refactor for tests

This commit is contained in:
mpostma 2021-03-23 11:00:50 +01:00
parent 1f51fc8baf
commit 3cc3637e2d
No known key found for this signature in database
GPG Key ID: CBC8A7C1D7A28C3A
22 changed files with 1676 additions and 1474 deletions

View File

@ -1,657 +0,0 @@
use std::collections::HashMap;
use std::fs::{create_dir_all, File};
use std::future::Future;
use std::path::{Path, PathBuf};
use std::sync::Arc;
use async_stream::stream;
use chrono::{DateTime, Utc};
use futures::pin_mut;
use futures::stream::StreamExt;
use heed::{CompactionOption, EnvOpenOptions};
use log::debug;
use serde::{Deserialize, Serialize};
use thiserror::Error;
use tokio::fs::remove_dir_all;
use tokio::sync::{mpsc, oneshot, RwLock};
use tokio::task::spawn_blocking;
use uuid::Uuid;
use super::update_handler::UpdateHandler;
use super::{get_arc_ownership_blocking, IndexSettings};
use crate::index::UpdateResult as UResult;
use crate::index::{Document, Index, SearchQuery, SearchResult, Settings};
use crate::index_controller::{
updates::{Failed, Processed, Processing},
UpdateMeta,
};
use crate::option::IndexerOpts;
pub type Result<T> = std::result::Result<T, IndexError>;
type AsyncMap<K, V> = Arc<RwLock<HashMap<K, V>>>;
type UpdateResult = std::result::Result<Processed<UpdateMeta, UResult>, Failed<UpdateMeta, String>>;
#[derive(Debug, Serialize, Deserialize, Clone)]
#[serde(rename_all = "camelCase")]
pub struct IndexMeta {
created_at: DateTime<Utc>,
updated_at: DateTime<Utc>,
primary_key: Option<String>,
}
impl IndexMeta {
fn new(index: &Index) -> Result<Self> {
let txn = index.read_txn()?;
Self::new_txn(index, &txn)
}
fn new_txn(index: &Index, txn: &heed::RoTxn) -> Result<Self> {
let created_at = index.created_at(&txn)?;
let updated_at = index.updated_at(&txn)?;
let primary_key = index.primary_key(&txn)?.map(String::from);
Ok(Self {
primary_key,
updated_at,
created_at,
})
}
}
enum IndexMsg {
CreateIndex {
uuid: Uuid,
primary_key: Option<String>,
ret: oneshot::Sender<Result<IndexMeta>>,
},
Update {
meta: Processing<UpdateMeta>,
data: std::fs::File,
ret: oneshot::Sender<Result<UpdateResult>>,
},
Search {
uuid: Uuid,
query: SearchQuery,
ret: oneshot::Sender<anyhow::Result<SearchResult>>,
},
Settings {
uuid: Uuid,
ret: oneshot::Sender<Result<Settings>>,
},
Documents {
uuid: Uuid,
attributes_to_retrieve: Option<Vec<String>>,
offset: usize,
limit: usize,
ret: oneshot::Sender<Result<Vec<Document>>>,
},
Document {
uuid: Uuid,
attributes_to_retrieve: Option<Vec<String>>,
doc_id: String,
ret: oneshot::Sender<Result<Document>>,
},
Delete {
uuid: Uuid,
ret: oneshot::Sender<Result<()>>,
},
GetMeta {
uuid: Uuid,
ret: oneshot::Sender<Result<IndexMeta>>,
},
UpdateIndex {
uuid: Uuid,
index_settings: IndexSettings,
ret: oneshot::Sender<Result<IndexMeta>>,
},
Snapshot {
uuid: Uuid,
path: PathBuf,
ret: oneshot::Sender<Result<()>>,
},
}
struct IndexActor<S> {
read_receiver: Option<mpsc::Receiver<IndexMsg>>,
write_receiver: Option<mpsc::Receiver<IndexMsg>>,
update_handler: Arc<UpdateHandler>,
store: S,
}
#[derive(Error, Debug)]
pub enum IndexError {
#[error("error with index: {0}")]
Error(#[from] anyhow::Error),
#[error("index already exists")]
IndexAlreadyExists,
#[error("Index doesn't exists")]
UnexistingIndex,
#[error("Heed error: {0}")]
HeedError(#[from] heed::Error),
#[error("Existing primary key")]
ExistingPrimaryKey,
}
#[async_trait::async_trait]
trait IndexStore {
async fn create(&self, uuid: Uuid, primary_key: Option<String>) -> Result<Index>;
async fn get(&self, uuid: Uuid) -> Result<Option<Index>>;
async fn delete(&self, uuid: Uuid) -> Result<Option<Index>>;
}
impl<S: IndexStore + Sync + Send> IndexActor<S> {
fn new(
read_receiver: mpsc::Receiver<IndexMsg>,
write_receiver: mpsc::Receiver<IndexMsg>,
store: S,
) -> Result<Self> {
let options = IndexerOpts::default();
let update_handler = UpdateHandler::new(&options).map_err(IndexError::Error)?;
let update_handler = Arc::new(update_handler);
let read_receiver = Some(read_receiver);
let write_receiver = Some(write_receiver);
Ok(Self {
read_receiver,
write_receiver,
store,
update_handler,
})
}
/// `run` poll the write_receiver and read_receiver concurrently, but while messages send
/// through the read channel are processed concurrently, the messages sent through the write
/// channel are processed one at a time.
