feat(rpc): add eth_flashblocksPeerStatus API for P2P peer visibility

[sieniven]

Summary

Expose the current flashblocks builder p2p peers status, via a new JSON-RPC method eth_flashblocksPeerStatus, enabling operators to monitor connected, disconnected, and never-connected static peers with timing information for debugging broadcast issues.

Note that the static peers configured inside flashblocks.p2p_known_peers are always shown first in the returned response of the api call.

🤖 Generated with Claude Code

[sieniven]

@claude review for me

[XLayer-Bot]

Clean implementation overall — the design is well-structured, the event-hook wiring through the libp2p swarm loop is correct, and the test suite covers the key state transitions. Three concerns below.

workflow run

[XLayer-Bot]

The summary counts are computed by string-comparing PeerInfo::connection_state (a String) against hard-coded literals:

let connected     = peers.iter().filter(|p| p.connection_state == "connected").count();
let disconnected  = peers.iter().filter(|p| p.connection_state == "disconnected").count();
let never_connected = peers.iter().filter(|p| p.connection_state == "never_connected").count();

The internal PeerConnectionState enum already models these variants exhaustively, so a typo in either the match arm string (line ~119–124) or these filter literals would silently produce wrong summary counts with no compile-time error. The tests embed the same string literals, so a consistent rename would slip past them too.

Suggestion: make PeerConnectionState public, add #[derive(Serialize)] with #[serde(rename_all = "snake_case")], and use it as the type of PeerInfo::connection_state. The summary can then match on the enum and the compiler catches any divergence.

[sieniven]

Resolved in c8060fa.

PeerConnectionState is now a public ConnectionState enum with #[derive(Serialize)] and #[serde(rename_all = "snake_case")]. PeerInfo::connection_state is typed as ConnectionState instead of String, and summary counts use enum comparison. Wire format is unchanged.

[XLayer-Bot]

fb_p2p_status.get().cloned() silently returns None if the OnceLock was never set. The OnceLock is filled inside try_build() (called from spawn_payload_service), which runs during component setup. In practice this happens before extend_rpc_modules fires, but this ordering is implicit and undocumented.

If the ordering ever drifts — or if a future code path skips the payload builder — the sequencer's eth_flashblocksPeerStatus would silently return null with no log message indicating a misconfiguration.

Suggest adding a warning log (or a debug-mode assertion) in sequencer mode when the tracker is absent at RPC init time, so the failure is immediately visible:

let peer_status = fb_p2p_status.get().cloned();
if xlayer_args.sequencer_mode && peer_status.is_none() {
    warn!(target: "reth::cli", "fb_p2p_status OnceLock not set — eth_flashblocksPeerStatus will return null");
}
let xlayer_rpc = DefaultRpcExt::new(flashblocks_state, peer_status);

[sieniven]

Resolved in c8060fa.

Added a tracing::warn! when sequencer_mode && peer_status.is_none() at RPC init time, so the misconfiguration is immediately visible in logs rather than silently returning null.

[sieniven]

@claude review again for me with the new fixes, as well as the enhanced unit test coverages and e2e test coverage.

[XLayer-Bot]

All three previously raised concerns have been fully addressed in c8060fa:

1. Type-safe connection state — is now a public enum with and . is typed as ; summary counts use enum equality. No more stringly-typed comparisons.

2. OnceLock ordering visibility — A is emitted when at RPC init time, making the misconfiguration immediately observable in logs rather than silently returning null.

3. Non-static peer accumulation — now removes non-static peers entirely from the tracker (static peers retain their state). This is covered by the new disconnect_non_static_peer_removes_entry unit test.

Unit test coverage is thorough: 12 tests in peer_status.rs covering static/dynamic peer lifecycle, reconnection, broadcast failure/success semantics, snapshot sorting, and no-op operations on unknown peers. The NLL borrow-checker pattern used in on_disconnected (reading entry.is_static then calling HashMap::remove) is valid — the borrow ends at its last-use point under non-lexical lifetimes.

E2E test coverage adds three tests with appropriate #[ignore] annotations on the two tests that require a conductor-enabled devnet with P2P keys, while the null-on-RPC-node test runs unconditionally.

workflow run

[JimmyShi22]

Design overview:

• State tracking: A PeerStatusTracker (Arc<RwLock<HashMap<PeerId, PeerEntry>>>) tracks per-peer connection state, stream health, connection count, and last broadcast time. Static peers (from config) are retained after disconnect; dynamic peers are evicted to avoid unbounded growth.
• Event hooks: Four libp2p swarm events (on_connected, on_disconnected, on_stream_opened, on_broadcast_result) drive state transitions directly in the swarm loop.
• Cross-crate wiring: An Arc<OnceLock<PeerStatusTracker>> created in main.rs bridges the builder and RPC crates — written once after the P2P node is up, read at RPC init time.
• RPC handler: DefaultRpcExt holds an Option<PeerStatusTracker> and calls snapshot() on each request — sequencer nodes return full status, non-sequencer nodes return null.

Overall the design is clean and the approach makes sense.

[JimmyShi22]

LGTM. One thing worth keeping in mind: the multiaddr field in the response includes internal IPs and ports, which could be sensitive if this API is ever exposed publicly. For internal-only deployments that's fine, but if this endpoint gets added to any public-facing RPC allowlist, it probably shouldn't be included.

[XLayer-Bot]

🔧 CI running — workflow run

[XLayer-Bot]

✅ XLayer Reth CI passed for commit c1e8fc0bd39b0e3746f08849aee45dddec71b1f6

Step	Result
format-check	✅ success
compile-check	✅ success
clippy	✅ success
tests	✅ success

View run