Enable reconstruction for nodes custodying more than 50% of columns and instrument tracing (#8052)

Co-Authored-By: Jimmy Chen <jchen.tc@gmail.com>

Co-Authored-By: Jimmy Chen <jimmy@sigmaprime.io>

beacon_node/beacon_chain/src/beacon_chain.rs

@@ -3299,10 +3299,14 @@ impl<T: BeaconChainTypes> BeaconChain<T> {
 
         let data_availability_checker = self.data_availability_checker.clone();
 
+        let current_span = Span::current();
         let result = self
             .task_executor
             .spawn_blocking_handle(
-                move || data_availability_checker.reconstruct_data_columns(&block_root),
+                move || {
+                    let _guard = current_span.enter();
+                    data_availability_checker.reconstruct_data_columns(&block_root)
+                },
                 "reconstruct_data_columns",
             )
             .ok_or(BeaconChainError::RuntimeShutdown)?

beacon_node/beacon_chain/src/data_availability_checker.rs

@@ -547,6 +547,7 @@ impl<T: BeaconChainTypes> DataAvailabilityChecker<T> {
         }
     }
 
+    #[instrument(skip_all, level = "debug")]
     pub fn reconstruct_data_columns(
         &self,
         block_root: &Hash256,

beacon_node/beacon_chain/src/data_availability_checker/overflow_lru_cache.rs

@@ -592,9 +592,9 @@ impl<T: BeaconChainTypes> DataAvailabilityCheckerInner<T> {
 
     /// Check whether data column reconstruction should be attempted.
     ///
-    /// Potentially trigger reconstruction if:
-    ///  - Our custody requirement is all columns (supernode), and we haven't got all columns
-    ///  - We have >= 50% of columns, but not all columns
+    /// Potentially trigger reconstruction if all the following satisfy:
+    ///  - Our custody requirement is more than 50% of total columns,
+    ///  - We haven't received all required columns
     ///  - Reconstruction hasn't been started for the block
     ///
     /// If reconstruction is required, returns `PendingComponents` which contains the
@@ -609,15 +609,25 @@ impl<T: BeaconChainTypes> DataAvailabilityCheckerInner<T> {
             return ReconstructColumnsDecision::No("block already imported");
         };
 
-        // If we're sampling all columns, it means we must be custodying all columns.
+        let Some(epoch) = pending_components
+            .verified_data_columns
+            .first()
+            .map(|c| c.as_data_column().epoch())
+        else {
+            return ReconstructColumnsDecision::No("not enough columns");
+        };
+
         let total_column_count = T::EthSpec::number_of_columns();
+        let sampling_column_count = self
+            .custody_context
+            .num_of_data_columns_to_sample(epoch, &self.spec);
         let received_column_count = pending_components.verified_data_columns.len();
 
         if pending_components.reconstruction_started {
             return ReconstructColumnsDecision::No("already started");
         }
-        if received_column_count >= total_column_count {
-            return ReconstructColumnsDecision::No("all columns received");
+        if received_column_count >= sampling_column_count {
+            return ReconstructColumnsDecision::No("all sampling columns received");
         }
         if received_column_count < total_column_count / 2 {
             return ReconstructColumnsDecision::No("not enough columns");

beacon_node/beacon_chain/src/validator_custody.rs

@@ -130,7 +130,7 @@ pub struct CustodyContext<E: EthSpec> {
     /// and enr values.
     validator_custody_count: AtomicU64,
     /// Is the node run as a supernode based on current cli parameters.
-    pub current_is_supernode: bool,
+    current_is_supernode: bool,
     /// The persisted value for `is_supernode` based on the previous run of this node.
     ///
     /// Note: We require this value because if a user restarts the node with a higher cli custody
@@ -307,6 +307,14 @@ impl<E: EthSpec> CustodyContext<E> {
             .expect("should compute node sampling size from valid chain spec")
     }
 
+    /// Returns whether the node should attempt reconstruction at a given epoch.
+    pub fn should_attempt_reconstruction(&self, epoch: Epoch, spec: &ChainSpec) -> bool {
+        let min_columns_for_reconstruction = E::number_of_columns() / 2;
+        // performing reconstruction is not necessary if sampling column count is exactly 50%,
+        // because the node doesn't need the remaining columns.
+        self.num_of_data_columns_to_sample(epoch, spec) > min_columns_for_reconstruction
+    }
+
     /// Returns the ordered list of column indices that should be sampled for data availability checking at the given epoch.
     ///
     /// # Parameters