async fn run(mut self) {
let mut read_receiver = self
.read_receiver
.take()
.expect("Index Actor must have a inbox at this point.");
let read_stream = stream! {
loop {
match read_receiver.recv().await {
Some(msg) => yield msg,
None => break,
}
}
};
let mut write_receiver = self
.write_receiver
.take()
.expect("Index Actor must have a inbox at this point.");
let write_stream = stream! {
loop {
match write_receiver.recv().await {
Some(msg) => yield msg,
None => break,
}
}
};
pin_mut!(write_stream);
pin_mut!(read_stream);
let fut1 = read_stream.for_each_concurrent(Some(10), |msg| self.handle_message(msg));
let fut2 = write_stream.for_each_concurrent(Some(1), |msg| self.handle_message(msg));
let fut1: Box<dyn Future<Output = ()> + Unpin + Send> = Box::new(fut1);
let fut2: Box<dyn Future<Output = ()> + Unpin + Send> = Box::new(fut2);
tokio::join!(fut1, fut2);
}
async fn handle_message(&self, msg: IndexMsg) {
use IndexMsg::*;
match msg {
CreateIndex {
uuid,
primary_key,
ret,
} => {
let _ = ret.send(self.handle_create_index(uuid, primary_key).await);
}
Update { ret, meta, data } => {
let _ = ret.send(self.handle_update(meta, data).await);
}
Search { ret, query, uuid } => {
let _ = ret.send(self.handle_search(uuid, query).await);
}
Settings { ret, uuid } => {
let _ = ret.send(self.handle_settings(uuid).await);
}
Documents {
ret,
uuid,
attributes_to_retrieve,
offset,
limit,
} => {
let _ = ret.send(
self.handle_fetch_documents(uuid, offset, limit, attributes_to_retrieve)
.await,
);
}
Document {
uuid,
attributes_to_retrieve,
doc_id,
ret,
} => {
let _ = ret.send(
self.handle_fetch_document(uuid, doc_id, attributes_to_retrieve)
.await,
);
}
Delete { uuid, ret } => {
let _ = ret.send(self.handle_delete(uuid).await);
}
GetMeta { uuid, ret } => {
let _ = ret.send(self.handle_get_meta(uuid).await);
}
UpdateIndex {
uuid,
index_settings,
ret,
} => {
let _ = ret.send(self.handle_update_index(uuid, index_settings).await);
}
Snapshot { uuid, path, ret } => {
let _ = ret.send(self.handle_snapshot(uuid, path).await);
}
}
}
async fn handle_search(&self, uuid: Uuid, query: SearchQuery) -> anyhow::Result<SearchResult> {
let index = self
.store
.get(uuid)
.await?
.ok_or(IndexError::UnexistingIndex)?;
spawn_blocking(move || index.perform_search(query)).await?
