Performance improvements:

- reduce contention on mutation_lock by having 256 of them
- better lmdb defaults

src/api/s3/put.rs

@@ -8,6 +8,11 @@ use hyper::{Request, Response};
 use md5::{digest::generic_array::*, Digest as Md5Digest, Md5};
 use sha2::Sha256;
 
+use opentelemetry::{
+	trace::{FutureExt as OtelFutureExt, TraceContextExt, Tracer},
+	Context,
+};
+
 use garage_table::*;
 use garage_util::async_hash::*;
 use garage_util::data::*;
@@ -279,12 +284,21 @@ async fn read_and_put_blocks<S: Stream<Item = Result<Bytes, Error>> + Unpin>(
 	first_block_hash: Hash,
 	chunker: &mut StreamChunker<S>,
 ) -> Result<(u64, GenericArray<u8, typenum::U16>, Hash), Error> {
+	let tracer = opentelemetry::global::tracer("garage");
+
 	let first_block = Bytes::from(first_block);
 
 	let md5hasher = AsyncHasher::<Md5>::new();
 	let sha256hasher = AsyncHasher::<Sha256>::new();
-	md5hasher.update(first_block.clone());
-	sha256hasher.update(first_block.clone());
+
+	futures::future::join(
+		md5hasher.update(first_block.clone()),
+		sha256hasher.update(first_block.clone()),
+	)
+	.with_context(Context::current_with_span(
+		tracer.start("Hash first block (md5, sha256)"),
+	))
+	.await;
 
 	let mut next_offset = first_block.len();
 	let mut put_curr_version_block = put_block_meta(
@@ -307,9 +321,15 @@ async fn read_and_put_blocks<S: Stream<Item = Result<Bytes, Error>> + Unpin>(
 		)?;
 		if let Some(block) = next_block {
 			let block = Bytes::from(block);
-			md5hasher.update(block.clone());
-			sha256hasher.update(block.clone());
-			let block_hash = async_blake2sum(block.clone()).await;
+			let (_, _, block_hash) = futures::future::join3(
+				md5hasher.update(block.clone()),
+				sha256hasher.update(block.clone()),
+				async_blake2sum(block.clone()),
+			)
+			.with_context(Context::current_with_span(
+				tracer.start("Hash block (md5, sha256, blake2)"),
+			))
+			.await;
 			let block_len = block.len();
 			put_curr_version_block = put_block_meta(
 				garage,

src/block/manager.rs

@@ -93,7 +93,7 @@ pub struct BlockManager {
 	compression_level: Option<i32>,
 	background_tranquility: u32,
 
-	mutation_lock: Mutex<BlockManagerLocked>,
+	mutation_lock: [Mutex<BlockManagerLocked>; 256],
 
 	pub(crate) rc: BlockRc,
 
@@ -150,8 +150,6 @@ impl BlockManager {
 			.netapp
 			.endpoint("garage_block/manager.rs/Rpc".to_string());
 
-		let manager_locked = BlockManagerLocked();
-
 		let metrics = BlockManagerMetrics::new(resync_queue.clone(), resync_errors.clone());
 
 		let block_manager = Arc::new(Self {
@@ -159,7 +157,7 @@ impl BlockManager {
 			data_dir,
 			compression_level,
 			background_tranquility,
-			mutation_lock: Mutex::new(manager_locked),
+			mutation_lock: [(); 256].map(|_| Mutex::new(BlockManagerLocked())),
 			rc,
 			resync_queue,
 			resync_notify: Notify::new(),
@@ -313,14 +311,21 @@ impl BlockManager {
 
 	/// Write a block to disk
 	async fn write_block(&self, hash: &Hash, data: &DataBlock) -> Result<BlockRpc, Error> {
+		let tracer = opentelemetry::global::tracer("garage");
+
 		let write_size = data.inner_buffer().len() as u64;
 
-		let res = self
-			.mutation_lock
+		let res = self.mutation_lock[hash.as_slice()[0] as usize]
 			.lock()
+			.with_context(Context::current_with_span(
+				tracer.start("Acquire mutation_lock"),
+			))
 			.await
 			.write_block(hash, data, self)
 			.bound_record_duration(&self.metrics.block_write_duration)
+			.with_context(Context::current_with_span(
+				tracer.start("BlockManagerLocked::write_block"),
+			))
 			.await?;
 
 		self.metrics.bytes_written.add(write_size);
@@ -370,7 +375,7 @@ impl BlockManager {
 		if data.verify(*hash).is_err() {
 			self.metrics.corruption_counter.add(1);
 
