Async FilesystemStore

Author: joostjager

lightning-persister/Cargo.toml

@@ -17,6 +17,9 @@ rustdoc-args = ["--cfg", "docsrs"]
 bitcoin = "0.32.2"
 lightning = { version = "0.2.0", path = "../lightning" }
 
+# TODO: Make conditional?
+tokio = { version = "1.35", features = [ "macros", "rt", "rt-multi-thread", "sync", "time", "fs", "io-util" ] }
+
 [target.'cfg(windows)'.dependencies]
 windows-sys = { version = "0.48.0", default-features = false, features = ["Win32_Storage_FileSystem", "Win32_Foundation"] }
 

lightning-persister/src/fs_store.rs

@@ -30,43 +30,29 @@ fn path_to_windows_str<T: AsRef<OsStr>>(path: &T) -> Vec<u16> {
 	path.as_ref().encode_wide().chain(Some(0)).collect()
 }
 
-// The number of read/write/remove/list operations after which we clean up our `locks` HashMap.
-const GC_LOCK_INTERVAL: usize = 25;
-
 /// A [`KVStoreSync`] implementation that writes to and reads from the file system.
 pub struct FilesystemStore {
 	data_dir: PathBuf,
 	tmp_file_counter: AtomicUsize,
-	gc_counter: AtomicUsize,
-	locks: Mutex<HashMap<PathBuf, Arc<RwLock<()>>>>,
+
+	// Per path lock that ensures that we don't have concurrent writes to the same file. The lock also encapsulates the
+	// latest written version per key.
+	locks: Mutex<HashMap<PathBuf, Arc<RwLock<u64>>>>,
 }
 
 impl FilesystemStore {
 	/// Constructs a new [`FilesystemStore`].
 	pub fn new(data_dir: PathBuf) -> Self {
 		let locks = Mutex::new(HashMap::new());
 		let tmp_file_counter = AtomicUsize::new(0);
-		let gc_counter = AtomicUsize::new(1);
-		Self { data_dir, tmp_file_counter, gc_counter, locks }
+		Self { data_dir, tmp_file_counter, locks }
 	}
 
 	/// Returns the data directory.
 	pub fn get_data_dir(&self) -> PathBuf {
 		self.data_dir.clone()
 	}
 
-	fn garbage_collect_locks(&self) {
-		let gc_counter = self.gc_counter.fetch_add(1, Ordering::AcqRel);
-
-		if gc_counter % GC_LOCK_INTERVAL == 0 {
-			// Take outer lock for the cleanup.
-			let mut outer_lock = self.locks.lock().unwrap();
-
-			// Garbage collect all lock entries that are not referenced anymore.
-			outer_lock.retain(|_, v| Arc::strong_count(&v) > 1);
-		}
-	}
-
 	fn get_dest_dir_path(
 		&self, primary_namespace: &str, secondary_namespace: &str,
 	) -> std::io::Result<PathBuf> {
@@ -90,36 +76,12 @@ impl FilesystemStore {
 
 		Ok(dest_dir_path)
 	}
-}
-
-impl KVStoreSync for FilesystemStore {
-	fn read(
-		&self, primary_namespace: &str, secondary_namespace: &str, key: &str,
-	) -> lightning::io::Result<Vec<u8>> {
-		check_namespace_key_validity(primary_namespace, secondary_namespace, Some(key), "read")?;
-
-		let mut dest_file_path = self.get_dest_dir_path(primary_namespace, secondary_namespace)?;
-		dest_file_path.push(key);
-
-		let mut buf = Vec::new();
-		{
-			let inner_lock_ref = {
-				let mut outer_lock = self.locks.lock().unwrap();
-				Arc::clone(&outer_lock.entry(dest_file_path.clone()).or_default())
-			};
-			let _guard = inner_lock_ref.read().unwrap();
-
-			let mut f = fs::File::open(dest_file_path)?;
-			f.read_to_end(&mut buf)?;
-		}
-
-		self.garbage_collect_locks();
-
-		Ok(buf)
-	}
 
-	fn write(
+	/// Writes a specific version of a key to the filesystem. If a newer version has been written already, this function
+	/// returns early without writing.
+	pub(crate) fn write_version(
 		&self, primary_namespace: &str, secondary_namespace: &str, key: &str, buf: &[u8],
+		version: Option<u64>,
 	) -> lightning::io::Result<()> {
 		check_namespace_key_validity(primary_namespace, secondary_namespace, Some(key), "write")?;
 
