[Forge] Add working_dir param to support running node on checkpoint d…

…ir, so that the existing data on disk is preserved. (aptos-labs#4591)

aptos-node/src/lib.rs

@@ -8,11 +8,11 @@ mod log_build_information;
 use anyhow::{anyhow, Context};
 use aptos_api::bootstrap as bootstrap_api;
 use aptos_build_info::build_information;
-use aptos_config::config::StateSyncConfig;
 use aptos_config::{
     config::{
         AptosDataClientConfig, BaseConfig, NetworkConfig, NodeConfig, PersistableConfig,
-        StorageServiceConfig,
+        RocksdbConfigs, StateSyncConfig, StorageServiceConfig, BUFFERED_STATE_TARGET_ITEMS,
+        DEFAULT_MAX_NUM_NODES_PER_LRU_CACHE_SHARD, NO_OP_STORAGE_PRUNER_CONFIG,
     },
     network_id::NetworkId,
     utils::get_genesis_txn,
@@ -26,6 +26,7 @@ use aptos_types::{
     account_config::CORE_CODE_ADDRESS, account_view::AccountView, chain_id::ChainId,
     on_chain_config::ON_CHAIN_CONFIG_REGISTRY, waypoint::Waypoint,
 };
+
 use aptos_vm::AptosVM;
 use aptosdb::AptosDB;
 use backup_service::start_backup_service;
@@ -53,8 +54,9 @@ use state_sync_driver::{
 use std::{
     boxed::Box,
     collections::{HashMap, HashSet},
+    fs,
     io::Write,
-    path::PathBuf,
+    path::{Path, PathBuf},
     sync::{
         atomic::{AtomicBool, AtomicUsize, Ordering},
         Arc,
@@ -569,8 +571,43 @@ fn bootstrap_indexer(
     Ok(None)
 }
 
+fn create_checkpoint_and_change_working_dir(
+    node_config: &mut NodeConfig,
+    working_dir: impl AsRef<Path>,
+) {
+    let source_dir = node_config.storage.dir();
+    node_config.set_data_dir(working_dir.as_ref().to_path_buf());
+    let checkpoint_dir = node_config.storage.dir();
+
+    assert!(source_dir != checkpoint_dir);
+
+    // Create rocksdb checkpoint.
+    fs::create_dir_all(&checkpoint_dir).unwrap();
+
+    AptosDB::open(
+        &source_dir,
+        false,                       /* readonly */
+        NO_OP_STORAGE_PRUNER_CONFIG, /* pruner */
+        RocksdbConfigs::default(),
+        false,
+        BUFFERED_STATE_TARGET_ITEMS,
+        DEFAULT_MAX_NUM_NODES_PER_LRU_CACHE_SHARD,
+    )
+    .expect("AptosDB open failure.")
+    .create_checkpoint(&checkpoint_dir)
+    .expect("AptosDB checkpoint creation failed.");
+
+    consensus::create_checkpoint(&source_dir, &checkpoint_dir)
+        .expect("ConsensusDB checkpoint creation failed.");
+    let state_sync_db =
+        state_sync_driver::metadata_storage::PersistentMetadataStorage::new(&source_dir);
+    state_sync_db
+        .create_checkpoint(&checkpoint_dir)
+        .expect("StateSyncDB checkpoint creation failed.");
+}
+
 pub fn setup_environment(
-    node_config: NodeConfig,
+    mut node_config: NodeConfig,
     remote_log_rx: Option<mpsc::Receiver<TelemetryLog>>,
     logger_filter_update_job: Option<LoggerFilterUpdater>,
 ) -> anyhow::Result<AptosHandle> {
@@ -580,6 +617,13 @@ pub fn setup_environment(
         inspection_service::inspection_service::start_inspection_service(node_config_clone)
     });
 
