v1.18 - Add detailed metrics reporting for packet filtering (#1423)

* Add detailed metrics reporting for packet filtering

* fix conflicts

Author: jstarry

core/src/banking_stage.rs

@@ -68,6 +68,7 @@ mod latest_unprocessed_votes;
 mod leader_slot_timing_metrics;
 mod multi_iterator_scanner;
 mod packet_deserializer;
+mod packet_filter;
 mod packet_receiver;
 mod read_write_account_set;
 #[allow(dead_code)]

core/src/banking_stage/immutable_deserialized_packet.rs

@@ -1,5 +1,5 @@
 use {
-    solana_cost_model::block_cost_limits::BUILT_IN_INSTRUCTION_COSTS,
+    super::packet_filter::PacketFilterFailure,
     solana_perf::packet::Packet,
     solana_runtime::transaction_priority_details::{
         GetTransactionPriorityDetails, TransactionPriorityDetails,
@@ -9,7 +9,6 @@ use {
         hash::Hash,
         message::Message,
         sanitize::SanitizeError,
-        saturating_add_assign,
         short_vec::decode_shortu16_len,
         signature::Signature,
         transaction::{
@@ -36,6 +35,8 @@ pub enum DeserializedPacketError {
     PrioritizationFailure,
     #[error("vote transaction failure")]
     VoteTransactionError,
+    #[error("Packet filter failure: {0}")]
+    FailedFilter(#[from] PacketFilterFailure),
 }
 
 #[derive(Debug, PartialEq, Eq)]
@@ -102,22 +103,6 @@ impl ImmutableDeserializedPacket {
         self.priority_details.clone()
     }
 
-    /// Returns true if the transaction's compute unit limit is at least as
-    /// large as the sum of the static builtins' costs.
-    /// This is a simple sanity check so the leader can discard transactions
-    /// which are statically known to exceed the compute budget, and will
-    /// result in no useful state-change.
-    pub fn compute_unit_limit_above_static_builtins(&self) -> bool {
-        let mut static_builtin_cost_sum: u64 = 0;
-        for (program_id, _) in self.transaction.get_message().program_instructions_iter() {
-            if let Some(ix_cost) = BUILT_IN_INSTRUCTION_COSTS.get(program_id) {
-                saturating_add_assign!(static_builtin_cost_sum, *ix_cost);
-            }
-        }
-
-        self.compute_unit_limit() >= static_builtin_cost_sum
-    }
-
     // This function deserializes packets into transactions, computes the blake3 hash of transaction
     // messages, and verifies secp256k1 instructions.
     pub fn build_sanitized_transaction(
@@ -196,7 +181,11 @@ mod tests {
         // 1. compute_unit_limit under static builtins
         // 2. compute_unit_limit equal to static builtins
         // 3. compute_unit_limit above static builtins
-        for (cu_limit, expectation) in [(250, false), (300, true), (350, true)] {
+        for (cu_limit, expectation) in [
+            (250, Err(PacketFilterFailure::InsufficientComputeLimit)),
+            (300, Ok(())),
+            (350, Ok(())),
+        ] {
             let keypair = Keypair::new();
             let bpf_program_id = Pubkey::new_unique();
             let ixs = vec![
@@ -213,7 +202,7 @@ mod tests {
             let packet = Packet::from_data(None, tx).unwrap();
             let deserialized_packet = ImmutableDeserializedPacket::new(packet).unwrap();
             assert_eq!(
-                deserialized_packet.compute_unit_limit_above_static_builtins(),
+                deserialized_packet.check_insufficent_compute_unit_limit(),
                 expectation
             );
         }

core/src/banking_stage/leader_slot_metrics.rs

@@ -1,6 +1,7 @@
 use {
     super::{
         leader_slot_timing_metrics::{LeaderExecuteAndCommitTimings, LeaderSlotTimingMetrics},
+        packet_deserializer::PacketReceiverStats,
         unprocessed_transaction_storage::InsertPacketBatchSummary,
     },
     solana_accounts_db::transaction_error_metrics::*,
@@ -64,6 +65,21 @@ struct LeaderSlotPacketCountMetrics {
     // total number of packets TPU received from sigverify that failed signature verification.
     newly_failed_sigverify_count: u64,
 