@@ -153,7 +115,18 @@ impl KVStoreSync for FilesystemStore {
 				let mut outer_lock = self.locks.lock().unwrap();
 				Arc::clone(&outer_lock.entry(dest_file_path.clone()).or_default())
 			};
-			let _guard = inner_lock_ref.write().unwrap();
+			let mut last_written_version = inner_lock_ref.write().unwrap();
+
+			// If a version is provided, we check if we already have a newer version written. This is used in async
+			// contexts to realize eventual consistency.
+			if let Some(version) = version {
+				if version <= *last_written_version {
+					// If the version is not greater, we don't write the file.
+					return Ok(());
+				}
+
+				*last_written_version = version;
+			}
 
 			#[cfg(not(target_os = "windows"))]
 			{
@@ -200,10 +173,39 @@ impl KVStoreSync for FilesystemStore {
 			}
 		};
 
-		self.garbage_collect_locks();
-
 		res
 	}
+}
+
+impl KVStoreSync for FilesystemStore {
+	fn read(
+		&self, primary_namespace: &str, secondary_namespace: &str, key: &str,
+	) -> lightning::io::Result<Vec<u8>> {
+		check_namespace_key_validity(primary_namespace, secondary_namespace, Some(key), "read")?;
+
+		let mut dest_file_path = self.get_dest_dir_path(primary_namespace, secondary_namespace)?;
+		dest_file_path.push(key);
+
+		let mut buf = Vec::new();
+		{
+			let inner_lock_ref = {
+				let mut outer_lock = self.locks.lock().unwrap();
+				Arc::clone(&outer_lock.entry(dest_file_path.clone()).or_default())
+			};
+			let _guard = inner_lock_ref.read().unwrap();
+
+			let mut f = fs::File::open(dest_file_path)?;
+			f.read_to_end(&mut buf)?;
+		}
+
+		Ok(buf)
+	}
+
+	fn write(
+		&self, primary_namespace: &str, secondary_namespace: &str, key: &str, buf: &[u8],
+	) -> lightning::io::Result<()> {
+		self.write_version(primary_namespace, secondary_namespace, key, buf, None)
+	}
 
 	fn remove(
 		&self, primary_namespace: &str, secondary_namespace: &str, key: &str, lazy: bool,
@@ -295,8 +297,6 @@ impl KVStoreSync for FilesystemStore {
 			}
 		}
 
-		self.garbage_collect_locks();
-
 		Ok(())
 	}
 
@@ -325,8 +325,6 @@ impl KVStoreSync for FilesystemStore {
 			keys.push(key);
 		}
 