+    // If working_dir is provided, we will make RocksDb checkpoint for consensus_db,
+    // state_sync_db, ledger_db and state_merkle_db to the checkpoint_path, and running the node
+    // on the new path, so that the existing data won't change. For now this is a testonly feature.
+    if let Some(working_dir) = node_config.base.working_dir.clone() {
+        create_checkpoint_and_change_working_dir(&mut node_config, working_dir);
+    }
+
     // Open the database
     let mut instant = Instant::now();
     let (aptos_db, db_rw) = DbReaderWriter::wrap(

config/src/config/mod.rs

@@ -101,6 +101,7 @@ pub struct NodeConfig {
 #[serde(default, deny_unknown_fields)]
 pub struct BaseConfig {
     pub data_dir: PathBuf,
+    pub working_dir: Option<PathBuf>,
     pub role: RoleType,
     pub waypoint: WaypointConfig,
 }
@@ -109,6 +110,7 @@ impl Default for BaseConfig {
     fn default() -> BaseConfig {
         BaseConfig {
             data_dir: PathBuf::from("/opt/aptos/data"),
+            working_dir: None,
             role: RoleType::Validator,
             waypoint: WaypointConfig::None,
         }
@@ -266,6 +268,13 @@ impl NodeConfig {
         &self.base.data_dir
     }
 
+    pub fn working_dir(&self) -> &Path {
+        match &self.base.working_dir {
+            Some(working_dir) => working_dir,
+            None => &self.base.data_dir,
+        }
+    }
+
     pub fn set_data_dir(&mut self, data_dir: PathBuf) {
         self.base.data_dir = data_dir.clone();
         self.consensus.set_data_dir(data_dir.clone());

consensus/src/consensusdb/mod.rs

@@ -24,6 +24,22 @@ use std::{collections::HashMap, iter::Iterator, path::Path, time::Instant};
 /// The name of the consensus db file
 pub const CONSENSUS_DB_NAME: &str = "consensus_db";
 
+/// Creates new physical DB checkpoint in directory specified by `checkpoint_path`.
+pub fn create_checkpoint<P: AsRef<Path> + Clone>(db_path: P, checkpoint_path: P) -> Result<()> {
+    let start = Instant::now();
+    let consensus_db_checkpoint_path = checkpoint_path.as_ref().join(CONSENSUS_DB_NAME);
+    std::fs::remove_dir_all(&consensus_db_checkpoint_path).unwrap_or(());
+    ConsensusDB::new(db_path)
+        .db
+        .create_checkpoint(&consensus_db_checkpoint_path)?;
+    info!(
+        path = consensus_db_checkpoint_path,
+        time_ms = %start.elapsed().as_millis(),
+        "Made ConsensusDB checkpoint."
+    );
+    Ok(())
+}
+
 pub struct ConsensusDB {
     db: DB,
 }

consensus/src/lib.rs

@@ -51,6 +51,8 @@ pub mod network_interface;
 /// Required by the smoke tests
 pub use consensusdb::CONSENSUS_DB_NAME;
 
+pub use consensusdb::create_checkpoint;
+
 #[cfg(feature = "fuzzing")]
 pub use round_manager::round_manager_fuzzing;
 

docker/compose/aptos-node/validator.yaml

@@ -16,7 +16,7 @@ consensus:
       type: "local"
     backend:
       type: "on_disk_storage"
-      path: /opt/aptos/data/secure-data.json
+      path: secure-data.json
       namespace: ~
     initial_safety_rules_config:
       from_file:

state-sync/state-sync-v2/state-sync-driver/src/metadata_storage.rs

@@ -161,6 +161,20 @@ impl PersistentMetadataStorage {
             ))
         })
     }
+
+    /// Creates new physical DB checkpoint in directory specified by `path`.
+    pub fn create_checkpoint<P: AsRef<Path>>(&self, path: P) -> Result<()> {
+        let start = Instant::now();
+        let state_sync_db_path = path.as_ref().join(STATE_SYNC_DB_NAME);
+        std::fs::remove_dir_all(&state_sync_db_path).unwrap_or(());
+        self.database.create_checkpoint(&state_sync_db_path)?;
+        info!(
+            path = state_sync_db_path,
+            time_ms = %start.elapsed().as_millis(),
+            "Made StateSyncDB checkpoint."
+        );
+        Ok(())
+    }
 }
 
