perf(db): decode and decompress directly from borrowed data instead (#2050)

crates/storage/codecs/src/lib.rs

@@ -96,8 +96,9 @@ where
     }
 
     fn from_compact(mut buf: &[u8], _: usize) -> (Self, &[u8]) {
-        let mut list = vec![];
         let length = buf.get_u16();
+        let mut list = Vec::with_capacity(length as usize);
+
         for _ in 0..length {
             #[allow(unused_assignments)]
             let mut element = T::default();
@@ -125,8 +126,9 @@ where
 
     /// To be used by fixed sized types like `Vec<H256>`.
     fn specialized_from_compact(mut buf: &[u8], len: usize) -> (Self, &[u8]) {
-        let mut list = vec![];
         let length = buf.get_u16();
+        let mut list = Vec::with_capacity(length as usize);
+
         for _ in 0..length {
             #[allow(unused_assignments)]
             let mut element = T::default();

crates/storage/db/Cargo.toml

@@ -12,7 +12,7 @@ description = "Staged syncing primitives used in reth."
 reth-primitives = { path = "../../primitives" }
 reth-interfaces = { path = "../../interfaces" }
 reth-codecs = { path = "../codecs" }
-reth-libmdbx = { path = "../libmdbx-rs", optional = true }
+reth-libmdbx = { path = "../libmdbx-rs", optional = true , features = ["return-borrowed"] }
 
 # codecs
 serde = { version = "1.0.*", default-features = false }

crates/storage/db/src/abstraction/table.rs

@@ -3,7 +3,7 @@ use crate::{
     transaction::{DbTx, DbTxMut},
     Error,
 };
-use reth_primitives::bytes::Bytes;
+
 use serde::Serialize;
 use std::{
     fmt::Debug,
@@ -22,7 +22,7 @@ pub trait Compress: Send + Sync + Sized + Debug {
 /// Trait that will transform the data to be read from the DB.
 pub trait Decompress: Send + Sync + Sized + Debug {
     /// Decompresses data coming from the database.
-    fn decompress<B: Into<Bytes>>(value: B) -> Result<Self, Error>;
+    fn decompress<B: AsRef<[u8]>>(value: B) -> Result<Self, Error>;
 }
 
 /// Trait that will transform the data to be saved in the DB.
@@ -37,7 +37,7 @@ pub trait Encode: Send + Sync + Sized + Debug {
 /// Trait that will transform the data to be read from the DB.
 pub trait Decode: Send + Sync + Sized + Debug {
     /// Decodes data coming from the database.
-    fn decode<B: Into<Bytes>>(value: B) -> Result<Self, Error>;
+    fn decode<B: AsRef<[u8]>>(value: B) -> Result<Self, Error>;
 }
 
 /// Generic trait that enforces the database key to implement [`Encode`] and [`Decode`].

crates/storage/db/src/tables/codecs/compact.rs

@@ -23,8 +23,8 @@ macro_rules! impl_compression_for_compact {
 
             impl Decompress for $name
             {
-                fn decompress<B: Into<bytes::Bytes>>(value: B) -> Result<$name, Error> {
-                    let value = value.into();
+                fn decompress<B: AsRef<[u8]>>(value: B) -> Result<$name, Error> {
+                    let value = value.as_ref();
                     let (obj, _) = Compact::from_compact(&value, value.len());
                     Ok(obj)
                 }

crates/storage/db/src/tables/codecs/scale.rs

@@ -2,7 +2,6 @@ use crate::{
     table::{Compress, Decompress},
     Error,
 };
-use parity_scale_codec::decode_from_bytes;
 use reth_primitives::*;
 
 mod sealed {
@@ -27,8 +26,8 @@ impl<T> Decompress for T
 where
     T: ScaleValue + parity_scale_codec::Decode + Sync + Send + std::fmt::Debug,
 {
-    fn decompress<B: Into<bytes::Bytes>>(value: B) -> Result<T, Error> {
-        decode_from_bytes(value.into()).map_err(|_| Error::DecodeError)
+    fn decompress<B: AsRef<[u8]>>(value: B) -> Result<T, Error> {
+        parity_scale_codec::Decode::decode(&mut value.as_ref()).map_err(|_| Error::DecodeError)
     }
 }
 

crates/storage/db/src/tables/models/accounts.rs

@@ -6,10 +6,7 @@ use crate::{
     Error,
 };
 use reth_codecs::Compact;
-use reth_primitives::{
-    bytes::{BufMut, Bytes},
-    Account, Address, TransitionId,
-};
+use reth_primitives::{bytes::BufMut, Account, Address, TransitionId};
 use serde::{Deserialize, Serialize};
 
