Hotfix v1.2.0 david (#6436)

* [backup-cli] print name when deleting stale metadata cache file * [backup-cli] replay-verify: resumable * [backup-cli] replay-verify: fix off-by-1 issue with --txns-to-skip --------- Co-authored-by: aldenhu <[email protected]>
aptos-labs · Feb 1, 2023 · 895ae28 · 895ae28
1 parent e169630
commit 895ae28
Show file tree

Hide file tree

Showing 4 changed files with 33 additions and 12 deletions.
diff --git a/execution/executor/src/chunk_executor.rs b/execution/executor/src/chunk_executor.rs
@@ -355,7 +355,7 @@ impl<V: VMExecutor> TransactionReplayer for ChunkExecutorInner<V> {
 
             transactions = remaining;
             transaction_infos = remaining_info;
-            offset = version;
+            offset = version + 1;
         }
         self.replay_impl(transactions, transaction_infos)
     }
@@ -412,7 +412,7 @@ impl<V: VMExecutor> ChunkExecutorInner<V> {
 
         info!(
             "Overiding the output of txn at version: {:?}",
-            latest_view.version().unwrap(),
+            latest_view.version().map_or(0, |v| v + 1),
         );
 
         let chunk_output = ChunkOutput::by_transaction_output(

diff --git a/storage/backup/backup-cli/src/bin/replay-verify.rs b/storage/backup/backup-cli/src/bin/replay-verify.rs
@@ -68,6 +68,7 @@ async fn main_impl() -> Result<()> {
     Logger::new().level(Level::Info).init();
 
     let opt = Opt::from_args();
+
     let restore_handler = Arc::new(AptosDB::open(
         opt.db_dir,
         false,                       /* read_only */

diff --git a/storage/backup/backup-cli/src/coordinators/replay_verify.rs b/storage/backup/backup-cli/src/coordinators/replay_verify.rs
@@ -89,26 +89,46 @@ impl ReplayVerifyCoordinator {
             "start_version should precede end_version."
         );
 
-        let state_snapshot = if self.start_version == 0 {
-            None
+        let run_mode = Arc::new(RestoreRunMode::Restore {
+            restore_handler: self.restore_handler,
+        });
+        let next_txn_version = run_mode.get_next_expected_transaction_version()?;
+        let (state_snapshot, replay_transactions_from_version) = if next_txn_version != 0 {
+            // DB is already in workable state
+            info!(
+                next_txn_version = next_txn_version,
+                "DB already has non-empty State DB.",
+            );
+            (None, next_txn_version)
+        } else if let Some(version) = run_mode.get_in_progress_state_snapshot()? {
+            info!(
+                version = version,
+                "Found in progress state snapshot restore",
+            );
+            (Some(metadata_view.expect_state_snapshot(version)?), version)
+        } else if self.start_version == 0 {
+            (None, 0)
         } else {
-            metadata_view.select_state_snapshot(self.start_version.wrapping_sub(1))?
+            let state_snapshot = metadata_view.select_state_snapshot(self.start_version - 1)?;
+            let replay_transactions_from_version =
+                state_snapshot.as_ref().map(|b| b.version + 1).unwrap_or(0);
+            (state_snapshot, replay_transactions_from_version)
         };
-        let replay_transactions_from_version =
-            state_snapshot.as_ref().map(|b| b.version + 1).unwrap_or(0);
+        ensure!(
+            next_txn_version <= self.start_version,
+            "DB version is already beyond start_version requested.",
+        );
+
         let transactions = metadata_view.select_transaction_backups(
             // transaction info at the snapshot must be restored otherwise the db will be confused
             // about the latest version after snapshot is restored.
             replay_transactions_from_version.saturating_sub(1),
             self.end_version,
         )?;
-
         let global_opt = GlobalRestoreOptions {
             target_version: self.end_version,
             trusted_waypoints: Arc::new(self.trusted_waypoints_opt.verify()?),
-            run_mode: Arc::new(RestoreRunMode::Restore {
-                restore_handler: self.restore_handler,
-            }),
+            run_mode,
             concurrent_downloads: self.concurrent_downloads,
             replay_concurrency_level: 0, // won't replay, doesn't matter
         };

diff --git a/storage/backup/backup-cli/src/metadata/cache.rs b/storage/backup/backup-cli/src/metadata/cache.rs
@@ -123,7 +123,7 @@ pub async fn sync_and_load(
     for h in stale_local_hashes {
         let file = cache_dir.join(h);
         remove_file(&file).await.err_notes(&file)?;
-        info!("Deleted stale metadata files in cache.");
+        info!(file_name = h, "Deleted stale metadata file in cache.");
     }
 
     let num_new_files = new_remote_hashes.len();