}
async fn handle_create_index(
&self,
uuid: Uuid,
primary_key: Option<String>,
) -> Result<IndexMeta> {
let index = self.store.create(uuid, primary_key).await?;
let meta = spawn_blocking(move || IndexMeta::new(&index))
.await
.map_err(|e| IndexError::Error(e.into()))??;
Ok(meta)
}
async fn handle_update(
&self,
meta: Processing<UpdateMeta>,
data: File,
) -> Result<UpdateResult> {
log::info!("Processing update {}", meta.id());
let uuid = meta.index_uuid();
let update_handler = self.update_handler.clone();
let index = match self.store.get(*uuid).await? {
Some(index) => index,
None => self.store.create(*uuid, None).await?,
};
spawn_blocking(move || update_handler.handle_update(meta, data, index))
.await
.map_err(|e| IndexError::Error(e.into()))
}
async fn handle_settings(&self, uuid: Uuid) -> Result<Settings> {
let index = self
.store
.get(uuid)
.await?
.ok_or(IndexError::UnexistingIndex)?;
spawn_blocking(move || index.settings().map_err(IndexError::Error))
.await
.map_err(|e| IndexError::Error(e.into()))?
}
async fn handle_fetch_documents(
&self,
uuid: Uuid,
offset: usize,
limit: usize,
attributes_to_retrieve: Option<Vec<String>>,
) -> Result<Vec<Document>> {
let index = self
.store
.get(uuid)
.await?
.ok_or(IndexError::UnexistingIndex)?;
spawn_blocking(move || {
index
.retrieve_documents(offset, limit, attributes_to_retrieve)
.map_err(IndexError::Error)
})
.await
.map_err(|e| IndexError::Error(e.into()))?
}
async fn handle_fetch_document(
&self,
uuid: Uuid,
doc_id: String,
attributes_to_retrieve: Option<Vec<String>>,
) -> Result<Document> {
let index = self
.store
.get(uuid)
.await?
.ok_or(IndexError::UnexistingIndex)?;
spawn_blocking(move || {
index
.retrieve_document(doc_id, attributes_to_retrieve)
.map_err(IndexError::Error)
})
.await
.map_err(|e| IndexError::Error(e.into()))?
}
async fn handle_delete(&self, uuid: Uuid) -> Result<()> {
let index = self.store.delete(uuid).await?;
if let Some(index) = index {
tokio::task::spawn(async move {
let index = index.0;
let store = get_arc_ownership_blocking(index).await;
spawn_blocking(move || {
store.prepare_for_closing().wait();
debug!("Index closed");
});
});
}
Ok(())
}
async fn handle_get_meta(&self, uuid: Uuid) -> Result<IndexMeta> {
match self.store.get(uuid).await? {
Some(index) => {
let meta = spawn_blocking(move || IndexMeta::new(&index))
.await
.map_err(|e| IndexError::Error(e.into()))??;
Ok(meta)
}
None => Err(IndexError::UnexistingIndex),
}
}
async fn handle_update_index(
&self,
uuid: Uuid,
index_settings: IndexSettings,
) -> Result<IndexMeta> {
let index = self
.store
.get(uuid)
.await?
.ok_or(IndexError::UnexistingIndex)?;
spawn_blocking(move || match index_settings.primary_key {
Some(ref primary_key) => {
let mut txn = index.write_txn()?;
if index.primary_key(&txn)?.is_some() {
return Err(IndexError::ExistingPrimaryKey);
}
index.put_primary_key(&mut txn, primary_key)?;
let meta = IndexMeta::new_txn(&index, &txn)?;
txn.commit()?;
Ok(meta)
}
None => {
let meta = IndexMeta::new(&index)?;
Ok(meta)
}
})
.await
.map_err(|e| IndexError::Error(e.into()))?
}
async fn handle_snapshot(&self, uuid: Uuid, mut path: PathBuf) -> Result<()> {
use tokio::fs::create_dir_all;
path.push("indexes");
create_dir_all(&path)
.await
.map_err(|e| IndexError::Error(e.into()))?;
if let Some(index) = self.store.get(uuid).await? {
let mut index_path = path.join(format!("index-{}", uuid));
create_dir_all(&index_path)
.await
.map_err(|e| IndexError::Error(e.into()))?;
index_path.push("data.mdb");
spawn_blocking(move || -> anyhow::Result<()> {
// Get write txn to wait for ongoing write transaction before snapshot.
let _txn = index.write_txn()?;
index
.env
.copy_to_path(index_path, CompactionOption::Enabled)?;
Ok(())
})
.await
.map_err(|e| IndexError::Error(e.into()))?
