Warn if vaildator appears to be repairing excessively (part 1)

solana-labs · Oct 10, 2023 · 185ab33 · 185ab33
1 parent 982d29c
commit 185ab33
Show file tree

Hide file tree

Showing 32 changed files with 241 additions and 76 deletions.
diff --git a/banking-bench/src/main.rs b/banking-bench/src/main.rs
@@ -436,7 +436,12 @@ fn main() {
         let cluster_info = {
             let keypair = Arc::new(Keypair::new());
             let node = Node::new_localhost_with_pubkey(&keypair.pubkey());
-            ClusterInfo::new(node.info, keypair, SocketAddrSpace::Unspecified)
+            ClusterInfo::new(
+                node.info,
+                keypair,
+                SocketAddrSpace::Unspecified,
+                /*known_validators*/ None,
+            )
         };
         let cluster_info = Arc::new(cluster_info);
         let tpu_disable_quic = matches.is_present("tpu_disable_quic");

diff --git a/core/benches/banking_stage.rs b/core/benches/banking_stage.rs
@@ -288,7 +288,12 @@ fn bench_banking(bencher: &mut Bencher, tx_type: TransactionType) {
         let cluster_info = {
             let keypair = Arc::new(Keypair::new());
             let node = Node::new_localhost_with_pubkey(&keypair.pubkey());
-            ClusterInfo::new(node.info, keypair, SocketAddrSpace::Unspecified)
+            ClusterInfo::new(
+                node.info,
+                keypair,
+                SocketAddrSpace::Unspecified,
+                /*known_validators*/ None,
+            )
         };
         let cluster_info = Arc::new(cluster_info);
         let (s, _r) = unbounded();

diff --git a/core/src/accounts_hash_verifier.rs b/core/src/accounts_hash_verifier.rs
@@ -601,7 +601,12 @@ mod tests {
     fn new_test_cluster_info() -> ClusterInfo {
         let keypair = Arc::new(Keypair::new());
         let contact_info = ContactInfo::new_localhost(&keypair.pubkey(), timestamp());
-        ClusterInfo::new(contact_info, keypair, SocketAddrSpace::Unspecified)
+        ClusterInfo::new(
+            contact_info,
+            keypair,
+            SocketAddrSpace::Unspecified,
+            /*known_validators*/ None,
+        )
     }
 
     #[test]

diff --git a/core/src/banking_stage.rs b/core/src/banking_stage.rs
@@ -655,8 +655,12 @@ mod tests {
     pub(crate) fn new_test_cluster_info(keypair: Option<Arc<Keypair>>) -> (Node, ClusterInfo) {
         let keypair = keypair.unwrap_or_else(|| Arc::new(Keypair::new()));
         let node = Node::new_localhost_with_pubkey(&keypair.pubkey());
-        let cluster_info =
-            ClusterInfo::new(node.info.clone(), keypair, SocketAddrSpace::Unspecified);
+        let cluster_info = ClusterInfo::new(
+            node.info.clone(),
+            keypair,
+            SocketAddrSpace::Unspecified,
+            /*known_validators*/ None,
+        );
         (node, cluster_info)
     }
 

diff --git a/core/src/cluster_slots_service.rs b/core/src/cluster_slots_service.rs
@@ -192,7 +192,12 @@ mod test {
         let keypair = Arc::new(Keypair::new());
         let pubkey = keypair.pubkey();
         let node_info = Node::new_localhost_with_pubkey(&pubkey);
-        let cluster_info = ClusterInfo::new(node_info.info, keypair, SocketAddrSpace::Unspecified);
+        let cluster_info = ClusterInfo::new(
+            node_info.info,
+            keypair,
+            SocketAddrSpace::Unspecified,
+            /*known_validators*/ None,
+        );
         ClusterSlotsService::update_lowest_slot(5, &cluster_info);
         cluster_info.flush_push_queue();
         let lowest = {

diff --git a/core/src/repair/ancestor_hashes_service.rs b/core/src/repair/ancestor_hashes_service.rs
@@ -1256,6 +1256,7 @@ mod test {
                 responder_node.info.clone(),
                 Arc::new(keypair),
                 SocketAddrSpace::Unspecified,
+                /*known_validators*/ None,
             );
             let responder_serve_repair = ServeRepair::new(
                 Arc::new(cluster_info),
@@ -1350,6 +1351,7 @@ mod test {
                 Node::new_localhost_with_pubkey(&keypair.pubkey()).info,
                 Arc::new(keypair),
                 SocketAddrSpace::Unspecified,
+                /*known_validators*/ None,
             ));
             let repair_whitelist = Arc::new(RwLock::new(HashSet::default()));
             let requester_serve_repair = ServeRepair::new(

diff --git a/core/src/repair/repair_service.rs b/core/src/repair/repair_service.rs
@@ -879,7 +879,12 @@ mod test {
     fn new_test_cluster_info() -> ClusterInfo {
         let keypair = Arc::new(Keypair::new());
         let contact_info = ContactInfo::new_localhost(&keypair.pubkey(), timestamp());
-        ClusterInfo::new(contact_info, keypair, SocketAddrSpace::Unspecified)
+        ClusterInfo::new(
+            contact_info,
+            keypair,
+            SocketAddrSpace::Unspecified,
+            /*known_validators*/ None,
+        )
     }
 