-		self.garbage_collect_locks();
-
 		Ok(keys)
 	}
 }

lightning-persister/src/fs_store_async.rs

@@ -0,0 +1,155 @@
+//! Objects related to [`FilesystemStoreAsync`] live here.
+
+use std::sync::{
+	atomic::{AtomicU64, Ordering},
+	Arc,
+};
+
+use crate::fs_store::FilesystemStore;
+use core::future::Future;
+use core::pin::Pin;
+use lightning::util::persist::{KVStore, KVStoreSync};
+
+/// An asynchronous extension of FilesystemStore, implementing the `KVStore` trait for async operations. It is shaped as
+/// a wrapper around an existing [`FilesystemStore`] so that the same locks are used. This allows both the sync and
+/// async interface to be used simultaneously.
+pub struct FilesystemStoreAsync {
+	inner: Arc<FilesystemStore>,
+
+	// Version counter to ensure that writes are applied in the correct order. It is assumed that read, list and remove
+	// operations aren't sensitive to the order of execution.
+	version_counter: AtomicU64,
+}
+
+impl FilesystemStoreAsync {
+	/// Creates a new instance of [`FilesystemStoreAsync`].
+	pub fn new(inner: Arc<FilesystemStore>) -> Self {
+		Self { inner, version_counter: AtomicU64::new(0) }
+	}
+}
+
+impl KVStore for FilesystemStoreAsync {
+	fn read(
+		&self, primary_namespace: &str, secondary_namespace: &str, key: &str,
+	) -> Pin<Box<dyn Future<Output = Result<Vec<u8>, lightning::io::Error>> + 'static + Send>> {
+		let primary_namespace = primary_namespace.to_string();
+		let secondary_namespace = secondary_namespace.to_string();
+		let key = key.to_string();
+		let this = Arc::clone(&self.inner);
+
+		Box::pin(async move {
+			tokio::task::spawn_blocking(move || {
+				this.read(&primary_namespace, &secondary_namespace, &key)
+			})
+			.await
+			.unwrap_or_else(|e| Err(lightning::io::Error::new(lightning::io::ErrorKind::Other, e)))
+		})
+	}
+
+	fn write(
+		&self, primary_namespace: &str, secondary_namespace: &str, key: &str, buf: &[u8],
+	) -> Pin<Box<dyn Future<Output = Result<(), lightning::io::Error>> + 'static + Send>> {
+		let primary_namespace = primary_namespace.to_string();
+		let secondary_namespace = secondary_namespace.to_string();
+		let key = key.to_string();
+		let buf = buf.to_vec();
+		let this = Arc::clone(&self.inner);
+
+		// Obtain a version number to retain the call sequence.
+		let version = self.version_counter.fetch_add(1, Ordering::SeqCst);
+
+		Box::pin(async move {
+			tokio::task::spawn_blocking(move || {
+				this.write_version(
+					&primary_namespace,
+					&secondary_namespace,
+					&key,
+					&buf,
+					Some(version),
+				)
+			})
+			.await
+			.unwrap_or_else(|e| Err(lightning::io::Error::new(lightning::io::ErrorKind::Other, e)))
+		})
+	}
+
+	fn remove(
+		&self, primary_namespace: &str, secondary_namespace: &str, key: &str, lazy: bool,
+	) -> Pin<Box<dyn Future<Output = Result<(), lightning::io::Error>> + 'static + Send>> {
+		let primary_namespace = primary_namespace.to_string();
+		let secondary_namespace = secondary_namespace.to_string();
+		let key = key.to_string();
+		let this = Arc::clone(&self.inner);
+
+		Box::pin(async move {
+			tokio::task::spawn_blocking(move || {
+				this.remove(&primary_namespace, &secondary_namespace, &key, lazy)
+			})
+			.await
+			.unwrap_or_else(|e| Err(lightning::io::Error::new(lightning::io::ErrorKind::Other, e)))
+		})
+	}
+
+	fn list(
+		&self, primary_namespace: &str, secondary_namespace: &str,
+	) -> Pin<Box<dyn Future<Output = Result<Vec<String>, lightning::io::Error>> + 'static + Send>> {
+		let primary_namespace = primary_namespace.to_string();
+		let secondary_namespace = secondary_namespace.to_string();
+		let this = Arc::clone(&self.inner);
+
+		Box::pin(async move {
+			tokio::task::spawn_blocking(move || this.list(&primary_namespace, &secondary_namespace))
+				.await
+				.unwrap_or_else(|e| {
+					Err(lightning::io::Error::new(lightning::io::ErrorKind::Other, e))
+				})
+		})
+	}
+}
+
+mod test {
+	use crate::{fs_store::FilesystemStore, fs_store_async::FilesystemStoreAsync};
+	use lightning::util::persist::KVStore;
+	use std::sync::Arc;
+
+	#[tokio::test]
+	async fn read_write_remove_list_persist() {
+		let mut temp_path = std::env::temp_dir();
+		temp_path.push("test_read_write_remove_list_persist");
+		let fs_store = Arc::new(FilesystemStore::new(temp_path));
+		let fs_store_async = FilesystemStoreAsync::new(Arc::clone(&fs_store));
+
+		let data1 = [42u8; 32];
+		let data2 = [43u8; 32];
+
+		let primary_namespace = "testspace";
+		let secondary_namespace = "testsubspace";
+		let key = "testkey";
+
+		// Test writing the same key twice with different data. Execute the asynchronous part out of order to ensure
+		// that eventual consistency works.
+		let fut1 = fs_store_async.write(primary_namespace, secondary_namespace, key, &data1);
+		let fut2 = fs_store_async.write(primary_namespace, secondary_namespace, key, &data2);
+
+		fut2.await.unwrap();
+		fut1.await.unwrap();
+
+		// Test list.
+		let listed_keys =
+			fs_store_async.list(primary_namespace, secondary_namespace).await.unwrap();
+		assert_eq!(listed_keys.len(), 1);
+		assert_eq!(listed_keys[0], key);
+
+		// Test read. We expect to read data2, as the write call was initiated later.
+		let read_data =
+			fs_store_async.read(primary_namespace, secondary_namespace, key).await.unwrap();
+		assert_eq!(data2, &*read_data);
+
+		// Test remove.
+		fs_store_async.remove(primary_namespace, secondary_namespace, key, false).await.unwrap();
+
+		let listed_keys =
+			fs_store_async.list(primary_namespace, secondary_namespace).await.unwrap();
+		assert_eq!(listed_keys.len(), 0);
+	}
+}

lightning-persister/src/lib.rs

@@ -9,6 +9,7 @@
 extern crate criterion;
 
 pub mod fs_store;
+pub mod fs_store_async;
 
 mod utils;