+    // total number of packets filtered due to sanitization failures during receiving from sigverify
+    failed_sanitization_count: u64,
+
+    // total number of packets filtered due to prioritization failures during receiving from sigverify
+    failed_prioritization_count: u64,
+
+    // total number of packets filtered due to insufficient compute limits during receiving from sigverify
+    insufficient_compute_limit_count: u64,
+
+    // total number of packets filtered due to excessive precompile signatures during receiving from sigverify
+    excessive_precompile_count: u64,
+
+    // total number of invalid vote packets filtered out during receiving from sigverify
+    invalid_votes_count: u64,
+
     // total number of dropped packet due to the thread's buffered packets capacity being reached.
     exceeded_buffer_limit_dropped_packets_count: u64,
 
@@ -148,111 +164,124 @@ impl LeaderSlotPacketCountMetrics {
     fn report(&self, id: u32, slot: Slot) {
         datapoint_info!(
             "banking_stage-leader_slot_packet_counts",
-            ("id", id as i64, i64),
-            ("slot", slot as i64, i64),
+            ("id", id, i64),
+            ("slot", slot, i64),
+            ("total_new_valid_packets", self.total_new_valid_packets, i64),
             (
-                "total_new_valid_packets",
-                self.total_new_valid_packets as i64,
+                "newly_failed_sigverify_count",
+                self.newly_failed_sigverify_count,
                 i64
             ),
             (
-                "newly_failed_sigverify_count",
-                self.newly_failed_sigverify_count as i64,
+                "failed_sanitization_count",
+                self.failed_sanitization_count,
                 i64
             ),
+            (
+                "failed_prioritization_count",
+                self.failed_prioritization_count,
+                i64
+            ),
+            (
+                "insufficient_compute_limit_count",
+                self.insufficient_compute_limit_count,
+                i64
+            ),
+            (
+                "excessive_precompile_count",
+                self.excessive_precompile_count,
+                i64
+            ),
+            ("invalid_votes_count", self.invalid_votes_count, i64),
             (
                 "exceeded_buffer_limit_dropped_packets_count",
-                self.exceeded_buffer_limit_dropped_packets_count as i64,
+                self.exceeded_buffer_limit_dropped_packets_count,
                 i64
             ),
             (
                 "newly_buffered_packets_count",
-                self.newly_buffered_packets_count as i64,
+                self.newly_buffered_packets_count,
                 i64
             ),
             (
                 "retryable_packets_filtered_count",
-                self.retryable_packets_filtered_count as i64,
+                self.retryable_packets_filtered_count,
                 i64
             ),
             (
                 "transactions_attempted_execution_count",
-                self.transactions_attempted_execution_count as i64,
+                self.transactions_attempted_execution_count,
                 i64
             ),
             (
                 "committed_transactions_count",
-                self.committed_transactions_count as i64,
+                self.committed_transactions_count,
                 i64
             ),
             (
                 "committed_transactions_with_successful_result_count",
-                self.committed_transactions_with_successful_result_count as i64,
+                self.committed_transactions_with_successful_result_count,
                 i64
             ),
             (
                 "retryable_errored_transaction_count",
-                self.retryable_errored_transaction_count as i64,
-                i64
-            ),
-            (
-                "retryable_packets_count",
-                self.retryable_packets_count as i64,
+                self.retryable_errored_transaction_count,
                 i64
             ),
+            ("retryable_packets_count", self.retryable_packets_count, i64),
             (
                 "nonretryable_errored_transactions_count",
-                self.nonretryable_errored_transactions_count as i64,
+                self.nonretryable_errored_transactions_count,
                 i64
             ),
             (
                 "executed_transactions_failed_commit_count",
-                self.executed_transactions_failed_commit_count as i64,
+                self.executed_transactions_failed_commit_count,
                 i64
             ),
             (
                 "account_lock_throttled_transactions_count",
-                self.account_lock_throttled_transactions_count as i64,
+                self.account_lock_throttled_transactions_count,
                 i64
             ),
             (
                 "account_locks_limit_throttled_transactions_count",
-                self.account_locks_limit_throttled_transactions_count as i64,
+                self.account_locks_limit_throttled_transactions_count,
                 i64
             ),
             (
                 "cost_model_throttled_transactions_count",
-                self.cost_model_throttled_transactions_count as i64,
+                self.cost_model_throttled_transactions_count,
                 i64
             ),
             (
                 "failed_forwarded_packets_count",
-                self.failed_forwarded_packets_count as i64,
+                self.failed_forwarded_packets_count,
                 i64
             ),
             (
                 "successful_forwarded_packets_count",
-                self.successful_forwarded_packets_count as i64,
+                self.successful_forwarded_packets_count,
                 i64
             ),
             (
                 "packet_batch_forward_failure_count",
-                self.packet_batch_forward_failure_count as i64,
+                self.packet_batch_forward_failure_count,
                 i64
             ),
             (
                 "cleared_from_buffer_after_forward_count",
-                self.cleared_from_buffer_after_forward_count as i64,
+                self.cleared_from_buffer_after_forward_count,
                 i64
             ),
             (
                 "forwardable_batches_count",
-                self.forwardable_batches_count as i64,
+                self.forwardable_batches_count,
                 i64
             ),
             (
                 "end_of_slot_unprocessed_buffer_len",
-                self.end_of_slot_unprocessed_buffer_len as i64,
+                self.end_of_slot_unprocessed_buffer_len,
                 i64
             ),
         );
@@ -559,24 +588,34 @@ impl LeaderSlotMetricsTracker {
     }
 