.map_err(|e| IndexError::Error(e.into()))?;
}
Ok(())
}
}
#[derive(Clone)]
pub struct IndexActorHandle {
read_sender: mpsc::Sender<IndexMsg>,
write_sender: mpsc::Sender<IndexMsg>,
}
impl IndexActorHandle {
pub fn new(path: impl AsRef<Path>, index_size: usize) -> anyhow::Result<Self> {
let (read_sender, read_receiver) = mpsc::channel(100);
let (write_sender, write_receiver) = mpsc::channel(100);
let store = HeedIndexStore::new(path, index_size);
let actor = IndexActor::new(read_receiver, write_receiver, store)?;
tokio::task::spawn(actor.run());
Ok(Self {
read_sender,
write_sender,
})
}
pub async fn create_index(&self, uuid: Uuid, primary_key: Option<String>) -> Result<IndexMeta> {
let (ret, receiver) = oneshot::channel();
let msg = IndexMsg::CreateIndex {
ret,
uuid,
primary_key,
};
let _ = self.read_sender.send(msg).await;
receiver.await.expect("IndexActor has been killed")
}
pub async fn update(
&self,
meta: Processing<UpdateMeta>,
data: std::fs::File,
) -> anyhow::Result<UpdateResult> {
let (ret, receiver) = oneshot::channel();
let msg = IndexMsg::Update { ret, meta, data };
let _ = self.write_sender.send(msg).await;
Ok(receiver.await.expect("IndexActor has been killed")?)
}
pub async fn search(&self, uuid: Uuid, query: SearchQuery) -> Result<SearchResult> {
let (ret, receiver) = oneshot::channel();
let msg = IndexMsg::Search { uuid, query, ret };
let _ = self.read_sender.send(msg).await;
Ok(receiver.await.expect("IndexActor has been killed")?)
}
pub async fn settings(&self, uuid: Uuid) -> Result<Settings> {
let (ret, receiver) = oneshot::channel();
let msg = IndexMsg::Settings { uuid, ret };
let _ = self.read_sender.send(msg).await;
Ok(receiver.await.expect("IndexActor has been killed")?)
}
pub async fn documents(
&self,
uuid: Uuid,
offset: usize,
limit: usize,
attributes_to_retrieve: Option<Vec<String>>,
) -> Result<Vec<Document>> {
let (ret, receiver) = oneshot::channel();
let msg = IndexMsg::Documents {
uuid,
ret,
offset,
attributes_to_retrieve,
limit,
};
let _ = self.read_sender.send(msg).await;
Ok(receiver.await.expect("IndexActor has been killed")?)
}
pub async fn document(
&self,
uuid: Uuid,
doc_id: String,
attributes_to_retrieve: Option<Vec<String>>,
) -> Result<Document> {
let (ret, receiver) = oneshot::channel();
let msg = IndexMsg::Document {
uuid,
ret,
doc_id,
attributes_to_retrieve,
};
let _ = self.read_sender.send(msg).await;
Ok(receiver.await.expect("IndexActor has been killed")?)
}
pub async fn delete(&self, uuid: Uuid) -> Result<()> {
let (ret, receiver) = oneshot::channel();
let msg = IndexMsg::Delete { uuid, ret };
let _ = self.write_sender.send(msg).await;
Ok(receiver.await.expect("IndexActor has been killed")?)
}
pub async fn get_index_meta(&self, uuid: Uuid) -> Result<IndexMeta> {
let (ret, receiver) = oneshot::channel();
let msg = IndexMsg::GetMeta { uuid, ret };
let _ = self.read_sender.send(msg).await;
Ok(receiver.await.expect("IndexActor has been killed")?)
}
pub async fn update_index(
&self,
uuid: Uuid,
index_settings: IndexSettings,
) -> Result<IndexMeta> {
let (ret, receiver) = oneshot::channel();
let msg = IndexMsg::UpdateIndex {
uuid,
index_settings,
ret,
};
let _ = self.read_sender.send(msg).await;
Ok(receiver.await.expect("IndexActor has been killed")?)
}
pub async fn snapshot(&self, uuid: Uuid, path: PathBuf) -> Result<()> {
let (ret, receiver) = oneshot::channel();
let msg = IndexMsg::Snapshot { uuid, path, ret };
let _ = self.read_sender.send(msg).await;
Ok(receiver.await.expect("IndexActor has been killed")?)
