Thread manager instantiation in the validator (#4603)

* enable agave to start with default config for thread manager * make it possible to create thread manager without consuming the config * relax thread name length constraints --------- Co-authored-by: Alex Pyattaev <[email protected]>
anza-xyz · Jan 28, 2025 · e48672c · e48672c
1 parent 94841a8
commit e48672c
Show file tree

Hide file tree

Showing 15 changed files with 294 additions and 52 deletions.
diff --git a/Cargo.lock b/Cargo.lock
diff --git a/core/Cargo.toml b/core/Cargo.toml
@@ -15,6 +15,7 @@ codecov = { repository = "solana-labs/solana", branch = "master", service = "git
 
 [dependencies]
 agave-banking-stage-ingress-types = { workspace = true }
+agave-thread-manager = { workspace = true }
 agave-transaction-view = { workspace = true }
 ahash = { workspace = true }
 anyhow = { workspace = true }

diff --git a/core/src/validator.rs b/core/src/validator.rs
@@ -30,6 +30,7 @@ use {
         tpu::{Tpu, TpuSockets, DEFAULT_TPU_COALESCE},
         tvu::{Tvu, TvuConfig, TvuSockets},
     },
+    agave_thread_manager::{ThreadManager, ThreadManagerConfig},
     anyhow::{anyhow, Context, Result},
     crossbeam_channel::{bounded, unbounded, Receiver},
     lazy_static::lazy_static,
@@ -313,6 +314,7 @@ pub struct ValidatorConfig {
     pub replay_forks_threads: NonZeroUsize,
     pub replay_transactions_threads: NonZeroUsize,
     pub tvu_shred_sigverify_threads: NonZeroUsize,
+    pub thread_manager_config: ThreadManagerConfig,
     pub delay_leader_block_for_pending_fork: bool,
 }
 
@@ -385,6 +387,7 @@ impl Default for ValidatorConfig {
             rayon_global_threads: NonZeroUsize::new(1).expect("1 is non-zero"),
             replay_forks_threads: NonZeroUsize::new(1).expect("1 is non-zero"),
             replay_transactions_threads: NonZeroUsize::new(1).expect("1 is non-zero"),
+            thread_manager_config: ThreadManagerConfig::default_for_agave(),
             tvu_shred_sigverify_threads: NonZeroUsize::new(1).expect("1 is non-zero"),
             delay_leader_block_for_pending_fork: false,
         }
@@ -588,6 +591,7 @@ pub struct Validator {
     repair_quic_endpoints: Option<[Endpoint; 3]>,
     repair_quic_endpoints_runtime: Option<TokioRuntime>,
     repair_quic_endpoints_join_handle: Option<repair::quic_endpoint::AsyncTryJoinHandle>,
+    thread_manager: ThreadManager,
 }
 
 impl Validator {
@@ -619,6 +623,7 @@ impl Validator {
 
         let start_time = Instant::now();
 
+        let thread_manager = ThreadManager::new(&config.thread_manager_config)?;
         // Initialize the global rayon pool first to ensure the value in config
         // is honored. Otherwise, some code accessing the global pool could
         // cause it to get initialized with Rayon's default (not ours)
@@ -1659,6 +1664,7 @@ impl Validator {
             repair_quic_endpoints,
             repair_quic_endpoints_runtime,
             repair_quic_endpoints_join_handle,
+            thread_manager,
         })
     }
 
@@ -1816,6 +1822,7 @@ impl Validator {
         self.poh_timing_report_service
             .join()
             .expect("poh_timing_report_service");
+        self.thread_manager.destroy();
     }
 }
 

diff --git a/local-cluster/src/validator_configs.rs b/local-cluster/src/validator_configs.rs
@@ -73,6 +73,7 @@ pub fn safe_clone_config(config: &ValidatorConfig) -> ValidatorConfig {
         replay_forks_threads: config.replay_forks_threads,
         replay_transactions_threads: config.replay_transactions_threads,
         tvu_shred_sigverify_threads: config.tvu_shred_sigverify_threads,
+        thread_manager_config: config.thread_manager_config.clone(),
         delay_leader_block_for_pending_fork: config.delay_leader_block_for_pending_fork,
     }
 }

diff --git a/programs/sbf/Cargo.lock b/programs/sbf/Cargo.lock
diff --git a/svm/examples/Cargo.lock b/svm/examples/Cargo.lock
diff --git a/thread-manager/Cargo.toml b/thread-manager/Cargo.toml
@@ -21,6 +21,7 @@ serde = { workspace = true, features = ["derive"] }
 solana-metrics = { workspace = true }
 thread-priority = { workspace = true }
 tokio = { workspace = true, features = ["time", "rt-multi-thread"] }
+toml = { workspace = true }
 
 [target.'cfg(target_os = "linux")'.dependencies]
 affinity = "0.1.2"
@@ -31,7 +32,6 @@ axum = { workspace = true }
 env_logger = { workspace = true }
 hyper = { workspace = true, features = ["http1", "client", "stream", "tcp"] }
 serde_json = { workspace = true }
-toml = { workspace = true }
 tower = { workspace = true }
 
 [features]