repository·master·Indexed 23 days ago
https://github.com/nethermindeth/nethermindA high-performance Ethereum execution client built on .NET, supporting Ethereum, Gnosis, and L2s like Optimism and Linea. Features include a modular plugin system for extending consensus engines and RPC namespaces, Snap sync for fast synchronization, and a dedicated RPC benchmarking suite for measuring state-reading JSON-RPC performance. The project also includes Nethermind.Kademlia, providing generic routing-table and lookup primitives.
The RPC benchmarking scripts are designed to measure an execution client's state-reading JSON-RPC performance (e.g., eth_call, eth_getBalance, trace_*, debug_*).
Key features include:
RpcTests.Generator is a tool that converts JSONL request recordings into RPC test case files compatible with the Erigon rpc-tests format. It supports two operational modes based on how the Nethermind RPC recorder was configured:
--JsonRpc.RpcRecorderState All. In this mode, the JSONL file already contains both requests and responses; the tool simply matches them by ID.--JsonRpc.RpcRecorderState Request. In this mode, the file contains only requests. The tool will send these requests to a provided reference client URL (--client) and record the responses.The output is a JSON file containing the request and response pairs.
[
{
"request": {
"jsonrpc": "2.0",
"id": 1,
"method": "eth_blockNumber",
"params": []
},
"response": {
"jsonrpc": "2.0",
"id": 1,
"result": "0x12345"
}
}
]newHeads via WebSocket and evaluating test conditions on every new block.When annotations are enabled, each instruction line includes an end-of-line comment containing performance metrics in the following format:
[TP:0.25 | Lat: 1 | Uops:1 | 1*FP_LD]
| Field | Meaning | Description |
|---|---|---|
TP | Reciprocal throughput | Cycles per instruction. Lower is faster. |
Lat | Latency | Cycles from input ready to output ready. |
Uops | Micro-op count | Number of micro-ops; fewer reduces execution engine pressure. |
Ports | Execution port usage | Shows functional units used (e.g., 1*p0156). |
Note on Mappings: The tool automatically maps Intel-style mnemonics (used by .NET JIT) to canonical uops.info forms (e.g., je $\rightarrow$ jz).
Nethermind.Kademlia provides generic Kademlia routing-table and lookup primitives. It is designed to be decoupled from the specific transport, authentication, and node record implementations of your protocol.
What it provides:
What you must provide:
The PersistedSnapshotMerger performs a compaction/merge of multiple sorted snapshot runs.
Merge Logic:
from_* fields from the oldest source with to_* and version from the newest source.XDC uses a 3-chain rule to finalize blocks. A block (the grandparent) is committed and finalized when three consecutive rounds of blocks exist.
Conditions for Finalization:
To finalize a grandparent based on a proposedBlock:
proposedRound - 1 == parent.roundproposedRound - 2 == grandparent.roundproposedRound > grandparent.round + 1Visual Flow:
Test files use a JSON format where requests and responses can be parameterized using C# expressions wrapped in {{ expression }}. These expressions are evaluated using DynamicExpresso and can access helpers from TestContext (e.g., EveryBlocks, Recent, Hex, Topic).
run (required): A boolean expression evaluated once per new head to decide if the test should run.request (required): The JSON-RPC request object with parameterized properties.response (optional): A fixed expected value to validate against. If this is provided, a --reference node is not required.ignore (optional): A single response path or an array of JSON paths to exclude from the comparison.test (optional): Metadata for the test.[
{
"run": "EveryBlocks(5)",
"test": {
"description": "WETH transfers to Uniswap V3"
},
"request": {
"jsonrpc": "2.0",
"method": "eth_getLogs",
"params": [{
"fromBlock": "{{Hex(Recent.Number - 9)}}",
"toBlock": "{{Hex(Recent.Number)}}",
"address": ["0xC02aaA39b223FE8D0A0e5C4F27eAD9083C756Cc2", "0xdAC17F958D2ee523a2206206994597C13D831ec7"],
"topics": [
"{{Topic.Transfer}}",
null,
"0x0000000000000000000000003fc91a3afd70395cd496c647d5a6cc9d4b2b7fad"
]
}],
"id": "{{Request.Number}}"
}
}
]The Engine API Proxy intercepts Engine API calls (specifically engine_forkchoiceUpdated, engine_newPayload, and engine_getPayload) to optimize the processing sequence. It supports four validation modes:
ForkChoiceUpdated pipeline.NewPayload pipeline.ForkChoiceUpdated pipeline and stores PayloadID without validation; validation occurs at the next new_payload request.PayloadAttributes from existing FCU requests instead of generating them.Note: For Lighthouse mode, use the Lighthouse consensus client with the --always-prepare-payload flag enabled. This is currently considered the most accurate validation mode.
The TimeoutCertificateManager ensures network liveness by handling timeouts when consensus progress stalls.
Timeout Flow:
Count >= threshold, a TimeoutCertificate (TC) is created.SyncInfo are broadcast to the network.HighestTC.TC.Round + 1.