}
}
struct HeedIndexStore {
index_store: AsyncMap<Uuid, Index>,
path: PathBuf,
index_size: usize,
}
impl HeedIndexStore {
fn new(path: impl AsRef<Path>, index_size: usize) -> Self {
let path = path.as_ref().join("indexes/");
let index_store = Arc::new(RwLock::new(HashMap::new()));
Self {
index_store,
path,
index_size,
}
}
}
#[async_trait::async_trait]
impl IndexStore for HeedIndexStore {
async fn create(&self, uuid: Uuid, primary_key: Option<String>) -> Result<Index> {
let path = self.path.join(format!("index-{}", uuid));
if path.exists() {
return Err(IndexError::IndexAlreadyExists);
}
let index_size = self.index_size;
let index = spawn_blocking(move || -> Result<Index> {
let index = open_index(&path, index_size)?;
if let Some(primary_key) = primary_key {
let mut txn = index.write_txn()?;
index.put_primary_key(&mut txn, &primary_key)?;
txn.commit()?;
}
Ok(index)
})
.await
.map_err(|e| IndexError::Error(e.into()))??;
self.index_store.write().await.insert(uuid, index.clone());
Ok(index)
}
async fn get(&self, uuid: Uuid) -> Result<Option<Index>> {
let guard = self.index_store.read().await;
match guard.get(&uuid) {
Some(index) => Ok(Some(index.clone())),
None => {
// drop the guard here so we can perform the write after without deadlocking;
drop(guard);
let path = self.path.join(format!("index-{}", uuid));
if !path.exists() {
return Ok(None);
}
let index_size = self.index_size;
let index = spawn_blocking(move || open_index(path, index_size))
.await
.map_err(|e| IndexError::Error(e.into()))??;
self.index_store.write().await.insert(uuid, index.clone());
Ok(Some(index))
}
}
}
async fn delete(&self, uuid: Uuid) -> Result<Option<Index>> {
let db_path = self.path.join(format!("index-{}", uuid));
remove_dir_all(db_path)
.await
.map_err(|e| IndexError::Error(e.into()))?;
let index = self.index_store.write().await.remove(&uuid);
Ok(index)
}
}
fn open_index(path: impl AsRef<Path>, size: usize) -> Result<Index> {
create_dir_all(&path).map_err(|e| IndexError::Error(e.into()))?;
let mut options = EnvOpenOptions::new();
options.map_size(size);
let index = milli::Index::new(options, &path).map_err(IndexError::Error)?;
Ok(Index(Arc::new(index)))
}

View File

@ -0,0 +1,331 @@
use std::fs::File;
use std::future::Future;
use std::path::PathBuf;
use std::sync::Arc;
use async_stream::stream;
use futures::pin_mut;
use futures::stream::StreamExt;
use heed::CompactionOption;
use log::debug;
use tokio::sync::mpsc;
use tokio::task::spawn_blocking;
use uuid::Uuid;
use crate::index::{Document, SearchQuery, SearchResult, Settings};
use crate::index_controller::update_handler::UpdateHandler;
use crate::index_controller::{updates::Processing, UpdateMeta, get_arc_ownership_blocking};
use crate::option::IndexerOpts;
use super::{IndexSettings, Result, IndexMsg, IndexStore, IndexError, UpdateResult, IndexMeta};
pub struct IndexActor<S> {
read_receiver: Option<mpsc::Receiver<IndexMsg>>,
write_receiver: Option<mpsc::Receiver<IndexMsg>>,
update_handler: Arc<UpdateHandler>,
store: S,
}
impl<S: IndexStore + Sync + Send> IndexActor<S> {
pub fn new(
read_receiver: mpsc::Receiver<IndexMsg>,
write_receiver: mpsc::Receiver<IndexMsg>,
store: S,
) -> Result<Self> {
let options = IndexerOpts::default();
let update_handler = UpdateHandler::new(&options).map_err(IndexError::Error)?;
let update_handler = Arc::new(update_handler);
let read_receiver = Some(read_receiver);
let write_receiver = Some(write_receiver);
Ok(Self {
read_receiver,
write_receiver,
store,
update_handler,
})
}
/// `run` poll the write_receiver and read_receiver concurrently, but while messages send
/// through the read channel are processed concurrently, the messages sent through the write
/// channel are processed one at a time.
