fix(metrics): pass the epoch manager to spawn_trie_metrics_loop (#12811)

This function creates its own epoch manager with
new_from_genesis_config(), but this is not how the main epoch manager is
created at startup, which is done with new_arc_handle(). These result in
epoch managers that return different EpochConfigs for a given protocol
version, since only the second one initializes an epoch config store
from the genesis config, and the first one generates epoch configs with
generate_epoch_config(). In this case it leads to a crash if the shard
layouts are different because the call to
`ShardUId::from_shard_id_and_layout()` will see the wrong shard IDs and
crash at `assert!(shard_layout.shard_ids().any(|i| i == shard_id));`

nearcore/src/lib.rs

@@ -250,17 +250,19 @@ pub fn start_with_config_and_synchronization(
         None
     };
 
+    let epoch_manager = EpochManager::new_arc_handle(
+        storage.get_hot_store(),
+        &config.genesis.config,
+        Some(home_dir),
+    );
+
     let trie_metrics_arbiter = spawn_trie_metrics_loop(
         config.clone(),
         storage.get_hot_store(),
         config.client_config.log_summary_period,
+        epoch_manager.clone(),
     )?;
 
-    let epoch_manager = EpochManager::new_arc_handle(
-        storage.get_hot_store(),
-        &config.genesis.config,
-        Some(home_dir),
-    );
     let genesis_epoch_config = epoch_manager.get_epoch_config(&EpochId::default())?;
     // Initialize genesis_state in store either from genesis config or dump before other components.
     // We only initialize if the genesis state is not already initialized in store.

nearcore/src/metrics.rs

@@ -2,7 +2,7 @@ use crate::NearConfig;
 use actix_rt::ArbiterHandle;
 use near_async::time::Duration;
 use near_chain::{Block, ChainStore, ChainStoreAccess};
-use near_epoch_manager::EpochManager;
+use near_epoch_manager::EpochManagerAdapter;
 use near_o11y::metrics::{
     exponential_buckets, try_create_histogram_vec, try_create_int_counter_vec,
     try_create_int_gauge, try_create_int_gauge_vec, HistogramVec, IntCounterVec, IntGauge,
@@ -107,15 +107,17 @@ fn log_trie_item(key: Vec<u8>, value: Vec<u8>) {
     }
 }
 
-fn export_postponed_receipt_count(near_config: &NearConfig, store: &Store) -> anyhow::Result<()> {
+fn export_postponed_receipt_count(
+    near_config: &NearConfig,
+    store: &Store,
+    epoch_manager: &dyn EpochManagerAdapter,
+) -> anyhow::Result<()> {
     let chain_store = ChainStore::new(
         store.clone(),
         near_config.genesis.config.genesis_height,
         near_config.client_config.save_trie_changes,
         near_config.genesis.config.transaction_validity_period,
     );
-    let epoch_manager =
-        EpochManager::new_from_genesis_config(store.clone(), &near_config.genesis.config)?;
 
     let head = chain_store.final_head()?;
     let block = chain_store.get_block(&head.last_block_hash)?;
@@ -194,6 +196,7 @@ pub fn spawn_trie_metrics_loop(
     near_config: NearConfig,
     store: Store,
     period: Duration,
+    epoch_manager: Arc<dyn EpochManagerAdapter>,
 ) -> anyhow::Result<ArbiterHandle> {
     tracing::debug!(target:"metrics", "Spawning the trie metrics loop.");
     let arbiter = actix_rt::Arbiter::new();
@@ -208,7 +211,7 @@ pub fn spawn_trie_metrics_loop(
             interval.tick().await;
 
             let start_time = std::time::Instant::now();
-            let result = export_postponed_receipt_count(&near_config, &store);
+            let result = export_postponed_receipt_count(&near_config, &store, epoch_manager.as_ref());
             if let Err(err) = result {
                 tracing::error!(target: "metrics", "Error when exporting postponed receipts count {err}.");
             };