validator-discovery: basic retrying logic (#3059)

* validator_discovery: less flexible, but simpler design

* fix test

* remove unused struct

* smol optimization

* validator_discovery: basic retrying logic

* add a test

* add more tests

* update the guide

* more test logic

* Require at least 2/3 connectivity.

* Fix test.

* Update node/network/gossip-support/src/lib.rs

Co-authored-by: André Silva <123550+andresilva@users.noreply.github.com>

* Update node/network/gossip-support/src/lib.rs

Co-authored-by: André Silva <123550+andresilva@users.noreply.github.com>

Co-authored-by: Robert Klotzner <robert.klotzner@gmx.at>
Co-authored-by: Robert Klotzner <eskimor@users.noreply.github.com>
Co-authored-by: André Silva <123550+andresilva@users.noreply.github.com>

polkadot/node/network/gossip-support/src/lib.rs

@@ -18,7 +18,10 @@
 //! and issuing a connection request to the validators relevant to
 //! the gossiping subsystems on every new session.
 
-use futures::FutureExt as _;
+#[cfg(test)]
+mod tests;
+
+use futures::{channel::oneshot, FutureExt as _};
 use polkadot_node_subsystem::{
 	messages::{
 		AllMessages, GossipSupportMessage, NetworkBridgeMessage,
@@ -44,6 +47,7 @@ pub struct GossipSupport {
 #[derive(Default)]
 struct State {
 	last_session_index: Option<SessionIndex>,
+	force_request: bool,
 }
 
 impl GossipSupport {
@@ -54,12 +58,18 @@ impl GossipSupport {
 		}
 	}
 
-	#[tracing::instrument(skip(self, ctx), fields(subsystem = LOG_TARGET))]
-	async fn run<Context>(self, mut ctx: Context)
+	async fn run<Context>(self, ctx: Context)
 	where
 		Context: SubsystemContext<Message = GossipSupportMessage>,
 	{
 		let mut state = State::default();
+		self.run_inner(ctx, &mut state).await;
+	}
+
+	async fn run_inner<Context>(self, mut ctx: Context, state: &mut State)
+	where
+		Context: SubsystemContext<Message = GossipSupportMessage>,
+	{
 		let Self { keystore } = self;
 		loop {
 			let message = match ctx.recv().await {
@@ -128,13 +138,16 @@ pub async fn connect_to_authorities(
 	ctx: &mut impl SubsystemContext,
 	validator_ids: Vec<AuthorityDiscoveryId>,
 	peer_set: PeerSet,
-) {
+) -> oneshot::Receiver<usize> {
+	let (failed, failed_rx) = oneshot::channel();
 	ctx.send_message(AllMessages::NetworkBridge(
 		NetworkBridgeMessage::ConnectToValidators {
 			validator_ids,
 			peer_set,
+			failed,
 		}
 	)).await;
+	failed_rx
 }
 
 impl State {
@@ -150,7 +163,7 @@ impl State {
 		for leaf in leaves {
 			let current_index = util::request_session_index_for_child(leaf, ctx.sender()).await.await??;
 			let maybe_new_session = match self.last_session_index {
-				Some(i) if i >= current_index => None,
+				Some(i) if current_index <= i && !self.force_request => None,
 				_ => Some((current_index, leaf)),
 			};
 
@@ -158,15 +171,22 @@ impl State {
 				tracing::debug!(target: LOG_TARGET, %new_session, "New session detected");
 				let authorities = determine_relevant_authorities(ctx, relay_parent).await?;
 				ensure_i_am_an_authority(keystore, &authorities).await?;
-				tracing::debug!(target: LOG_TARGET, num = ?authorities.len(), "Issuing a connection request");
+				let num = authorities.len();
+				tracing::debug!(target: LOG_TARGET, %num, "Issuing a connection request");
 
-				connect_to_authorities(
+				let failures = connect_to_authorities(
 					ctx,
 					authorities,
 					PeerSet::Validation,
 				).await;
 
+				// we await for the request to be processed
+				// this is fine, it should take much less time than one session
+				let failures = failures.await.unwrap_or(num);
+
 				self.last_session_index = Some(new_session);
+				// issue another request if at least a third of the authorities were not resolved
+				self.force_request = failures >= num / 3;
 			}
 		}