 impl MetadataStorageInterface for PersistentMetadataStorage {

storage/aptosdb/src/lib.rs

@@ -65,8 +65,6 @@ use aptos_crypto::hash::HashValue;
 use aptos_infallible::Mutex;
 use aptos_logger::prelude::*;
 use aptos_rocksdb_options::gen_rocksdb_options;
-use aptos_types::proof::TransactionAccumulatorSummary;
-use aptos_types::state_store::state_storage_usage::StateStorageUsage;
 use aptos_types::{
     account_address::AccountAddress,
     account_config::{new_block_event_key, NewBlockEvent},
@@ -77,12 +75,13 @@ use aptos_types::{
     ledger_info::LedgerInfoWithSignatures,
     proof::{
         accumulator::InMemoryAccumulator, AccumulatorConsistencyProof, SparseMerkleProofExt,
-        TransactionInfoListWithProof,
+        TransactionAccumulatorSummary, TransactionInfoListWithProof,
     },
     state_proof::StateProof,
     state_store::{
         state_key::StateKey,
         state_key_prefix::StateKeyPrefix,
+        state_storage_usage::StateStorageUsage,
         state_value::{StateValue, StateValueChunkWithProof},
         table::{TableHandle, TableInfo},
     },
@@ -108,12 +107,15 @@ use std::{
     time::{Duration, Instant},
 };
 
-use crate::pruner::{
-    ledger_pruner_manager::LedgerPrunerManager, ledger_store::ledger_store_pruner::LedgerPruner,
-    state_pruner_manager::StatePrunerManager, state_store::StateMerklePruner,
+use crate::{
+    pruner::{
+        ledger_pruner_manager::LedgerPrunerManager,
+        ledger_store::ledger_store_pruner::LedgerPruner, state_pruner_manager::StatePrunerManager,
+        state_store::StateMerklePruner,
+    },
+    stale_node_index::StaleNodeIndexSchema,
+    stale_node_index_cross_epoch::StaleNodeIndexCrossEpochSchema,
 };
-use crate::stale_node_index::StaleNodeIndexSchema;
-use crate::stale_node_index_cross_epoch::StaleNodeIndexCrossEpochSchema;
 use storage_interface::{
     state_delta::StateDelta, state_view::DbStateView, DbReader, DbWriter, ExecutedTrees, Order,
     StateSnapshotReceiver, MAX_REQUEST_LIMIT,
@@ -655,6 +657,8 @@ impl AptosDB {
         let start = Instant::now();
         let ledger_db_path = path.as_ref().join(LEDGER_DB_NAME);
         let state_merkle_db_path = path.as_ref().join(STATE_MERKLE_DB_NAME);
+        std::fs::remove_dir_all(&ledger_db_path).unwrap_or(());
+        std::fs::remove_dir_all(&state_merkle_db_path).unwrap_or(());
         self.ledger_db.create_checkpoint(&ledger_db_path)?;
         self.state_merkle_db
             .create_checkpoint(&state_merkle_db_path)?;

terraform/helm/aptos-node/files/configs/validator-base.yaml

@@ -19,7 +19,7 @@ consensus:
       type: "local"
     backend:
       type: "on_disk_storage"
-      path: /opt/aptos/data/secure-data.json
+      path: secure-data.json
       namespace: ~
     initial_safety_rules_config:
       from_file:

testsuite/forge/src/backend/local/node.rs

@@ -282,7 +282,7 @@ impl Node for LocalNode {
         let node_config = self.config();
         let ledger_db_path = node_config.storage.dir().join(LEDGER_DB_NAME);
         let state_db_path = node_config.storage.dir().join(STATE_MERKLE_DB_NAME);
-        let secure_storage_path = node_config.base.data_dir.join("secure_storage.json");
+        let secure_storage_path = node_config.working_dir().join("secure_storage.json");
         let state_sync_db_path = node_config.storage.dir().join(STATE_SYNC_DB_NAME);
 
         debug!(