Add test to demonstrate the failure scenario (#1999)

The change adds a test to show the failure scenario that caused #1812 to
be rolled back (more context:
https://github.com/paritytech/polkadot-sdk/issues/493#issuecomment-1772009924)

Summary of the scenario:
1. Node has finished downloading up to block 1000 from the peers, from
the canonical chain.
2. Peers are undergoing re-org around this time. One of the peers has
switched to a non-canonical chain, announces block 1001 from that chain
3. Node downloads 1001 from the peer, and tries to import which would
fail (as we don't have the parent block 1000 from the other chain)

---------

Co-authored-by: Dmitry Markin <dmitry@markin.tech>

Cargo.lock

@@ -18466,7 +18466,7 @@ version = "0.3.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "21bebf2b7c9e0a515f6e0f8c51dc0f8e4696391e6f1ff30379559f8365fb0df7"
 dependencies = [
- "rustix 0.38.8",
+ "rustix 0.38.21",
  "windows-sys 0.48.0",
 ]
 

substrate/client/network/sync/src/lib.rs

@@ -3364,4 +3364,129 @@ mod test {
 		pending_responses.remove(&peers[1]);
 		assert_eq!(pending_responses.len(), 0);
 	}
+
+	/// The test demonstrates https://github.com/paritytech/polkadot-sdk/issues/2094.
+	/// TODO: convert it into desired behavior test once the issue is fixed (see inline comments).
+	/// The issue: we currently rely on block numbers instead of block hash
+	/// to download blocks from peers. As a result, we can end up with blocks
+	/// from different forks as shown by the test.
+	#[test]
+	#[should_panic]
+	fn request_across_forks() {
+		sp_tracing::try_init_simple();
+
+		let (_chain_sync_network_provider, chain_sync_network_handle) =
+			NetworkServiceProvider::new();
+		let mut client = Arc::new(TestClientBuilder::new().build());
+		let blocks = (0..100).map(|_| build_block(&mut client, None, false)).collect::<Vec<_>>();
+
+		let fork_a_blocks = {
+			let mut client = Arc::new(TestClientBuilder::new().build());
+			let mut fork_blocks = blocks[..]
+				.into_iter()
+				.inspect(|b| {
+					assert!(matches!(client.block(*b.header.parent_hash()), Ok(Some(_))));
+					block_on(client.import(BlockOrigin::Own, (*b).clone())).unwrap()
+				})
+				.cloned()
+				.collect::<Vec<_>>();
+			for _ in 0..10 {
+				fork_blocks.push(build_block(&mut client, None, false));
+			}
+			fork_blocks
+		};
+
+		let fork_b_blocks = {
+			let mut client = Arc::new(TestClientBuilder::new().build());
+			let mut fork_blocks = blocks[..]
+				.into_iter()
+				.inspect(|b| {
+					assert!(matches!(client.block(*b.header.parent_hash()), Ok(Some(_))));
+					block_on(client.import(BlockOrigin::Own, (*b).clone())).unwrap()
+				})
+				.cloned()
+				.collect::<Vec<_>>();
+			for _ in 0..10 {
+				fork_blocks.push(build_block(&mut client, None, true));
+			}
+			fork_blocks
+		};
+
+		let mut sync = ChainSync::new(
+			SyncMode::Full,
+			client.clone(),
+			ProtocolName::from("test-block-announce-protocol"),
+			5,
+			64,
+			None,
+			chain_sync_network_handle,
+		)
+		.unwrap();
+
+		// Add the peers, all at the common ancestor 100.
+		let common_block = blocks.last().unwrap();
+		let peer_id1 = PeerId::random();
+		sync.new_peer(peer_id1, common_block.hash(), *common_block.header().number())
+			.unwrap();
+		let peer_id2 = PeerId::random();
+		sync.new_peer(peer_id2, common_block.hash(), *common_block.header().number())
+			.unwrap();
+
+		// Peer 1 announces 107 from fork 1, 100-107 get downloaded.
+		{
+			let block = (&fork_a_blocks[106]).clone();
+			let peer = peer_id1;
+			log::trace!(target: LOG_TARGET, "<1> {peer} announces from fork 1");
+			send_block_announce(block.header().clone(), peer, &mut sync);
+			let request = get_block_request(&mut sync, FromBlock::Hash(block.hash()), 7, &peer);
+			let mut resp_blocks = fork_a_blocks[100_usize..107_usize].to_vec();
+			resp_blocks.reverse();
+			let response = create_block_response(resp_blocks.clone());
+			let res = sync.on_block_data(&peer, Some(request), response).unwrap();
+			assert!(matches!(
+				res,
+				OnBlockData::Import(ImportBlocksAction{ origin: _, blocks }) if blocks.len() == 7_usize
+			),);
+			assert_eq!(sync.best_queued_number, 107);
+			assert_eq!(sync.best_queued_hash, block.hash());
+			assert!(sync.is_known(&block.header.parent_hash()));
+		}
+
+		// Peer 2 also announces 107 from fork 1.
+		{
+			let prev_best_number = sync.best_queued_number;
+			let prev_best_hash = sync.best_queued_hash;
+			let peer = peer_id2;
+			log::trace!(target: LOG_TARGET, "<2> {peer} announces from fork 1");
+			for i in 100..107 {
+				let block = (&fork_a_blocks[i]).clone();
+				send_block_announce(block.header().clone(), peer, &mut sync);
+				assert!(sync.block_requests().is_empty());
+			}
+			assert_eq!(sync.best_queued_number, prev_best_number);
+			assert_eq!(sync.best_queued_hash, prev_best_hash);
+		}
+
+		// Peer 2 undergoes reorg, announces 108 from fork 2, gets downloaded even though we
+		// don't have the parent from fork 2.
+		{
+			let block = (&fork_b_blocks[107]).clone();
+			let peer = peer_id2;
+			log::trace!(target: LOG_TARGET, "<3> {peer} announces from fork 2");
+			send_block_announce(block.header().clone(), peer, &mut sync);
+			// TODO: when the issue is fixed, this test can be changed to test the
+			// expected behavior instead. The needed changes would be:
+			// 1. Remove the `#[should_panic]` directive
+			// 2. These should be changed to check that sync.block_requests().is_empty(), after the
+			//    block is announced.
+			let request = get_block_request(&mut sync, FromBlock::Hash(block.hash()), 1, &peer);
+			let response = create_block_response(vec![block.clone()]);
+			let res = sync.on_block_data(&peer, Some(request), response).unwrap();
+			assert!(matches!(
+				res,
+				OnBlockData::Import(ImportBlocksAction{ origin: _, blocks }) if blocks.len() == 1_usize
+			),);
+			assert!(sync.is_known(&block.header.parent_hash()));
+		}
+	}
 }