Improving blob propagation post-PeerDAS with Decentralized Blob Building

beacon_node/beacon_chain/benches/benches.rs

@@ -37,12 +37,15 @@ fn all_benches(c: &mut Criterion) {
 
     let kzg = get_kzg(&spec);
     for blob_count in [1, 2, 3, 6] {
-        let kzg = kzg.clone();
-        let (signed_block, blob_sidecars) = create_test_block_and_blobs::<E>(blob_count, &spec);
+        let (signed_block, blobs) = create_test_block_and_blobs::<E>(blob_count, &spec);
 
-        let column_sidecars =
-            blobs_to_data_column_sidecars(&blob_sidecars, &signed_block, &kzg.clone(), &spec)
-                .unwrap();
+        let column_sidecars = blobs_to_data_column_sidecars(
+            &blobs.iter().collect::<Vec<_>>(),
+            &signed_block,
+            &kzg,
+            &spec,
+        )
+        .unwrap();
 
         let spec = spec.clone();
 

beacon_node/beacon_chain/src/blob_verification.rs

@@ -12,7 +12,6 @@ use crate::{metrics, BeaconChainError};
 use kzg::{Error as KzgError, Kzg, KzgCommitment};
 use slog::debug;
 use ssz_derive::{Decode, Encode};
-use ssz_types::VariableList;
 use std::time::Duration;
 use tree_hash::TreeHash;
 use types::blob_sidecar::BlobIdentifier;
@@ -156,11 +155,6 @@ impl From<BeaconStateError> for GossipBlobError {
     }
 }
 
-pub type GossipVerifiedBlobList<T> = VariableList<
-    GossipVerifiedBlob<T>,
-    <<T as BeaconChainTypes>::EthSpec as EthSpec>::MaxBlobsPerBlock,
->;
-
 /// A wrapper around a `BlobSidecar` that indicates it has been approved for re-gossiping on
 /// the p2p network.
 #[derive(Debug)]
@@ -335,6 +329,25 @@ impl<E: EthSpec> KzgVerifiedBlobList<E> {
             verified_blobs: blobs,
         })
     }
+
+    /// Create a `KzgVerifiedBlobList` from `blobs` that are already KZG verified.
+    ///
+    /// This should be used with caution, as used incorrectly it could result in KZG verification
+    /// being skipped and invalid blobs being deemed valid.
+    pub fn from_verified<I: IntoIterator<Item = Arc<BlobSidecar<E>>>>(
+        blobs: I,
+        seen_timestamp: Duration,
+    ) -> Self {
+        Self {
+            verified_blobs: blobs
+                .into_iter()
+                .map(|blob| KzgVerifiedBlob {
+                    blob,
+                    seen_timestamp,
+                })
+                .collect(),
+        }
+    }
 }
 
 impl<E: EthSpec> IntoIterator for KzgVerifiedBlobList<E> {

beacon_node/beacon_chain/src/block_verification.rs

@@ -683,7 +683,7 @@ pub struct SignatureVerifiedBlock<T: BeaconChainTypes> {
     consensus_context: ConsensusContext<T::EthSpec>,
 }
 
-/// Used to await the result of executing payload with a remote EE.
+/// Used to await the result of executing payload with an EE.
 type PayloadVerificationHandle = JoinHandle<Option<Result<PayloadVerificationOutcome, BlockError>>>;
 
 /// A wrapper around a `SignedBeaconBlock` that indicates that this block is fully verified and
@@ -750,7 +750,8 @@ pub fn build_blob_data_column_sidecars<T: BeaconChainTypes>(
         &metrics::DATA_COLUMN_SIDECAR_COMPUTATION,
         &[&blobs.len().to_string()],
     );
-    let sidecars = blobs_to_data_column_sidecars(&blobs, block, &chain.kzg, &chain.spec)
+    let blob_refs = blobs.iter().collect::<Vec<_>>();
+    let sidecars = blobs_to_data_column_sidecars(&blob_refs, block, &chain.kzg, &chain.spec)
         .discard_timer_on_break(&mut timer)?;
     drop(timer);
     Ok(sidecars)
@@ -1343,7 +1344,6 @@ impl<T: BeaconChainTypes> ExecutionPendingBlock<T> {
         /*
          *  Perform cursory checks to see if the block is even worth processing.
          */
-
         check_block_relevancy(block.as_block(), block_root, chain)?;
 
