fix(drive): uncommitted state if db transaction fails (#2305)

shumkov · QuantumExplorer · commit 315836524c83 · 2024-11-10T00:15:50.000+01:00
diff --git a/packages/rs-drive-abci/src/abci/handler/finalize_block.rs b/packages/rs-drive-abci/src/abci/handler/finalize_block.rs
@@ -5,6 +5,7 @@ use crate::execution::types::block_execution_context::v0::BlockExecutionContextV
 use crate::platform_types::cleaned_abci_messages::finalized_block_cleaned_request::v0::FinalizeBlockCleanedRequest;
 use crate::platform_types::platform_state::v0::PlatformStateV0Methods;
 use crate::rpc::core::CoreRPCLike;
+use dpp::dashcore::Network;
 use std::sync::atomic::Ordering;
 use tenderdash_abci::proto::abci as proto;
 
@@ -66,7 +67,30 @@ where
         ));
     }
 
-    app.commit_transaction(platform_version)?;
+    let result = app.commit_transaction(platform_version);
+
+    // We had a sequence of errors on the mainnet started since block 32326.
+    // We got RocksDB's "transaction is busy" error because of a bug (https://github.com/dashpay/platform/pull/2309).
+    // Due to another bug in Tenderdash (https://github.com/dashpay/tenderdash/pull/966),
+    // validators just proceeded to the next block partially committing the state and updating the cache.
+    // Full nodes are stuck and proceeded after re-sync.
+    // For the mainnet chain, we enable these fixes at the block when we consider the state is consistent.
+    let config = &app.platform().config;
+
+    if app.platform().config.network == Network::Dash
+        && config.abci.chain_id == "evo1"
+        && block_height < 33000
+    {
+        // Old behavior on mainnet below block 33000
+        result?;
+    } else {
+        // In case if transaction commit failed we still have caches in memory that
+        // corresponds to the data that we weren't able to commit.
+        // The simplified solution is to restart the Drive, so all caches
+        // will be restored from the disk and try to process this block again.
+        // TODO: We need a better handling of the transaction is busy error with retry logic.
+        result.expect("commit transaction");
+    }
 
     app.platform()
         .committed_block_height_guard
diff --git a/packages/rs-drive-abci/src/abci/handler/info.rs b/packages/rs-drive-abci/src/abci/handler/info.rs
@@ -3,6 +3,7 @@ use crate::abci::AbciError;
 use crate::error::Error;
 use crate::platform_types::platform_state::v0::PlatformStateV0Methods;
 use crate::rpc::core::CoreRPCLike;
+use dpp::dashcore::Network;
 use dpp::version::DESIRED_PLATFORM_VERSION;
 use tenderdash_abci::proto::abci as proto;
 
@@ -21,28 +22,67 @@ where
 
     let platform_state = app.platform().state.load();
 
-    let state_app_hash = platform_state
+    let last_block_height = platform_state.last_committed_block_height() as i64;
+
+    // Verify that Platform State corresponds to Drive commited state
+    let platform_state_app_hash = platform_state
         .last_committed_block_app_hash()
-        .map(|app_hash| app_hash.to_vec())
         .unwrap_or_default();
 
+    let grove_version = &platform_state
+        .current_platform_version()?
+        .drive
+        .grove_version;
+
+    let drive_storage_root_hash = app
+        .platform()
+        .drive
+        .grove
+        .root_hash(None, grove_version)
+        .unwrap()?;
+
+    // We had a sequence of errors on the mainnet started since block 32326.
+    // We got RocksDB's "transaction is busy" error because of a bug (https://github.com/dashpay/platform/pull/2309).
+    // Due to another bug in Tenderdash (https://github.com/dashpay/tenderdash/pull/966),
+    // validators just proceeded to the next block partially committing the state and updating the cache.
+    // Full nodes are stuck and proceeded after re-sync.
+    // For the mainnet chain, we enable these fixes at the block when we consider the state is consistent.
+    let config = &app.platform().config;
+
+    #[allow(clippy::collapsible_if)]
+    if !(config.network == Network::Dash
+        && config.abci.chain_id == "evo1"
+        && last_block_height < 33000)
+    {
+        // App hash in memory must be equal to app hash on disk
+        if drive_storage_root_hash != platform_state_app_hash {
+            // We panic because we can't recover from this situation.
+            // Better to restart the Drive, so we might self-heal the node
+            // reloading state form the disk
+            panic!(
+                "drive and platform state app hash mismatch: drive_storage_root_hash: {:?}, platform_state_app_hash: {:?}",
+                drive_storage_root_hash, platform_state_app_hash
+            );
+        }
+    }
+
     let desired_protocol_version = DESIRED_PLATFORM_VERSION.protocol_version;
 
     let response = proto::ResponseInfo {
         data: "".to_string(),
         app_version: desired_protocol_version as u64,
-        last_block_height: platform_state.last_committed_block_height() as i64,
+        last_block_height,
         version: env!("CARGO_PKG_VERSION").to_string(),
-        last_block_app_hash: state_app_hash.clone(),
+        last_block_app_hash: platform_state_app_hash.to_vec(),
     };
 
