fix flaky test_db_restart (#15291)

aptos-labs · Nov 20, 2024 · 1a4fe24 · 1a4fe24
1 parent e2af7c4
commit 1a4fe24
Show file tree

Hide file tree

Showing 4 changed files with 35 additions and 34 deletions.
diff --git a/consensus/src/transaction_shuffler/use_case_aware/mod.rs b/consensus/src/transaction_shuffler/use_case_aware/mod.rs
@@ -2,10 +2,7 @@
 // SPDX-License-Identifier: Apache-2.0
 
 use crate::transaction_shuffler::TransactionShuffler;
-use aptos_types::transaction::{
-    use_case::{UseCaseAwareTransaction, UseCaseKey},
-    SignedTransaction,
-};
+use aptos_types::transaction::{use_case::UseCaseKey, SignedTransaction};
 use iterator::ShuffledTransactionIterator;
 use std::fmt::Debug;
 
@@ -45,7 +42,9 @@ pub struct UseCaseAwareShuffler {
 
 #[cfg(any(test, feature = "fuzzing"))]
 impl UseCaseAwareShuffler {
-    pub fn shuffle_generic<Txn: UseCaseAwareTransaction + Debug>(
+    pub fn shuffle_generic<
+        Txn: aptos_types::transaction::use_case::UseCaseAwareTransaction + Debug,
+    >(
         &self,
         txns: Vec<Txn>,
     ) -> Vec<Txn> {

diff --git a/crates/aptos-rest-client/src/lib.rs b/crates/aptos-rest-client/src/lib.rs
@@ -765,7 +765,7 @@ impl Client {
                         if let Some(state) = aptos_error_response.state {
                             if expiration_timestamp_secs <= state.timestamp_usecs / 1_000_000 {
                                 if reached_mempool {
-                                    return Err(anyhow!("Transaction expired. It is guaranteed it will not be committed on chain.").into());
+                                    return Err(anyhow!("Used to be pending and now not found. Transaction expired. It is guaranteed it will not be committed on chain.").into());
                                 } else {
                                     // We want to know whether we ever got Pending state from the mempool,
                                     // to warn in case we didn't.

diff --git a/testsuite/forge/src/interface/aptos.rs b/testsuite/forge/src/interface/aptos.rs
@@ -332,20 +332,12 @@ pub async fn reconfig(
     transaction_factory: &TransactionFactory,
     root_account: Arc<LocalAccount>,
 ) -> State {
-    let aptos_version = client.get_aptos_version().await.unwrap();
-    let current = aptos_version.into_inner();
-    let current_version = *current.major.inner();
     let txns = {
-        vec![
-            root_account.sign_with_transaction_builder(transaction_factory.clone().payload(
-                aptos_stdlib::version_set_for_next_epoch(current_version + 1),
-            )),
-            root_account.sign_with_transaction_builder(
-                transaction_factory
-                    .clone()
-                    .payload(aptos_stdlib::aptos_governance_force_end_epoch_test_only()),
-            ),
-        ]
+        vec![root_account.sign_with_transaction_builder(
+            transaction_factory
+                .clone()
+                .payload(aptos_stdlib::aptos_governance_force_end_epoch_test_only()),
+        )]
     };
 
     submit_and_wait_reconfig(client, txns).await

diff --git a/testsuite/smoke-test/src/storage.rs b/testsuite/smoke-test/src/storage.rs
@@ -466,14 +466,16 @@ async fn do_transfer_or_reconfig(info: &mut AptosPublicInfo) -> Result<()> {
     const LOTS_MONEY: u64 = 100_000_000;
     let r = rand::random::<u64>() % 10;
     if r < 3 {
-        // reconfig
+        info!(
+            "{LINE} background task: triggering reconfig. Root account seq_num: {}. Ledger info: {:?}",
+            info.root_account().sequence_number(),
+            info.client().get_ledger_information().await.unwrap(),
+        );
         info.reconfig().await;
-    } else if r == 9 {
-        // drain backlog
-        let mut sender = info.create_and_fund_user_account(LOTS_MONEY).await?;
-        let receiver = info.create_and_fund_user_account(LOTS_MONEY).await?;
-        let pending_txn = info.transfer(&mut sender, &receiver, 1).await?;
-        info.client().wait_for_transaction(&pending_txn).await?;
+        info!(
+            "{LINE} background task: Reconfig done. Root account seq_num: {}",
+            info.root_account().sequence_number(),
+        );
     } else {
         let mut sender = info.create_and_fund_user_account(LOTS_MONEY).await?;
         let receiver = info.create_and_fund_user_account(LOTS_MONEY).await?;
@@ -531,23 +533,30 @@ async fn test_db_restart() {
     for round in 0..3 {
         info!("{LINE} Restart round {round}");
         for (v, vid) in restarting_validator_ids.iter().enumerate() {
-            info!("{LINE} Round {round}: Restarting validator {v}.");
-            info!(
-                "{LINE} ledger info: {:?}",
-                client.get_ledger_information().await.unwrap(),
-            );
             let validator = swarm.validator_mut(*vid).unwrap();
             // sometimes trigger reconfig right before the restart, to expose edge cases around
             // epoch change
             if rand::random::<usize>() % 3 == 0 {
-                info!("{LINE} Triggering reconfig right before restarting.");
+                info!(
+                    "{LINE} Triggering reconfig right before restarting. Root account seq_num: {}. Ledger info: {:?}",
+                    pub_chain_info.root_account().sequence_number(),
+                    client.get_ledger_information().await.unwrap(),
+                );
                 reconfig(
-                    &validator.rest_client(),
+                    &client,
                     &pub_chain_info.transaction_factory(),
                     pub_chain_info.root_account(),
                 )
                 .await;
+                info!(
+                    "{LINE} Reconfig done. Root account seq_num: {}",
+                    pub_chain_info.root_account().sequence_number(),
+                )
             }
+            info!(
+                "{LINE} Round {round}: Restarting validator {v}. ledger info: {:?}",
+                client.get_ledger_information().await.unwrap(),
+            );
             validator.restart().await.unwrap();
             swarm
                 .wait_for_all_nodes_to_catchup(Duration::from_secs(60))
@@ -558,11 +567,12 @@ async fn test_db_restart() {
         }
     }
 
-    info!("{LINE} Stopping background traffic, and check again that all validators are alive.");
+    info!("{LINE} Stopping background traffic, and make sure background task didn't panic.");
     quit_flag.store(true, Ordering::Release);
     // Make sure background thread didn't panic.
     background_traffic.await.unwrap();
 
+    info!("{LINE} Check again that all validators are alive.");
     swarm
         .wait_for_all_nodes_to_catchup(Duration::from_secs(60))
         .await