pub async fn run(mut self) {
let mut read_receiver = self
.read_receiver
.take()
.expect("Index Actor must have a inbox at this point.");
let read_stream = stream! {
loop {
match read_receiver.recv().await {
Some(msg) => yield msg,
None => break,
}
}
};
let mut write_receiver = self
.write_receiver
.take()
.expect("Index Actor must have a inbox at this point.");
let write_stream = stream! {
loop {
match write_receiver.recv().await {
Some(msg) => yield msg,
None => break,
}
}
};
pin_mut!(write_stream);
pin_mut!(read_stream);
let fut1 = read_stream.for_each_concurrent(Some(10), |msg| self.handle_message(msg));
let fut2 = write_stream.for_each_concurrent(Some(1), |msg| self.handle_message(msg));
let fut1: Box<dyn Future<Output = ()> + Unpin + Send> = Box::new(fut1);
let fut2: Box<dyn Future<Output = ()> + Unpin + Send> = Box::new(fut2);
tokio::join!(fut1, fut2);
}
async fn handle_message(&self, msg: IndexMsg) {
use IndexMsg::*;
match msg {
CreateIndex {
uuid,
primary_key,
ret,
} => {
let _ = ret.send(self.handle_create_index(uuid, primary_key).await);
}
Update { ret, meta, data } => {
let _ = ret.send(self.handle_update(meta, data).await);
}
Search { ret, query, uuid } => {
let _ = ret.send(self.handle_search(uuid, query).await);
}
Settings { ret, uuid } => {
let _ = ret.send(self.handle_settings(uuid).await);
}
Documents {
ret,
uuid,
attributes_to_retrieve,
offset,
limit,
} => {
let _ = ret.send(
self.handle_fetch_documents(uuid, offset, limit, attributes_to_retrieve)
.await,
);
}
Document {
uuid,
attributes_to_retrieve,
doc_id,
ret,
} => {
let _ = ret.send(
self.handle_fetch_document(uuid, doc_id, attributes_to_retrieve)
.await,
);
}
Delete { uuid, ret } => {
let _ = ret.send(self.handle_delete(uuid).await);
}
GetMeta { uuid, ret } => {
let _ = ret.send(self.handle_get_meta(uuid).await);
}
UpdateIndex {
uuid,
index_settings,
ret,
} => {
let _ = ret.send(self.handle_update_index(uuid, index_settings).await);
}
Snapshot { uuid, path, ret } => {
let _ = ret.send(self.handle_snapshot(uuid, path).await);
}
}
}
async fn handle_search(&self, uuid: Uuid, query: SearchQuery) -> anyhow::Result<SearchResult> {
let index = self
.store
.get(uuid)
.await?
.ok_or(IndexError::UnexistingIndex)?;
spawn_blocking(move || index.perform_search(query)).await?
}
async fn handle_create_index(
&self,
uuid: Uuid,
primary_key: Option<String>,
) -> Result<IndexMeta> {
let index = self.store.create(uuid, primary_key).await?;
let meta = spawn_blocking(move || IndexMeta::new(&index))
.await
.map_err(|e| IndexError::Error(e.into()))??;
Ok(meta)
}
async fn handle_update(
&self,
meta: Processing<UpdateMeta>,
data: File,
) -> Result<UpdateResult> {
debug!("Processing update {}", meta.id());
let uuid = meta.index_uuid();
let update_handler = self.update_handler.clone();
let index = match self.store.get(*uuid).await? {
Some(index) => index,
None => self.store.create(*uuid, None).await?,
};
spawn_blocking(move || update_handler.handle_update(meta, data, index))
.await
.map_err(|e| IndexError::Error(e.into()))
}
async fn handle_settings(&self, uuid: Uuid) -> Result<Settings> {
let index = self
.store
.get(uuid)
.await?
.ok_or(IndexError::UnexistingIndex)?;
spawn_blocking(move || index.settings().map_err(IndexError::Error))
.await
.map_err(|e| IndexError::Error(e.into()))?
}
async fn handle_fetch_documents(
&self,
uuid: Uuid,
offset: usize,
limit: usize,
attributes_to_retrieve: Option<Vec<String>>,
) -> Result<Vec<Document>> {
let index = self
.store
.get(uuid)
.await?
.ok_or(IndexError::UnexistingIndex)?;
spawn_blocking(move || {
index
.retrieve_documents(offset, limit, attributes_to_retrieve)
.map_err(IndexError::Error)
})
.await
.map_err(|e| IndexError::Error(e.into()))?
