Implement tracing spans for data columm RPC requests and responses (#7831)

#7830

beacon_node/network/src/sync/network_context.rs

@@ -29,6 +29,7 @@ use lighthouse_network::service::api_types::{
     DataColumnsByRootRequester, Id, SingleLookupReqId, SyncRequestId,
 };
 use lighthouse_network::{Client, NetworkGlobals, PeerAction, PeerId, ReportSource};
+use lighthouse_tracing::SPAN_OUTGOING_RANGE_REQUEST;
 use parking_lot::RwLock;
 pub use requests::LookupVerifyError;
 use requests::{
@@ -45,7 +46,7 @@ use std::time::Duration;
 #[cfg(test)]
 use task_executor::TaskExecutor;
 use tokio::sync::mpsc;
-use tracing::{debug, error, warn};
+use tracing::{Span, debug, debug_span, error, warn};
 use types::blob_sidecar::FixedBlobSidecarList;
 use types::{
     BlobSidecar, ColumnIndex, DataColumnSidecar, DataColumnSidecarList, EthSpec, ForkContext,
@@ -55,6 +56,18 @@ use types::{
 pub mod custody;
 mod requests;
 
+macro_rules! new_range_request_span {
+    ($self:expr, $name:literal, $parent:expr, $peer_id:expr) => {{
+        let client = $self.client_type(&$peer_id).kind;
+        debug_span!(
+            parent: $parent,
+            $name,
+            peer_id = %$peer_id,
+            client = %client
+        )
+    }};
+}
+
 /// Max retries for block components after which we fail the batch.
 pub const MAX_COLUMN_RETRIES: usize = 3;
 
@@ -444,10 +457,16 @@ impl<T: BeaconChainTypes> SyncNetworkContext<T> {
         request: BlocksByRangeRequest,
         failed_columns: &HashSet<ColumnIndex>,
     ) -> Result<(), String> {
-        let Some(requester) = self
+        let Some((requester, parent_request_span)) = self
             .components_by_range_requests
-            .keys()
-            .find_map(|r| if r.id == id { Some(r.requester) } else { None })
+            .iter()
+            .find_map(|(key, value)| {
+                if key.id == id {
+                    Some((key.requester, value.request_span.clone()))
+                } else {
+                    None
+                }
+            })
         else {
             return Err("request id not present".to_string());
         };
@@ -485,6 +504,12 @@ impl<T: BeaconChainTypes> SyncNetworkContext<T> {
                         columns,
                     },
                     id,
+                    new_range_request_span!(
+                        self,
+                        "outgoing_columns_by_range_retry",
+                        parent_request_span.clone(),
+                        peer_id
+                    ),
                 )
             })
             .collect::<Result<Vec<_>, _>>()
@@ -511,6 +536,13 @@ impl<T: BeaconChainTypes> SyncNetworkContext<T> {
         peers: &HashSet<PeerId>,
         peers_to_deprioritize: &HashSet<PeerId>,
     ) -> Result<Id, RpcRequestSendError> {
+        let range_request_span = debug_span!(
+            parent: None,
+            SPAN_OUTGOING_RANGE_REQUEST,
+            range_req_id = %requester,
+            peers = peers.len()
+        );
+        let _guard = range_request_span.clone().entered();
         let active_request_count_by_peer = self.active_request_count_by_peer();
 
         let Some(block_peer) = peers
@@ -561,7 +593,17 @@ impl<T: BeaconChainTypes> SyncNetworkContext<T> {
             requester,
         };
 
-        let blocks_req_id = self.send_blocks_by_range_request(block_peer, request.clone(), id)?;
+        let blocks_req_id = self.send_blocks_by_range_request(
+            block_peer,
+            request.clone(),
+            id,
+            new_range_request_span!(
+                self,
+                "outgoing_blocks_by_range",
+                range_request_span.clone(),
+                block_peer
+            ),
+        )?;
 