-			self.mutation_lock
+			self.mutation_lock[hash.as_slice()[0] as usize]
 				.lock()
 				.await
 				.move_block_to_corrupted(hash, self)
@@ -384,8 +389,7 @@ impl BlockManager {
 
 	/// Check if this node should have a block, but don't actually have it
 	async fn need_block(&self, hash: &Hash) -> Result<bool, Error> {
-		let BlockStatus { exists, needed } = self
-			.mutation_lock
+		let BlockStatus { exists, needed } = self.mutation_lock[hash.as_slice()[0] as usize]
 			.lock()
 			.await
 			.check_block_status(hash, self)
@@ -608,8 +612,7 @@ impl BlockManager {
 	}
 
 	async fn resync_block(&self, hash: &Hash) -> Result<(), Error> {
-		let BlockStatus { exists, needed } = self
-			.mutation_lock
+		let BlockStatus { exists, needed } = self.mutation_lock[hash.as_slice()[0] as usize]
 			.lock()
 			.await
 			.check_block_status(hash, self)
@@ -694,7 +697,7 @@ impl BlockManager {
 				who.len()
 			);
 
-			self.mutation_lock
+			self.mutation_lock[hash.as_slice()[0] as usize]
 				.lock()
 				.await
 				.delete_if_unneeded(hash, self)

src/model/garage.rs

@@ -104,11 +104,16 @@ impl Garage {
 				std::fs::create_dir_all(&db_path).expect("Unable to create LMDB data directory");
 				let map_size = garage_db::lmdb_adapter::recommended_map_size();
 
-				let db = db::lmdb_adapter::heed::EnvOpenOptions::new()
-					.max_dbs(100)
-					.map_size(map_size)
-					.open(&db_path)
-					.expect("Unable to open LMDB DB");
+				use db::lmdb_adapter::heed;
+				let mut env_builder = heed::EnvOpenOptions::new();
+				env_builder.max_dbs(100);
+				env_builder.max_readers(500);
+				env_builder.map_size(map_size);
+				unsafe {
+					env_builder.flag(heed::flags::Flags::MdbNoSync);
+					env_builder.flag(heed::flags::Flags::MdbNoMetaSync);
+				}
+				let db = env_builder.open(&db_path).expect("Unable to open LMDB DB");
 				db::lmdb_adapter::LmdbDb::init(db)
 			}
 			e => {

src/util/async_hash.rs

@@ -1,7 +1,7 @@
 use bytes::Bytes;
 use digest::Digest;
 
-use tokio::sync::mpsc;
+use tokio::sync::{mpsc, oneshot};
 use tokio::task::JoinHandle;
 
 use crate::data::*;
@@ -27,25 +27,28 @@ pub async fn async_blake2sum(data: Bytes) -> Hash {
 // ----
 
 pub struct AsyncHasher<D: Digest> {
-	sendblk: mpsc::UnboundedSender<Bytes>,
+	sendblk: mpsc::UnboundedSender<(Bytes, oneshot::Sender<()>)>,
 	task: JoinHandle<digest::Output<D>>,
 }
 
 impl<D: Digest> AsyncHasher<D> {
 	pub fn new() -> Self {
-		let (sendblk, mut recvblk) = mpsc::unbounded_channel::<Bytes>();
+		let (sendblk, mut recvblk) = mpsc::unbounded_channel::<(Bytes, oneshot::Sender<()>)>();
 		let task = tokio::task::spawn_blocking(move || {
 			let mut digest = D::new();
-			while let Some(blk) = recvblk.blocking_recv() {
+			while let Some((blk, ch)) = recvblk.blocking_recv() {
 				digest.update(&blk[..]);
+				let _ = ch.send(());
 			}
 			digest.finalize()
 		});
 		Self { sendblk, task }
 	}
 
-	pub fn update(&self, b: Bytes) {
-		self.sendblk.send(b).unwrap()
+	pub async fn update(&self, b: Bytes) {
+		let (tx, rx) = oneshot::channel();
+		self.sendblk.send((b, tx)).unwrap();
+		let _ = rx.await;
 	}
 
 	pub async fn finalize(self) -> digest::Output<D> {