 /// Account as it is saved inside [`AccountChangeSet`][crate::tables::AccountChangeSet].
@@ -89,12 +86,10 @@ impl Encode for TransitionIdAddress {
 }
 
 impl Decode for TransitionIdAddress {
-    fn decode<B: Into<Bytes>>(value: B) -> Result<Self, Error> {
-        let value: Bytes = value.into();
-
-        let num =
-            u64::from_be_bytes(value.as_ref()[..8].try_into().map_err(|_| Error::DecodeError)?);
-        let hash = Address::from_slice(&value.slice(8..));
+    fn decode<B: AsRef<[u8]>>(value: B) -> Result<Self, Error> {
+        let value = value.as_ref();
+        let num = u64::from_be_bytes(value[..8].try_into().map_err(|_| Error::DecodeError)?);
+        let hash = Address::from_slice(&value[8..]);
 
         Ok(TransitionIdAddress((num, hash)))
     }
@@ -121,7 +116,7 @@ mod test {
         let encoded = Encode::encode(key);
         assert_eq!(encoded, bytes);
 
-        let decoded: TransitionIdAddress = Decode::decode(encoded.to_vec()).unwrap();
+        let decoded: TransitionIdAddress = Decode::decode(encoded).unwrap();
         assert_eq!(decoded, key);
     }
 

crates/storage/db/src/tables/models/integer_list.rs

@@ -4,7 +4,7 @@ use crate::{
     table::{Compress, Decompress},
     Error,
 };
-use reth_primitives::{bytes::Bytes, IntegerList};
+use reth_primitives::IntegerList;
 
 impl Compress for IntegerList {
     type Compressed = Vec<u8>;
@@ -15,7 +15,7 @@ impl Compress for IntegerList {
 }
 
 impl Decompress for IntegerList {
-    fn decompress<B: Into<Bytes>>(value: B) -> Result<Self, Error> {
-        IntegerList::from_bytes(&value.into()).map_err(|_| Error::DecodeError)
+    fn decompress<B: AsRef<[u8]>>(value: B) -> Result<Self, Error> {
+        IntegerList::from_bytes(value.as_ref()).map_err(|_| Error::DecodeError)
     }
 }

crates/storage/db/src/tables/models/mod.rs

@@ -14,7 +14,7 @@ use crate::{
     table::{Decode, Encode},
     Error,
 };
-use reth_primitives::{bytes::Bytes, Address, H256};
+use reth_primitives::{Address, H256};
 
 /// Macro that implements [`Encode`] and [`Decode`] for uint types.
 macro_rules! impl_uints {
@@ -31,8 +31,7 @@ macro_rules! impl_uints {
 
             impl Decode for $name
             {
-                fn decode<B: Into<Bytes>>(value: B) -> Result<Self, Error> {
-                    let value: Bytes = value.into();
+                fn decode<B: AsRef<[u8]>>(value: B) -> Result<Self, Error> {
                     Ok(
                         $name::from_be_bytes(
                             value.as_ref().try_into().map_err(|_| Error::DecodeError)?
@@ -54,8 +53,8 @@ impl Encode for Vec<u8> {
 }
 
 impl Decode for Vec<u8> {
-    fn decode<B: Into<Bytes>>(value: B) -> Result<Self, Error> {
-        Ok(value.into().to_vec())
+    fn decode<B: AsRef<[u8]>>(value: B) -> Result<Self, Error> {
+        Ok(value.as_ref().to_vec())
     }
 }
 