         let blobs_req_id = if matches!(batch_type, ByRangeRequestType::BlocksAndBlobs) {
             Some(self.send_blobs_by_range_request(
@@ -571,6 +613,12 @@ impl<T: BeaconChainTypes> SyncNetworkContext<T> {
                     count: *request.count(),
                 },
                 id,
+                new_range_request_span!(
+                    self,
+                    "outgoing_blobs_by_range",
+                    range_request_span.clone(),
+                    block_peer
+                ),
             )?)
         } else {
             None
@@ -589,6 +637,12 @@ impl<T: BeaconChainTypes> SyncNetworkContext<T> {
                                 columns,
                             },
                             id,
+                            new_range_request_span!(
+                                self,
+                                "outgoing_columns_by_range",
+                                range_request_span.clone(),
+                                peer_id
+                            ),
                         )
                     })
                     .collect::<Result<Vec<_>, _>>()
@@ -605,6 +659,7 @@ impl<T: BeaconChainTypes> SyncNetworkContext<T> {
                     self.chain.sampling_columns_for_epoch(epoch).to_vec(),
                 )
             }),
+            range_request_span,
         );
         self.components_by_range_requests.insert(id, info);
 
@@ -833,6 +888,8 @@ impl<T: BeaconChainTypes> SyncNetworkContext<T> {
             // block and the peer must have it.
             true,
             BlocksByRootRequestItems::new(request),
+            // Not implemented
+            Span::none(),
         );
 
         Ok(LookupRequestResult::RequestSent(id.req_id))
@@ -927,6 +984,8 @@ impl<T: BeaconChainTypes> SyncNetworkContext<T> {
             // have imported the block+blobs.
             true,
             BlobsByRootRequestItems::new(request),
+            // Not implemented
+            Span::none(),
         );
 
         Ok(LookupRequestResult::RequestSent(id.req_id))
@@ -970,6 +1029,9 @@ impl<T: BeaconChainTypes> SyncNetworkContext<T> {
             peer_id,
             expect_max_responses,
             DataColumnsByRootRequestItems::new(request),
+            // Span is tracked in `self.custody_columns_by_root_requests` in the
+            // `ActiveCustodyRequest` struct.
+            Span::none(),
         );
 
         Ok(LookupRequestResult::RequestSent(id))
@@ -1065,6 +1127,7 @@ impl<T: BeaconChainTypes> SyncNetworkContext<T> {
         peer_id: PeerId,
         request: BlocksByRangeRequest,
         parent_request_id: ComponentsByRangeRequestId,
+        request_span: Span,
     ) -> Result<BlocksByRangeRequestId, RpcRequestSendError> {
         let id = BlocksByRangeRequestId {
             id: self.next_id(),
@@ -1094,6 +1157,7 @@ impl<T: BeaconChainTypes> SyncNetworkContext<T> {
             // know if there are missed blocks.
             false,
             BlocksByRangeRequestItems::new(request),
+            request_span,
         );
         Ok(id)
     }
@@ -1103,6 +1167,7 @@ impl<T: BeaconChainTypes> SyncNetworkContext<T> {
         peer_id: PeerId,
         request: BlobsByRangeRequest,
         parent_request_id: ComponentsByRangeRequestId,
+        request_span: Span,
     ) -> Result<BlobsByRangeRequestId, RpcRequestSendError> {
         let id = BlobsByRangeRequestId {
             id: self.next_id(),
@@ -1136,6 +1201,7 @@ impl<T: BeaconChainTypes> SyncNetworkContext<T> {
             // know if there are missed blocks.
             false,
             BlobsByRangeRequestItems::new(request, max_blobs_per_block),
+            request_span,
         );
         Ok(id)
     }
@@ -1145,6 +1211,7 @@ impl<T: BeaconChainTypes> SyncNetworkContext<T> {
         peer_id: PeerId,
         request: DataColumnsByRangeRequest,
         parent_request_id: ComponentsByRangeRequestId,
+        request_span: Span,
     ) -> Result<(DataColumnsByRangeRequestId, Vec<u64>), RpcRequestSendError> {
         let requested_columns = request.columns.clone();
         let id = DataColumnsByRangeRequestId {
@@ -1177,6 +1244,7 @@ impl<T: BeaconChainTypes> SyncNetworkContext<T> {
             // know if there are missed blocks.
             false,
             DataColumnsByRangeRequestItems::new(request),
+            request_span,
         );
         Ok((id, requested_columns))
     }