     tracing::debug!(
         desired_protocol_version,
         software_version = env!("CARGO_PKG_VERSION"),
         block_version = request.block_version,
         p2p_version = request.p2p_version,
-        app_hash = hex::encode(state_app_hash),
-        height = platform_state.last_committed_block_height(),
+        app_hash = hex::encode(platform_state_app_hash),
+        last_block_height,
         "Handshake with consensus engine",
     );
 
diff --git a/packages/rs-drive-abci/src/abci/handler/prepare_proposal.rs b/packages/rs-drive-abci/src/abci/handler/prepare_proposal.rs
@@ -11,6 +11,7 @@ use crate::platform_types::platform_state::v0::PlatformStateV0Methods;
 use crate::platform_types::state_transitions_processing_result::StateTransitionExecutionResult;
 use crate::rpc::core::CoreRPCLike;
 use dpp::dashcore::hashes::Hash;
+use dpp::dashcore::Network;
 use dpp::version::TryIntoPlatformVersioned;
 use drive::grovedb_storage::Error::RocksDBError;
 use tenderdash_abci::proto::abci as proto;
@@ -35,6 +36,48 @@ where
 
     let platform_state = app.platform().state.load();
 
+    // Verify that Platform State corresponds to Drive commited state
+    let platform_state_app_hash = platform_state
+        .last_committed_block_app_hash()
+        .unwrap_or_default();
+
+    let grove_version = &platform_state
+        .current_platform_version()?
+        .drive
+        .grove_version;
+
+    let drive_storage_root_hash = app
+        .platform()
+        .drive
+        .grove
+        .root_hash(None, grove_version)
+        .unwrap()?;
+
+    // We had a sequence of errors on the mainnet started since block 32326.
+    // We got RocksDB's "transaction is busy" error because of a bug (https://github.com/dashpay/platform/pull/2309).
+    // Due to another bug in Tenderdash (https://github.com/dashpay/tenderdash/pull/966),
+    // validators just proceeded to the next block partially committing the state and updating the cache.
+    // Full nodes are stuck and proceeded after re-sync.
+    // For the mainnet chain, we enable these fixes at the block when we consider the state is consistent.
+    let config = &app.platform().config;
+
+    #[allow(clippy::collapsible_if)]
+    if !(config.network == Network::Dash
+        && config.abci.chain_id == "evo1"
+        && request.height < 33000)
+    {
+        // App hash in memory must be equal to app hash on disk
+        if drive_storage_root_hash != platform_state_app_hash {
+            // We panic because we can't recover from this situation.
+            // Better to restart the Drive, so we might self-heal the node
+            // reloading state form the disk
+            panic!(
+                "drive and platform state app hash mismatch: drive_storage_root_hash: {:?}, platform_state_app_hash: {:?}",
+                drive_storage_root_hash, platform_state_app_hash
+            );
+        }
+    }
+
     let last_committed_core_height = platform_state.last_committed_core_height();
 
     let starting_platform_version = platform_state.current_platform_version()?;
diff --git a/packages/rs-drive-abci/src/abci/handler/process_proposal.rs b/packages/rs-drive-abci/src/abci/handler/process_proposal.rs
@@ -12,6 +12,7 @@ use crate::platform_types::block_execution_outcome;
 use crate::platform_types::platform_state::v0::PlatformStateV0Methods;
 use crate::platform_types::state_transitions_processing_result::StateTransitionExecutionResult;
 use crate::rpc::core::CoreRPCLike;
+use dpp::dashcore::Network;
 use dpp::version::TryIntoPlatformVersioned;
 use drive::grovedb_storage::Error::RocksDBError;
 use tenderdash_abci::proto::abci as proto;
@@ -179,6 +180,48 @@ where
 
     let platform_state = app.platform().state.load();
 
+    // Verify that Platform State corresponds to Drive commited state
+    let platform_state_app_hash = platform_state
+        .last_committed_block_app_hash()
+        .unwrap_or_default();
+
+    let grove_version = &platform_state
+        .current_platform_version()?
+        .drive
+        .grove_version;
+
+    let drive_storage_root_hash = app
+        .platform()
+        .drive
+        .grove
+        .root_hash(None, grove_version)
+        .unwrap()?;
+
+    // We had a sequence of errors on the mainnet started since block 32326.
+    // We got RocksDB's "transaction is busy" error because of a bug (https://github.com/dashpay/platform/pull/2309).
+    // Due to another bug in Tenderdash (https://github.com/dashpay/tenderdash/pull/966),
+    // validators just proceeded to the next block partially committing the state and updating the cache.
+    // Full nodes are stuck and proceeded after re-sync.
+    // For the mainnet chain, we enable these fixes at the block when we consider the state is consistent.
+    let config = &app.platform().config;
+
+    #[allow(clippy::collapsible_if)]
+    if !(app.platform().config.network == Network::Dash
+        && config.abci.chain_id == "evo1"
+        && request.height < 33000)
+    {
+        // App hash in memory must be equal to app hash on disk
+        if drive_storage_root_hash != platform_state_app_hash {
+            // We panic because we can't recover from this situation.
+            // Better to restart the Drive, so we might self-heal the node
+            // reloading state form the disk
+            panic!(
+                "drive and platform state app hash mismatch: drive_storage_root_hash: {:?}, platform_state_app_hash: {:?}",
+                drive_storage_root_hash, platform_state_app_hash
+            );
+        }
+    }
+
     let starting_platform_version = platform_state.current_platform_version()?;
 
     // Running the proposal executes all the state transitions for the block
