Allow contact debug interval to be adjusted (solana-labs#13737)

takeuchi5512 · Nov 20, 2020 · c1eb350 · c1eb350
1 parent ed82bf7
commit c1eb350
Show file tree

Hide file tree

Showing 3 changed files with 28 additions and 6 deletions.
diff --git a/core/src/cluster_info.rs b/core/src/cluster_info.rs
@@ -107,6 +107,7 @@ const MAX_PRUNE_DATA_NODES: usize = 32;
 const GOSSIP_PING_TOKEN_SIZE: usize = 32;
 const GOSSIP_PING_CACHE_CAPACITY: usize = 16384;
 const GOSSIP_PING_CACHE_TTL: Duration = Duration::from_secs(640);
+pub const DEFAULT_CONTACT_DEBUG_INTERVAL: u64 = 10_000;
 
 #[derive(Debug, PartialEq, Eq)]
 pub enum ClusterInfoError {
@@ -298,6 +299,7 @@ pub struct ClusterInfo {
     stats: GossipStats,
     socket: UdpSocket,
     local_message_pending_push_queue: RwLock<Vec<(CrdsValue, u64)>>,
+    contact_debug_interval: u64,
 }
 
 impl Default for ClusterInfo {
@@ -553,6 +555,7 @@ impl ClusterInfo {
             stats: GossipStats::default(),
             socket: UdpSocket::bind("0.0.0.0:0").unwrap(),
             local_message_pending_push_queue: RwLock::new(vec![]),
+            contact_debug_interval: DEFAULT_CONTACT_DEBUG_INTERVAL,
         };
         {
             let mut gossip = me.gossip.write().unwrap();
@@ -586,9 +589,14 @@ impl ClusterInfo {
                     .unwrap()
                     .clone(),
             ),
+            contact_debug_interval: self.contact_debug_interval,
         }
     }
 
+    pub fn set_contact_debug_interval(&mut self, new: u64) {
+        self.contact_debug_interval = new;
+    }
+
     pub fn update_contact_info<F>(&self, modify: F)
     where
         F: FnOnce(&mut ContactInfo),
@@ -1792,7 +1800,9 @@ impl ClusterInfo {
                 loop {
                     let start = timestamp();
                     thread_mem_usage::datapoint("solana-gossip");
-                    if start - last_contact_info_trace > 10000 {
+                    if self.contact_debug_interval != 0
+                        && start - last_contact_info_trace > self.contact_debug_interval
+                    {
                         // Log contact info every 10 seconds
                         info!(
                             "\n{}\n\n{}",

diff --git a/core/src/validator.rs b/core/src/validator.rs
@@ -3,7 +3,7 @@
 use crate::{
     broadcast_stage::BroadcastStageType,
     cache_block_time_service::{CacheBlockTimeSender, CacheBlockTimeService},
-    cluster_info::{ClusterInfo, Node},
+    cluster_info::{ClusterInfo, Node, DEFAULT_CONTACT_DEBUG_INTERVAL},
     cluster_info_vote_listener::VoteTracker,
     completed_data_sets_service::CompletedDataSetsService,
     consensus::{reconcile_blockstore_roots_with_tower, Tower},
@@ -105,6 +105,7 @@ pub struct ValidatorConfig {
     pub cuda: bool,
     pub require_tower: bool,
     pub debug_keys: Option<Arc<HashSet<Pubkey>>>,
+    pub contact_debug_interval: u64,
 }
 
 impl Default for ValidatorConfig {
@@ -140,6 +141,7 @@ impl Default for ValidatorConfig {
             cuda: false,
             require_tower: false,
             debug_keys: None,
+            contact_debug_interval: DEFAULT_CONTACT_DEBUG_INTERVAL,
         }
     }
 }
@@ -333,10 +335,9 @@ impl Validator {
             }
         }
 
-        let cluster_info = Arc::new(ClusterInfo::new(
-            node.info.clone(),
-            identity_keypair.clone(),
-        ));
+        let mut cluster_info = ClusterInfo::new(node.info.clone(), identity_keypair.clone());
+        cluster_info.set_contact_debug_interval(config.contact_debug_interval);
+        let cluster_info = Arc::new(cluster_info);
         let mut block_commitment_cache = BlockCommitmentCache::default();
         block_commitment_cache.initialize_slots(bank.slot());
         let block_commitment_cache = Arc::new(RwLock::new(block_commitment_cache));

diff --git a/validator/src/main.rs b/validator/src/main.rs
@@ -1099,6 +1099,14 @@ pub fn main() {
                 .help("Number of slots between generating snapshots, \
                       0 to disable snapshots"),
         )
+        .arg(
+            Arg::with_name("contact_debug_interval")
+                .long("contact-debug-interval")
+                .value_name("CONTACT_DEBUG_INTERVAL")
+                .takes_value(true)
+                .default_value("10000")
+                .help("Milliseconds between printing contact debug from gossip."),
+        )
         .arg(
             Arg::with_name("accounts_hash_interval_slots")
                 .long("accounts-hash-slots")
@@ -1435,6 +1443,8 @@ pub fn main() {
         bind_address
     };
 
+    let contact_debug_interval = value_t_or_exit!(matches, "contact_debug_interval", u64);
+
     let restricted_repair_only_mode = matches.is_present("restricted_repair_only_mode");
     let mut validator_config = ValidatorConfig {
         require_tower: matches.is_present("require_tower"),
@@ -1499,6 +1509,7 @@ pub fn main() {
         wal_recovery_mode,
         poh_verify: !matches.is_present("skip_poh_verify"),
         debug_keys,
+        contact_debug_interval,
         ..ValidatorConfig::default()
     };