UX: Find way to keep EV-node RPC endpoint available on error scenarios when Full node halts gracefully

[Manav-Aggarwal]

Currently running into an issue trying to debug where a full node is running into an error when syncing, in that case, the full node halts and stops gracefully so I can't query its RPC endpoint for blocks and headers. It'd be nice like to have a way to query the full node's RPC endpoint even after it halts or just maybe keep the node process up with just the core process stopped.

[tac0turtle]

@claude could you create a solution where the node could be started in rpc only mode where everything but the sequencer block production is started

[claude]

Claude finished @tac0turtle's task —— View job • claude/issue-2512-20250808-0746 • Create PR ➔

---

✅ Implementation Complete: RPC-Only Mode for EV-Node

I've successfully implemented RPC-only mode for ev-node to address the issue where RPC endpoints become unavailable when full nodes halt during sync errors.

Solution Overview

The implementation adds a new --rpc-only flag that creates a minimal node mode where:

• RPC server remains active for debugging queries (Store, P2P, Health services)
• No block production or syncing - prevents conflicts that cause node halting
• Basic P2P connectivity maintained for network access
• Mutual exclusion with other node modes (aggregator, light)

Key Changes

Configuration:

• Added RPCOnly flag to NodeConfig with CLI support (--rollkit.node.rpc_only)
• Updated default config and added validation for conflicting modes
• Enhanced config validation to prevent multiple node modes simultaneously

Node Implementation:

• Created RPCOnlyNode type that starts minimal services
• Modified NewNode() routing logic to support RPC-only mode
• Added comprehensive tests for validation and functionality

Usage:

# Start node in RPC-only mode
./testapp start --rollkit.node.rpc_only=true

# RPC endpoints remain available for debugging:
# - Store service: query blocks, headers, etc.
# - P2P service: network connectivity info  
# - Health service: node status

Files Modified

• pkg/config/config.go - Added RPCOnly flag and validation
• pkg/config/defaults.go - Updated default config
• node/node.go - Updated node creation logic
• node/rpc_only.go - New RPCOnlyNode implementation
• node/rpc_only_test.go - Tests for new functionality

Todo List:

• Analyze current node architecture and startup process
• Understand sequencer integration and how to disable it
• Find configuration system and add RPC-only mode flag
• Modify node initialization to conditionally skip sequencer
• Ensure RPC endpoints remain available in RPC-only mode
• Test the implementation with build commands
• Commit and push changes

This solution directly addresses the original issue by providing a way to keep RPC endpoints accessible for debugging even when main node processes encounter errors during synchronization.

---

[julienrbrt]

Is this really useful? Personally I'd rather have the node halt, so you know there's an issue directly, instead of having to continuously fetch for the latest height and see if it is advancing.

What use case you would want to query a node that doesn't sync?

[tac0turtle]

We can do a rpc only mode if we really wanted but tend to agree with halting over staying alive in the case of errors. We can just start the rpc server from a command and feed in the store.