}
async fn handle_fetch_document(
&self,
uuid: Uuid,
doc_id: String,
attributes_to_retrieve: Option<Vec<String>>,
) -> Result<Document> {
let index = self
.store
.get(uuid)
.await?
.ok_or(IndexError::UnexistingIndex)?;
spawn_blocking(move || {
index
.retrieve_document(doc_id, attributes_to_retrieve)
.map_err(IndexError::Error)
})
.await
.map_err(|e| IndexError::Error(e.into()))?
}
async fn handle_delete(&self, uuid: Uuid) -> Result<()> {
let index = self.store.delete(uuid).await?;
if let Some(index) = index {
tokio::task::spawn(async move {
let index = index.0;
let store = get_arc_ownership_blocking(index).await;
spawn_blocking(move || {
store.prepare_for_closing().wait();
debug!("Index closed");
});
});
}
Ok(())
}
async fn handle_get_meta(&self, uuid: Uuid) -> Result<IndexMeta> {
match self.store.get(uuid).await? {
Some(index) => {
let meta = spawn_blocking(move || IndexMeta::new(&index))
.await
.map_err(|e| IndexError::Error(e.into()))??;
Ok(meta)
}
None => Err(IndexError::UnexistingIndex),
}
}
async fn handle_update_index(
&self,
uuid: Uuid,
index_settings: IndexSettings,
) -> Result<IndexMeta> {
let index = self
.store
.get(uuid)
.await?
.ok_or(IndexError::UnexistingIndex)?;
spawn_blocking(move || match index_settings.primary_key {
Some(ref primary_key) => {
let mut txn = index.write_txn()?;
if index.primary_key(&txn)?.is_some() {
return Err(IndexError::ExistingPrimaryKey);
}
index.put_primary_key(&mut txn, primary_key)?;
let meta = IndexMeta::new_txn(&index, &txn)?;
txn.commit()?;
Ok(meta)
}
None => {
let meta = IndexMeta::new(&index)?;
Ok(meta)
}
})
.await
.map_err(|e| IndexError::Error(e.into()))?
}
async fn handle_snapshot(&self, uuid: Uuid, mut path: PathBuf) -> Result<()> {
use tokio::fs::create_dir_all;
path.push("indexes");
create_dir_all(&path)
.await
.map_err(|e| IndexError::Error(e.into()))?;
if let Some(index) = self.store.get(uuid).await? {
let mut index_path = path.join(format!("index-{}", uuid));
create_dir_all(&index_path)
.await
.map_err(|e| IndexError::Error(e.into()))?;
index_path.push("data.mdb");
spawn_blocking(move || -> anyhow::Result<()> {
// Get write txn to wait for ongoing write transaction before snapshot.
let _txn = index.write_txn()?;
index
.env
.copy_to_path(index_path, CompactionOption::Enabled)?;
Ok(())
})
.await
.map_err(|e| IndexError::Error(e.into()))?
.map_err(|e| IndexError::Error(e.into()))?;
}
Ok(())
}
}

View File

@ -0,0 +1,141 @@
use std::path::{PathBuf, Path};
use tokio::sync::{mpsc, oneshot};
use uuid::Uuid;
use crate::index::{Document, SearchQuery, SearchResult, Settings};
use crate::index_controller::IndexSettings;
use crate::index_controller::{updates::Processing, UpdateMeta};
use super::{IndexActorHandle, IndexMsg, IndexMeta, UpdateResult, Result, IndexActor, MapIndexStore};
#[derive(Clone)]
pub struct IndexActorHandleImpl {
read_sender: mpsc::Sender<IndexMsg>,
write_sender: mpsc::Sender<IndexMsg>,
}
#[async_trait::async_trait]
impl IndexActorHandle for IndexActorHandleImpl {
async fn create_index(&self, uuid: Uuid, primary_key: Option<String>) -> Result<IndexMeta> {
let (ret, receiver) = oneshot::channel();
let msg = IndexMsg::CreateIndex {
ret,
uuid,
primary_key,
};
let _ = self.read_sender.send(msg).await;
receiver.await.expect("IndexActor has been killed")
}
async fn update(
&self,
meta: Processing<UpdateMeta>,
data: std::fs::File,
) -> anyhow::Result<UpdateResult> {
let (ret, receiver) = oneshot::channel();
let msg = IndexMsg::Update { ret, meta, data };
let _ = self.read_sender.send(msg).await;
Ok(receiver.await.expect("IndexActor has been killed")?)
