title: mini-redis项目-2-存储层 toc: true cover: 'https://img.paulzzh.com/touhou/random?2' date: 2022-12-05 19:58:43 categories: Rust tags: [Rust, 数据库, Redis]
本篇接上一篇《mini-redis项目-1-简介》,讲解mini-redis存储层的实现;
源代码:
系列文章:
<br/>
<!--more-->在 mini-redis 中为了简单起见,数据是直接通过 HashMap 中存储了;
对于实际生产环境中使用的数据库来说,通常会使用:
<br/>
数据操作接口被定义在 traits.rs 文件的 KvStore Trait 中;
src/storage/traits.rs
pub trait KvStore {
fn get(&self, key: &str) -> Option<Bytes>;
/// Set the value associated with a key along with an optional expiration
/// Duration.
///
/// If a value is already associated with the key, it is removed.
fn set(&self, key: String, value: Bytes, expire: Option<Duration>);
/// Returns a `Receiver` for the requested channel.
///
/// The returned `Receiver` is used to receive values broadcast by `PUBLISH`
/// commands.
fn subscribe(&self, key: String) -> broadcast::Receiver<Bytes>;
/// Publish a message to the channel. Returns the number of subscribers
/// listening on the channel.
fn publish(&self, key: &str, value: Bytes) -> usize;
}
主要是 get、set、pub、sub 四个接口,非常的简单;
其中 set 可以设置键的过期时间;
<br/>
Store 结构体中存放实际的数据,定义如下:
src/storage/store.rs
#[derive(Debug)]
pub(crate) struct Store {
pub(crate) entries: HashMap<String, Entry>,
pub(crate) pub_sub: HashMap<String, broadcast::Sender<Bytes>>,
pub(crate) expirations: BTreeMap<(Instant, u64), String>,
pub(crate) next_id: u64,
pub(crate) shutdown: bool,
}
#[derive(Debug)]
pub(crate) struct Entry {
pub(crate) id: u64,
pub(crate) data: Bytes,
pub(crate) expires_at: Option<Instant>,
}
impl Store {
pub(crate) fn new() -> Store {
Store {
entries: HashMap::new(),
pub_sub: HashMap::new(),
expirations: BTreeMap::new(),
next_id: 0,
shutdown: false,
}
}
pub(crate) fn next_expiration(&self) -> Option<Instant> {
self.expirations.keys().next().map(|expire| expire.0)
}
}
Store 的逻辑也非常简单:
这里的 Store 层用了
pub(crate)
即整个 crate 可见,是一个简化操作;个人感觉应当将 Store 进行封装,这样如果实际上底层的存储结构变化,对于上次的操作是无感知的!
<br/>
Db 封装了上面的 Store,并实现了 KvStore Trait,对外提供了一组数据操作接口,下面来看具体实现;
Db实现:
src/storage/db.rs
#[derive(Debug, Clone)]
pub(crate) struct Db {
shared: Arc<SharedDb>,
}
impl Db {
pub(crate) fn new() -> Db {
let shared = Arc::new(SharedDb::new());
tokio::spawn(Db::purge_expired_tasks(shared.clone()));
Db { shared }
}
async fn purge_expired_tasks(shared: Arc<SharedDb>) {
while !shared.is_shutdown() {
if let Some(when) = shared.purge_expired_keys() {
tokio::select! {
_ = time::sleep_until(when) => {}
_ = shared.background_task.notified() => {}
}
} else {
shared.background_task.notified().await;
}
}
info!("Purge background task shut down")
}
fn shutdown_purge_task(&self) {
let mut store = self.shared.store.lock().unwrap();
store.shutdown = true;
drop(store);
self.shared.background_task.notify_one();
}
}
由于可能会存在多个异步任务同时获取数据库的场景(例如:键过期、Channel订阅都会用到数据库);
因此需要将 Store 进行封装至 Arc 中,同时封装了 SharedDb(下文会讲);
Db 中定义了 new 方法,用于初始化数据库,并且启动 purge_expired_tasks
任务;
当服务器未关闭时:
purge_expired_tasks
会调用 shared.purge_expired_keys
清除最近的过期的键,并获取下一个最新过期的键的时间并等待;shared.background_task
再次通知了当前异步任务,此时大概率是加入了更早的过期键,需要重新调用 shared.purge_expired_keys
获取最新的最近的过期时间;shared.background_task.notified().await
;这样,就完成了过期键的处理!