@@ -67,8 +66,8 @@ impl Encode for Address {
 }
 
 impl Decode for Address {
-    fn decode<B: Into<Bytes>>(value: B) -> Result<Self, Error> {
-        Ok(Address::from_slice(&value.into()[..]))
+    fn decode<B: AsRef<[u8]>>(value: B) -> Result<Self, Error> {
+        Ok(Address::from_slice(value.as_ref()))
     }
 }
 impl Encode for H256 {
@@ -79,8 +78,8 @@ impl Encode for H256 {
 }
 
 impl Decode for H256 {
-    fn decode<B: Into<Bytes>>(value: B) -> Result<Self, Error> {
-        Ok(H256::from_slice(&value.into()[..]))
+    fn decode<B: AsRef<[u8]>>(value: B) -> Result<Self, Error> {
+        Ok(H256::from_slice(value.as_ref()))
     }
 }
 
@@ -92,7 +91,7 @@ impl Encode for String {
 }
 
 impl Decode for String {
-    fn decode<B: Into<Bytes>>(value: B) -> Result<Self, Error> {
-        String::from_utf8(value.into().to_vec()).map_err(|_| Error::DecodeError)
+    fn decode<B: AsRef<[u8]>>(value: B) -> Result<Self, Error> {
+        String::from_utf8(value.as_ref().to_vec()).map_err(|_| Error::DecodeError)
     }
 }

crates/storage/db/src/tables/models/sharded_key.rs

@@ -4,7 +4,7 @@ use crate::{
     table::{Decode, Encode},
     Error,
 };
-use reth_primitives::{bytes::Bytes, TransitionId};
+use reth_primitives::TransitionId;
 
 /// Number of indices in one shard.
 pub const NUM_OF_INDICES_IN_SHARD: usize = 100;
@@ -48,14 +48,14 @@ impl<T> Decode for ShardedKey<T>
 where
     T: Decode,
 {
-    fn decode<B: Into<Bytes>>(value: B) -> Result<Self, Error> {
-        let value: Bytes = value.into();
+    fn decode<B: AsRef<[u8]>>(value: B) -> Result<Self, Error> {
+        let value = value.as_ref();
+
         let tx_num_index = value.len() - 8;
 
-        let highest_tx_number = u64::from_be_bytes(
-            value.as_ref()[tx_num_index..].try_into().map_err(|_| Error::DecodeError)?,
-        );
-        let key = T::decode(value.slice(..tx_num_index))?;
+        let highest_tx_number =
+            u64::from_be_bytes(value[tx_num_index..].try_into().map_err(|_| Error::DecodeError)?);
+        let key = T::decode(&value[..tx_num_index])?;
 
         Ok(ShardedKey::new(key, highest_tx_number))
     }

crates/storage/db/src/tables/models/storage_sharded_key.rs

@@ -4,7 +4,7 @@ use crate::{
     table::{Decode, Encode},
     Error,
 };
-use reth_primitives::{bytes::Bytes, TransitionId, H160, H256};
+use reth_primitives::{TransitionId, H160, H256};
 
 use super::ShardedKey;
 
@@ -44,15 +44,14 @@ impl Encode for StorageShardedKey {
 }
 
 impl Decode for StorageShardedKey {
-    fn decode<B: Into<Bytes>>(value: B) -> Result<Self, Error> {
-        let value: Bytes = value.into();
+    fn decode<B: AsRef<[u8]>>(value: B) -> Result<Self, Error> {
+        let value = value.as_ref();
         let tx_num_index = value.len() - 8;
 
-        let highest_tx_number = u64::from_be_bytes(
-            value.as_ref()[tx_num_index..].try_into().map_err(|_| Error::DecodeError)?,
-        );
-        let address = H160::decode(value.slice(..20))?;
-        let storage_key = H256::decode(value.slice(20..52))?;
+        let highest_tx_number =
+            u64::from_be_bytes(value[tx_num_index..].try_into().map_err(|_| Error::DecodeError)?);
+        let address = H160::decode(&value[..20])?;
+        let storage_key = H256::decode(&value[20..52])?;
 
