.
+    #[clap(long, env = "FLEET_GPU_ENDPOINT")]
+    pub local_gpu_endpoint: Option<Url>,
+
+    /// Persistent local worker state. Required with --local-gpu-endpoint.
+    #[clap(long, env = "FLEET_APP_STORAGE_ROOT", requires = "local_gpu_endpoint")]
+    pub state_dir: Option<std::path::PathBuf>,
+
+    /// Use an existing production guest instead of the embedded ELF.
+    #[clap(long, env = "ETH_PROOFS_ELF_FILE")]
+    pub elf_file: Option<std::path::PathBuf>,
+
+    /// Reject a guest whose digest differs from this app's configured program.
+    #[clap(long, env = "FLEET_APP_ELF_ID")]
+    pub expected_elf_id: Option<String>,
+
+    /// First block for a new durable stream; restarts resume persisted progress.
+    #[clap(long)]
+    pub start_block: Option<u64>,
+
+    /// Witness preparation concurrency (currently 1; GPU proving remains serial).
+    #[clap(long, default_value_t = 1)]
+    pub prepare_concurrency: usize,
+
+    /// Pause proving when this many completed reports await acknowledgment.
+    #[clap(long, default_value_t = 128)]
+    pub max_pending_reports: usize,
+
     /// Moongate server endpoint.
     #[clap(long, env)]
     pub moongate_endpoint: Option<String>,