同时,Db 定义了 shutdown_purge_task
方法,提供给 DbDropGuard,用于在服务器停止后清理资源以及其他任务:
drop(store)
就显式释放了数据库资源,但是这个方法不能由在 Drop Db 的时候调用,因为还需要停止其他异步任务;self.shared.background_task.notify_one()
通知异步过期键任务,此时由于 shutdown 变为了 true,异步任务会直接退出!<br/>
前面提到了可能会存在多个异步任务同时获取数据库,因此将 Db 封装至了SharedDb 中:
src/storage/db.rs
#[derive(Debug)]
struct SharedDb {
store: Mutex<Store>,
background_task: Notify,
}
impl SharedDb {
fn new() -> Self {
SharedDb {
store: Mutex::new(Store::new()),
background_task: Notify::new(),
}
}
fn purge_expired_keys(&self) -> Option<Instant> {
let mut store = self.store.lock().unwrap();
if store.shutdown {
return None;
}
let store = &mut *store;
let now = Instant::now();
while let Some((&(when, id), key)) = store.expirations.iter().next() {
if when > now {
return Some(when);
}
store.entries.remove(key);
store.expirations.remove(&(when, id));
}
None
}
fn is_shutdown(&self) -> bool {
self.store.lock().unwrap().shutdown
}
}
SharedDb 中的内容非常简单:
store: Mutex<Store>
:互斥锁保护底层数据不会并发访问;background_task: Notify
:和前面提到的异步任务交互;同时定义了几个方法,new、is_shutdown 很简单不说了;
和上文对应,purge_expired_keys
方法由 Db 中的异步任务调用,在服务器没有关闭时,通过 while let Some((&(when, id), key)) = store.expirations.iter().next()
按照过期时间获取过期键:
<br/>
下面是 Db 实现的 KvStore Trait:
src/storage/db.rs
impl KvStore for Db {
fn get(&self, key: &str) -> Option<Bytes> {
let store = self.shared.store.lock().unwrap();
store.entries.get(key).map(|entry| entry.data.clone())
}
fn set(&self, key: String, value: Bytes, expire: Option<Duration>) {
let mut store = self.shared.store.lock().unwrap();
let id = store.next_id;
store.next_id += 1;
let mut notify = false;
let expires_at = expire.map(|duration| {
let when = Instant::now() + duration;
notify = store
.next_expiration()
.map(|expiration| expiration > when)
.unwrap_or(true);
store.expirations.insert((when, id), key.clone());
when
});
let prev = store.entries.insert(
key,
Entry {
id,
data: value,
expires_at,
},
);
if let Some(prev) = prev {
if let Some(when) = prev.expires_at {
// clear expiration
store.expirations.remove(&(when, prev.id));
}
}
drop(store);
if notify {
self.shared.background_task.notify_one();
}
}
fn subscribe(&self, key: String) -> broadcast::Receiver<Bytes> {
use std::collections::hash_map::Entry;
let mut store = self.shared.store.lock().unwrap();
match store.pub_sub.entry(key) {
Entry::Occupied(e) => e.get().subscribe(),
Entry::Vacant(e) => {
let (tx, rx) = broadcast::channel(1024);
e.insert(tx);
rx
}
}
}
fn publish(&self, key: &str, value: Bytes) -> usize {
debug!("publish: (key={}, len(value)={})", key, value.len());
let state = self.shared.store.lock().unwrap();
state
.pub_sub
.get(key)
.map(|tx| tx.send(value).unwrap_or(0))
.unwrap_or(0)
}
}
下面分别来看;
<br/>
Get 方法比较简单:
fn get(&self, key: &str) -> Option<Bytes> {
// Acquire the lock, get the entry and clone the value.