     // Packet inflow/outflow/processing metrics
-    pub(crate) fn increment_total_new_valid_packets(&mut self, count: u64) {
+    pub(crate) fn increment_received_packet_counts(&mut self, stats: PacketReceiverStats) {
         if let Some(leader_slot_metrics) = &mut self.leader_slot_metrics {
+            let metrics = &mut leader_slot_metrics.packet_count_metrics;
+            let PacketReceiverStats {
+                passed_sigverify_count,
+                failed_sigverify_count,
+                invalid_vote_count,
+                failed_prioritization_count,
+                failed_sanitization_count,
+                excessive_precompile_count,
+                insufficient_compute_limit_count,
+            } = stats;
+
+            saturating_add_assign!(metrics.total_new_valid_packets, passed_sigverify_count);
+            saturating_add_assign!(metrics.newly_failed_sigverify_count, failed_sigverify_count);
+            saturating_add_assign!(metrics.invalid_votes_count, invalid_vote_count);
             saturating_add_assign!(
-                leader_slot_metrics
-                    .packet_count_metrics
-                    .total_new_valid_packets,
-                count
+                metrics.failed_prioritization_count,
+                failed_prioritization_count
             );
-        }
-    }
-
-    pub(crate) fn increment_newly_failed_sigverify_count(&mut self, count: u64) {
-        if let Some(leader_slot_metrics) = &mut self.leader_slot_metrics {
+            saturating_add_assign!(metrics.failed_sanitization_count, failed_sanitization_count);
             saturating_add_assign!(
-                leader_slot_metrics
-                    .packet_count_metrics
-                    .newly_failed_sigverify_count,
-                count
+                metrics.excessive_precompile_count,
+                excessive_precompile_count
+            );
+            saturating_add_assign!(
+                metrics.insufficient_compute_limit_count,
+                insufficient_compute_limit_count
             );
         }
     }