         // Define a future that will verify the execution payload with an execution engine.

beacon_node/beacon_chain/src/chain_config.rs

@@ -88,6 +88,10 @@ pub struct ChainConfig {
     pub malicious_withhold_count: usize,
     /// Enable peer sampling on blocks.
     pub enable_sampling: bool,
+    /// Number of batches that supernodes split data columns into during publishing by a non-proposer.
+    pub supernode_data_column_publication_batches: usize,
+    /// The delay applied by supernodes between the sending of each data column batch.
+    pub supernode_data_column_publication_batch_interval: Duration,
 }
 
 impl Default for ChainConfig {
@@ -121,6 +125,8 @@ impl Default for ChainConfig {
             enable_light_client_server: false,
             malicious_withhold_count: 0,
             enable_sampling: false,
+            supernode_data_column_publication_batches: 4,
+            supernode_data_column_publication_batch_interval: Duration::from_millis(200),
         }
     }
 }

beacon_node/beacon_chain/src/data_availability_checker.rs

@@ -18,7 +18,7 @@ use task_executor::TaskExecutor;
 use types::blob_sidecar::{BlobIdentifier, BlobSidecar, FixedBlobSidecarList};
 use types::{
     BlobSidecarList, ChainSpec, DataColumnIdentifier, DataColumnSidecar, DataColumnSidecarList,
-    Epoch, EthSpec, Hash256, RuntimeVariableList, SignedBeaconBlock, Slot,
+    Epoch, EthSpec, Hash256, RuntimeVariableList, SignedBeaconBlock,
 };
 
 mod error;
@@ -201,7 +201,6 @@ impl<T: BeaconChainTypes> DataAvailabilityChecker<T> {
     pub fn put_rpc_blobs(
         &self,
         block_root: Hash256,
-        epoch: Epoch,
         blobs: FixedBlobSidecarList<T::EthSpec>,
     ) -> Result<Availability<T::EthSpec>, AvailabilityCheckError> {
         let seen_timestamp = self
@@ -212,15 +211,12 @@ impl<T: BeaconChainTypes> DataAvailabilityChecker<T> {
         // Note: currently not reporting which specific blob is invalid because we fetch all blobs
         // from the same peer for both lookup and range sync.
 
-        let verified_blobs = KzgVerifiedBlobList::new(
-            Vec::from(blobs).into_iter().flatten(),
-            &self.kzg,
-            seen_timestamp,
-        )
-        .map_err(AvailabilityCheckError::InvalidBlobs)?;
+        let verified_blobs =
+            KzgVerifiedBlobList::new(blobs.iter().flatten().cloned(), &self.kzg, seen_timestamp)
+                .map_err(AvailabilityCheckError::InvalidBlobs)?;
 
         self.availability_cache
-            .put_kzg_verified_blobs(block_root, epoch, verified_blobs, &self.log)
+            .put_kzg_verified_blobs(block_root, verified_blobs, &self.log)
     }
 
     /// Put a list of custody columns received via RPC into the availability cache. This performs KZG
@@ -229,7 +225,6 @@ impl<T: BeaconChainTypes> DataAvailabilityChecker<T> {
     pub fn put_rpc_custody_columns(
         &self,
         block_root: Hash256,
-        epoch: Epoch,
         custody_columns: DataColumnSidecarList<T::EthSpec>,
     ) -> Result<Availability<T::EthSpec>, AvailabilityCheckError> {
         // TODO(das): report which column is invalid for proper peer scoring
@@ -248,12 +243,32 @@ impl<T: BeaconChainTypes> DataAvailabilityChecker<T> {
 
         self.availability_cache.put_kzg_verified_data_columns(
             block_root,
-            epoch,
             verified_custody_columns,
             &self.log,
         )
     }
 