         Ok(Self { address, sharded_key: ShardedKey::new(storage_key, highest_tx_number) })
     }

crates/storage/db/src/tables/utils.rs

@@ -3,7 +3,7 @@ use crate::{
     table::{Decode, Decompress, Table},
     Error,
 };
-use reth_primitives::bytes::Bytes;
+
 use std::borrow::Cow;
 
 #[macro_export]
@@ -49,10 +49,15 @@ where
     T::Key: Decode,
     T::Value: Decompress,
 {
-    Ok((
-        Decode::decode(Bytes::from(kv.0.into_owned()))?,
-        Decompress::decompress(Bytes::from(kv.1.into_owned()))?,
-    ))
+    let key = match kv.0 {
+        Cow::Borrowed(k) => Decode::decode(k)?,
+        Cow::Owned(k) => Decode::decode(k)?,
+    };
+    let value = match kv.1 {
+        Cow::Borrowed(v) => Decompress::decompress(v)?,
+        Cow::Owned(v) => Decompress::decompress(v)?,
+    };
+    Ok((key, value))
 }
 
 /// Helper function to decode only a value from a `(key, value)` pair.
@@ -60,7 +65,10 @@ pub(crate) fn decode_value<'a, T>(kv: (Cow<'a, [u8]>, Cow<'a, [u8]>)) -> Result<
 where
     T: Table,
 {
-    Decompress::decompress(Bytes::from(kv.1.into_owned()))
+    Ok(match kv.1 {
+        Cow::Borrowed(v) => Decompress::decompress(v)?,
+        Cow::Owned(v) => Decompress::decompress(v)?,
+    })
 }
 
 /// Helper function to decode a value. It can be a key or subkey.
@@ -68,5 +76,8 @@ pub(crate) fn decode_one<T>(value: Cow<'_, [u8]>) -> Result<T::Value, Error>
 where
     T: Table,
 {
-    Decompress::decompress(Bytes::from(value.into_owned()))
+    Ok(match value {
+        Cow::Borrowed(v) => Decompress::decompress(v)?,
+        Cow::Owned(v) => Decompress::decompress(v)?,
+    })
 }

crates/storage/libmdbx-rs/Cargo.toml

@@ -23,6 +23,10 @@ ffi = { package = "reth-mdbx-sys", path = "./mdbx-sys" }
 
 lifetimed-bytes = { version = "0.1", optional = true }
 
+[features]
+default = []
+return-borrowed = []
+
 [dev-dependencies]
 pprof = { version = "0.11", features = ["flamegraph", "frame-pointer", "criterion"] }
 criterion = "0.4"

crates/storage/libmdbx-rs/src/codec.rs

@@ -1,4 +1,4 @@
-use crate::{error::mdbx_result, Error, TransactionKind};
+use crate::{Error, TransactionKind};
 use derive_more::*;
 use std::{borrow::Cow, slice};
 
@@ -29,15 +29,27 @@ impl<'tx> TableObject<'tx> for Cow<'tx, [u8]> {
 
     #[doc(hidden)]
     unsafe fn decode_val<K: TransactionKind>(
-        txn: *const ffi::MDBX_txn,
+        _txn: *const ffi::MDBX_txn,
         data_val: &ffi::MDBX_val,
     ) -> Result<Self, Error> {
-        let is_dirty = (!K::ONLY_CLEAN) && mdbx_result(ffi::mdbx_is_dirty(txn, data_val.iov_base))?;
-
         let s = slice::from_raw_parts(data_val.iov_base as *const u8, data_val.iov_len);
 
+        #[cfg(feature = "return-borrowed")]
+        {
+            Ok(Cow::Borrowed(s))
+        }
+
+        #[cfg(not(feature = "return-borrowed"))]
+        {
+            let is_dirty = (!K::ONLY_CLEAN) &&
+                crate::error::mdbx_result::mdbx_result(ffi::mdbx_is_dirty(
+                    _txn,
+                    data_val.iov_base,
+                ))?;
+
             Ok(if is_dirty { Cow::Owned(s.to_vec()) } else { Cow::Borrowed(s) })
         }
+    }
 }
 
 #[cfg(feature = "lifetimed-bytes")]