Parachains-Aura: Only produce once per slot

cumulus/client/consensus/aura/src/collator.rs

@@ -258,6 +258,7 @@ where
 pub struct SlotClaim<Pub> {
 	author_pub: Pub,
 	pre_digest: DigestItem,
+	slot: Slot,
 	timestamp: Timestamp,
 }
 
@@ -272,7 +273,7 @@ impl<Pub> SlotClaim<Pub> {
 		P::Public: Codec,
 		P::Signature: Codec,
 	{
-		SlotClaim { author_pub, timestamp, pre_digest: aura_internal::pre_digest::<P>(slot) }
+		SlotClaim { author_pub, timestamp, pre_digest: aura_internal::pre_digest::<P>(slot), slot }
 	}
 
 	/// Get the author's public key.
@@ -285,6 +286,11 @@ impl<Pub> SlotClaim<Pub> {
 		&self.pre_digest
 	}
 
+	/// Get the slot assigned to this claim.
+	pub fn slot(&self) -> Slot {
+		self.slot
+	}
+
 	/// Get the timestamp corresponding to the relay-chain slot this claim was
 	/// generated against.
 	pub fn timestamp(&self) -> Timestamp {

cumulus/client/consensus/aura/src/collators/basic.rs

@@ -141,6 +141,8 @@ where
 			collator_util::Collator::<Block, P, _, _, _, _, _>::new(params)
 		};
 
+		let mut last_processed_slot = 0;
+
 		while let Some(request) = collation_requests.next().await {
 			macro_rules! reject_with_error {
 				($err:expr) => {{
@@ -192,6 +194,18 @@ where
 				Err(e) => reject_with_error!(e),
 			};
 
+			// With async backing this function will be called every relay chain block.
+			//
+			// Most parachains currently run with 12 seconds slots and thus, they would try to
+			// produce multiple blocks per slot which very likely would fail on chain. Thus, we have
+			// this "hack" to only produce on block per slot.
+			//
+			// With https://github.com/paritytech/polkadot-sdk/issues/3168 this implementation will be
+			// obsolete and also the underlying issue will be fixed.
+			if last_processed_slot >= *claim.slot() {
+				continue
+			}
+
 			let (parachain_inherent_data, other_inherent_data) = try_request!(
 				collator
 					.create_inherent_data(
@@ -228,6 +242,8 @@ where
 				request.complete(None);
 				tracing::debug!(target: crate::LOG_TARGET, "No block proposal");
 			}
+
+			last_processed_slot = *claim.slot();
 		}
 	}
 }

cumulus/client/consensus/aura/src/lib.rs

@@ -42,7 +42,14 @@ use sp_core::crypto::Pair;
 use sp_inherents::CreateInherentDataProviders;
 use sp_keystore::KeystorePtr;
 use sp_runtime::traits::{Block as BlockT, Header as HeaderT, Member, NumberFor};
-use std::{convert::TryFrom, marker::PhantomData, sync::Arc};
+use std::{
+	convert::TryFrom,
+	marker::PhantomData,
+	sync::{
+		atomic::{AtomicU64, Ordering},
+		Arc,
+	},
+};
 
 mod import_queue;
 
@@ -61,6 +68,7 @@ pub struct AuraConsensus<B, CIDP, W> {
 	create_inherent_data_providers: Arc<CIDP>,
 	aura_worker: Arc<Mutex<W>>,
 	slot_duration: SlotDuration,
+	last_slot_processed: Arc<AtomicU64>,
 	_phantom: PhantomData<B>,
 }
 
@@ -70,6 +78,7 @@ impl<B, CIDP, W> Clone for AuraConsensus<B, CIDP, W> {
 			create_inherent_data_providers: self.create_inherent_data_providers.clone(),
 			aura_worker: self.aura_worker.clone(),
 			slot_duration: self.slot_duration,
+			last_slot_processed: self.last_slot_processed.clone(),
 			_phantom: PhantomData,
 		}
 	}
@@ -156,6 +165,7 @@ where
 		Box::new(AuraConsensus {
 			create_inherent_data_providers: Arc::new(create_inherent_data_providers),
 			aura_worker: Arc::new(Mutex::new(worker)),
+			last_slot_processed: Default::default(),
 			slot_duration,
 			_phantom: PhantomData,
 		})
@@ -221,6 +231,18 @@ where
 			Some((validation_data.max_pov_size / 2) as usize),
 		);
 
+		// With async backing this function will be called every relay chain block.
+		//
+		// Most parachains currently run with 12 seconds slots and thus, they would try to produce
+		// multiple blocks per slot which very likely would fail on chain. Thus, we have this "hack"
+		// to only produce on block per slot.
+		//
+		// With https://github.com/paritytech/polkadot-sdk/issues/3168 this implementation will be
+		// obsolete and also the underlying issue will be fixed.
+		if self.last_slot_processed.fetch_max(*info.slot, Ordering::Relaxed) >= *info.slot {
+			return None
+		}
+
 		let res = self.aura_worker.lock().await.on_slot(info).await?;
 
 		Some(ParachainCandidate { block: res.block, proof: res.storage_proof })

prdoc/pr_3308.prdoc

@@ -0,0 +1,14 @@
+title: Parachains-Aura: Only produce once per slot
+
+doc:
+  - audience: Node Dev
+    description: |
+      With the introduction of asynchronous backing the relay chain allows parachain to include blocks every 6 seconds.
+      The Cumulus Aura implementations, besides the lookahead collator, are building blocks when there is a free slot for
+      the parachain in the relay chain. Most parachains are still running with a 12s slot duration and not allowing
+      to build multiple blocks per slot. But, the block production logic will be triggered every 6s, resulting in error
+      logs like: "no space left for the block in the unincluded segment". This is solved by ensuring that we don't build
+      multiple blocks per slot.
+
+crates:
+  - name: "cumulus-client-consensus-aura"