     #[test]

diff --git a/core/src/repair/serve_repair.rs b/core/src/repair/serve_repair.rs
@@ -1969,7 +1969,12 @@ mod tests {
     fn new_test_cluster_info() -> ClusterInfo {
         let keypair = Arc::new(Keypair::new());
         let contact_info = ContactInfo::new_localhost(&keypair.pubkey(), timestamp());
-        ClusterInfo::new(contact_info, keypair, SocketAddrSpace::Unspecified)
+        ClusterInfo::new(
+            contact_info,
+            keypair,
+            SocketAddrSpace::Unspecified,
+            /*known_validators*/ None,
+        )
     }
 
     #[test]

diff --git a/core/src/replay_stage.rs b/core/src/replay_stage.rs
@@ -4162,6 +4162,7 @@ pub(crate) mod tests {
             Node::new_localhost_with_pubkey(&my_pubkey).info,
             Arc::new(my_keypairs.node_keypair.insecure_clone()),
             SocketAddrSpace::Unspecified,
+            /*known_validators*/ None,
         );
         assert_eq!(my_pubkey, cluster_info.id());
 

diff --git a/core/src/tvu.rs b/core/src/tvu.rs
@@ -409,8 +409,12 @@ pub mod tests {
         let (repair_quic_endpoint_sender, _repair_quic_endpoint_receiver) =
             tokio::sync::mpsc::channel(/*buffer:*/ 128);
         //start cluster_info1
-        let cluster_info1 =
-            ClusterInfo::new(target1.info.clone(), keypair, SocketAddrSpace::Unspecified);
+        let cluster_info1 = ClusterInfo::new(
+            target1.info.clone(),
+            keypair,
+            SocketAddrSpace::Unspecified,
+            /*known_validators*/ None,
+        );
         cluster_info1.insert_info(leader.info);
         let cref1 = Arc::new(cluster_info1);
 

diff --git a/core/src/validator.rs b/core/src/validator.rs
@@ -728,6 +728,7 @@ impl Validator {
             node.info.clone(),
             identity_keypair.clone(),
             socket_addr_space,
+            config.known_validators.clone(),
         );
         cluster_info.set_contact_debug_interval(config.contact_debug_interval);
         cluster_info.set_entrypoints(cluster_entrypoints);
@@ -974,7 +975,6 @@ impl Validator {
                 ledger_path,
                 config.validator_exit.clone(),
                 exit.clone(),
-                config.known_validators.clone(),
                 rpc_override_health_check.clone(),
                 startup_verification_complete,
                 optimistically_confirmed_bank.clone(),
@@ -2610,6 +2610,7 @@ mod tests {
             ContactInfo::new_localhost(&node_keypair.pubkey(), timestamp()),
             node_keypair,
             SocketAddrSpace::Unspecified,
+            /*known_validators*/ None,
         );
 
         let (genesis_config, _mint_keypair) = create_genesis_config(1);

diff --git a/core/src/window_service.rs b/core/src/window_service.rs
@@ -361,7 +361,7 @@ impl WindowService {
         let (duplicate_sender, duplicate_receiver) = unbounded();
 
         let t_check_duplicate = Self::start_check_duplicate_thread(
-            cluster_info,
+            cluster_info.clone(),
             exit.clone(),
             blockstore.clone(),
             duplicate_receiver,
@@ -442,6 +442,7 @@ impl WindowService {
                 };
                 let mut metrics = BlockstoreInsertionMetrics::default();
                 let mut ws_metrics = WindowServiceMetrics::default();
+
                 let mut last_print = Instant::now();
                 while !exit.load(Ordering::Relaxed) {
                     if let Err(e) = run_insert(
@@ -581,6 +582,7 @@ mod test {
             contact_info,
             Arc::new(keypair),
             SocketAddrSpace::Unspecified,
+            /*known_validators*/ None,
         );
         run_check_duplicate(
             &cluster_info,
@@ -615,6 +617,7 @@ mod test {
             contact_info,
             Arc::new(keypair),
             SocketAddrSpace::Unspecified,
+            /*known_validators*/ None,
         ));
 
         // Start duplicate thread receiving and inserting duplicates

diff --git a/core/tests/epoch_accounts_hash.rs b/core/tests/epoch_accounts_hash.rs
@@ -130,6 +130,7 @@ impl TestEnvironment {
             ContactInfo::new_localhost(&node_id.pubkey(), timestamp()),
             Arc::clone(&node_id),
             SocketAddrSpace::Unspecified,
+            /*known_validators*/ None,
         ));
 
         let pruned_banks_receiver =

diff --git a/core/tests/snapshots.rs b/core/tests/snapshots.rs
@@ -505,7 +505,12 @@ fn test_concurrent_snapshot_packaging(
             timestamp(), // wallclock
             0u16,        // shred_version
         );
-        ClusterInfo::new(contact_info, keypair, SocketAddrSpace::Unspecified)
+        ClusterInfo::new(
+            contact_info,
+            keypair,
+            SocketAddrSpace::Unspecified,
+            /*known_validators*/ None,
+        )
     });
 
     let (snapshot_package_sender, snapshot_package_receiver) = crossbeam_channel::unbounded();
@@ -955,6 +960,7 @@ fn test_snapshots_with_background_services(
         ContactInfo::new_localhost(&node_keypair.pubkey(), timestamp()),
         node_keypair,
         SocketAddrSpace::Unspecified,
+        /*known_validators*/ None,
     ));
 
     let (pruned_banks_sender, pruned_banks_receiver) = unbounded();