//
// Because data is stored using `Bytes`, a clone here is a shallow
// clone. Data is not copied.
let store = self.shared.store.lock().unwrap();
store.entries.get(key).map(|entry| entry.data.clone())
}
首先获取存储的锁,然后通过 store 中的 HashMap entries 直接获取键值即可;
<font color="#f00">**这里注意:我们使用的是 `bytes::Bytes`,因此这里 clone 的实际上是一个胖指针,而非将整个数据copy了一份!**</font>
<br/>
Set方法如下:
fn set(&self, key: String, value: Bytes, expire: Option<Duration>) {
let mut store = self.shared.store.lock().unwrap();
let id = store.next_id;
store.next_id += 1;
let mut notify = false;
let expires_at = expire.map(|duration| {
let when = Instant::now() + duration;
notify = store
.next_expiration()
.map(|expiration| expiration > when)
.unwrap_or(true);
store.expirations.insert((when, id), key.clone());
when
});
let prev = store.entries.insert(
key,
Entry {
id,
data: value,
expires_at,
},
);
if let Some(prev) = prev {
if let Some(when) = prev.expires_at {
store.expirations.remove(&(when, prev.id));
}
}
drop(store);
if notify {
self.shared.background_task.notify_one();
}
}
我们为当前 set 的 key:
purge_expired_keys
方法,重新刷新过期任务!随后,将键加入 store.entries 中,如果是过期键还会加入 store.expirations 中;
然后,需要判断如果 set 的 key 之前在数据库中存在了,则需要将前一个过期任务删除;
最后,显式 drop 掉我们获取的 store 的锁,这样最后通知刷新过期键异步任务后,这个异步任务不用再等待这个函数释放锁了~
<br/>
subscribe方法如下:
fn subscribe(&self, key: String) -> broadcast::Receiver<Bytes> {
use std::collections::hash_map::Entry;
let mut store = self.shared.store.lock().unwrap();
match store.pub_sub.entry(key) {
Entry::Occupied(e) => e.get().subscribe(),
Entry::Vacant(e) => {
let (tx, rx) = broadcast::channel(1024);
e.insert(tx);
rx
}
}
}
subscribe 方法的逻辑也非常简单:
subscribe
了,则直接返回一个新的 Receiver 即可;subscribe
则创建一个新的channel,并将 Receiver 加入 store.pub_sub 存储后返回Receiver即可;<br/>
有了 Receiver,则 publish 方法的实现更加简单:
fn publish(&self, key: &str, value: Bytes) -> usize {
let state = self.shared.store.lock().unwrap();
state
.pub_sub
.get(key)
.map(|tx| tx.send(value).unwrap_or(0))
.unwrap_or(0)
}
直接获取 state.pub_sub
中 key 对应的 channel 并调用 send 方法即可;
需要注意的是:如果不存在这个 key,或者发生失败,则返回 0 大小的消息,表示未发送成功;
<br/>
最后就是 DbDropGuard 的实现了,DbDropGuard 实际上是 Db 的一个封装;
当通过 DbDropGuard 获取 Db 时,实际上是一个 Db 的 clone,而 Db 内部是一个 Arc 计数智能指针;
因此,当所有使用 Db 都被回收,仍然需要 DbDropGuard 被 drop 后才会对数据库进行回收(DbDropGuard 中持有第一个 Db):
src/storage/db.rs
/// A wrapper around a `Db` instance. This exists to allow orderly cleanup
/// of the `Db` by signalling the background purge task to shut down when
/// this struct is dropped.
#[derive(Debug)]
pub(crate) struct DbDropGuard {
/// The `Db` instance that will be shut down when this `DbHolder` struct
/// is dropped.
db: Db,
}
impl DbDropGuard {
/// Create a new `DbHolder`, wrapping a `Db` instance. When this is dropped
/// the `Db`'s purge task will be shut down.
pub(crate) fn new() -> DbDropGuard {
DbDropGuard { db: Db::new() }
}
/// Get the shared database. Internally, this is an `Arc`,
/// so a clone only increments the ref count.
pub(crate) fn db(&self) -> Db {
self.db.clone()
}
}
impl Drop for DbDropGuard {
fn drop(&mut self) {
// Signal the 'Db' instance to shut down the task that purges expired keys
self.db.shutdown_purge_task();
}
}
同时,为 DbDropGuard 实现了 Drop Trait,当 DbDropGuard 被回收后,会调用前文中提到的 db 中的 shutdown_purge_task
方法进行资源回收;
<br/>
本小节完成了对 mini-redis 存储层的实现,对外封装了 DbDropGuard;
在使用时只需创建 DbDropGuard 对象,然后通过其内部封装的智能指针 Db 操作底层的 Store 即可!
实际上,真实的存储结构远远要比这里的 HashMap 要复杂,但是本文也能起到一定抛砖引玉的作用!
<br/>
源代码:
系列文章:
<br/>