+    /// Put a list of blobs received from the EL pool into the availability cache.
+    ///
+    /// This DOES NOT perform KZG verification because the KZG proofs should have been constructed
+    /// immediately prior to calling this function so they are assumed to be valid.
+    pub fn put_engine_blobs(
+        &self,
+        block_root: Hash256,
+        blobs: FixedBlobSidecarList<T::EthSpec>,
+    ) -> Result<Availability<T::EthSpec>, AvailabilityCheckError> {
+        let seen_timestamp = self
+            .slot_clock
+            .now_duration()
+            .ok_or(AvailabilityCheckError::SlotClockError)?;
+
+        let verified_blobs =
+            KzgVerifiedBlobList::from_verified(blobs.iter().flatten().cloned(), seen_timestamp);
+
+        self.availability_cache
+            .put_kzg_verified_blobs(block_root, verified_blobs, &self.log)
+    }
+
     /// Check if we've cached other blobs for this block. If it completes a set and we also
     /// have a block cached, return the `Availability` variant triggering block import.
     /// Otherwise cache the blob sidecar.
@@ -265,7 +280,6 @@ impl<T: BeaconChainTypes> DataAvailabilityChecker<T> {
     ) -> Result<Availability<T::EthSpec>, AvailabilityCheckError> {
         self.availability_cache.put_kzg_verified_blobs(
             gossip_blob.block_root(),
-            gossip_blob.epoch(),
             vec![gossip_blob.into_inner()],
             &self.log,
         )
@@ -279,20 +293,16 @@ impl<T: BeaconChainTypes> DataAvailabilityChecker<T> {
     #[allow(clippy::type_complexity)]
     pub fn put_gossip_data_columns(
         &self,
-        slot: Slot,
         block_root: Hash256,
         gossip_data_columns: Vec<GossipVerifiedDataColumn<T>>,
     ) -> Result<Availability<T::EthSpec>, AvailabilityCheckError> {
-        let epoch = slot.epoch(T::EthSpec::slots_per_epoch());
-
         let custody_columns = gossip_data_columns
             .into_iter()
             .map(|c| KzgVerifiedCustodyDataColumn::from_asserted_custody(c.into_inner()))
             .collect::<Vec<_>>();
 
         self.availability_cache.put_kzg_verified_data_columns(
             block_root,
-            epoch,
             custody_columns,
             &self.log,
         )
@@ -595,12 +605,7 @@ impl<T: BeaconChainTypes> DataAvailabilityChecker<T> {
         );
 
         self.availability_cache
-            .put_kzg_verified_data_columns(
-                *block_root,
-                slot.epoch(T::EthSpec::slots_per_epoch()),
-                data_columns_to_publish.clone(),
-                &self.log,
-            )
+            .put_kzg_verified_data_columns(*block_root, data_columns_to_publish.clone(), &self.log)
             .map(|availability| {
                 DataColumnReconstructionResult::Success((
                     availability,

beacon_node/beacon_chain/src/data_availability_checker/error.rs

@@ -1,4 +1,5 @@
 use kzg::{Error as KzgError, KzgCommitment};
+use types::data_column_sidecar::DataColumnSidecarError;
 use types::{BeaconStateError, ColumnIndex, Hash256};
 
 #[derive(Debug)]
@@ -23,6 +24,7 @@ pub enum Error {
     BlockReplayError(state_processing::BlockReplayError),
     RebuildingStateCaches(BeaconStateError),
     SlotClockError,
+    DataColumnSidecarError(DataColumnSidecarError),
 }
 
 #[derive(PartialEq, Eq)]
@@ -46,7 +48,8 @@ impl Error {
             | Error::BlockReplayError(_)
             | Error::UnableToDetermineImportRequirement
             | Error::RebuildingStateCaches(_)
-            | Error::SlotClockError => ErrorCategory::Internal,
+            | Error::SlotClockError
+            | Error::DataColumnSidecarError(_) => ErrorCategory::Internal,
             Error::InvalidBlobs { .. }
             | Error::InvalidColumn { .. }
             | Error::ReconstructColumnsError { .. }