}
async fn search(&self, uuid: Uuid, query: SearchQuery) -> Result<SearchResult> {
let (ret, receiver) = oneshot::channel();
let msg = IndexMsg::Search { uuid, query, ret };
let _ = self.read_sender.send(msg).await;
Ok(receiver.await.expect("IndexActor has been killed")?)
}
async fn settings(&self, uuid: Uuid) -> Result<Settings> {
let (ret, receiver) = oneshot::channel();
let msg = IndexMsg::Settings { uuid, ret };
let _ = self.read_sender.send(msg).await;
Ok(receiver.await.expect("IndexActor has been killed")?)
}
async fn documents(
&self,
uuid: Uuid,
offset: usize,
limit: usize,
attributes_to_retrieve: Option<Vec<String>>,
) -> Result<Vec<Document>> {
let (ret, receiver) = oneshot::channel();
let msg = IndexMsg::Documents {
uuid,
ret,
offset,
attributes_to_retrieve,
limit,
};
let _ = self.read_sender.send(msg).await;
Ok(receiver.await.expect("IndexActor has been killed")?)
}
async fn document(
&self,
uuid: Uuid,
doc_id: String,
attributes_to_retrieve: Option<Vec<String>>,
) -> Result<Document> {
let (ret, receiver) = oneshot::channel();
let msg = IndexMsg::Document {
uuid,
ret,
doc_id,
attributes_to_retrieve,
};
let _ = self.read_sender.send(msg).await;
Ok(receiver.await.expect("IndexActor has been killed")?)
}
async fn delete(&self, uuid: Uuid) -> Result<()> {
let (ret, receiver) = oneshot::channel();
let msg = IndexMsg::Delete { uuid, ret };
let _ = self.read_sender.send(msg).await;
Ok(receiver.await.expect("IndexActor has been killed")?)
}
async fn get_index_meta(&self, uuid: Uuid) -> Result<IndexMeta> {
let (ret, receiver) = oneshot::channel();
let msg = IndexMsg::GetMeta { uuid, ret };
let _ = self.read_sender.send(msg).await;
Ok(receiver.await.expect("IndexActor has been killed")?)
}
async fn update_index(
&self,
uuid: Uuid,
index_settings: IndexSettings,
) -> Result<IndexMeta> {
let (ret, receiver) = oneshot::channel();
let msg = IndexMsg::UpdateIndex {
uuid,
index_settings,
ret,
};
let _ = self.read_sender.send(msg).await;
Ok(receiver.await.expect("IndexActor has been killed")?)
}
async fn snapshot(&self, uuid: Uuid, path: PathBuf) -> Result<()> {
let (ret, receiver) = oneshot::channel();
let msg = IndexMsg::Snapshot { uuid, path, ret };
let _ = self.read_sender.send(msg).await;
Ok(receiver.await.expect("IndexActor has been killed")?)
}
}
impl IndexActorHandleImpl {
pub fn new(path: impl AsRef<Path>, index_size: usize) -> anyhow::Result<Self> {
let (read_sender, read_receiver) = mpsc::channel(100);
let (write_sender, write_receiver) = mpsc::channel(100);
let store = MapIndexStore::new(path, index_size);
let actor = IndexActor::new(read_receiver, write_receiver, store)?;
tokio::task::spawn(actor.run());
Ok(Self {
read_sender,
write_sender,
})
}
}

View File

@ -0,0 +1,64 @@
use std::path::PathBuf;
use tokio::sync::oneshot;
use uuid::Uuid;
use crate::index::{Document, SearchQuery, SearchResult, Settings};
use crate::index_controller::{
updates::Processing,
UpdateMeta,
};
use super::{IndexSettings, IndexMeta, UpdateResult, Result};
pub enum IndexMsg {
CreateIndex {
uuid: Uuid,
primary_key: Option<String>,
ret: oneshot::Sender<Result<IndexMeta>>,
},
Update {
meta: Processing<UpdateMeta>,
data: std::fs::File,
ret: oneshot::Sender<Result<UpdateResult>>,
},
Search {
uuid: Uuid,
query: SearchQuery,
ret: oneshot::Sender<anyhow::Result<SearchResult>>,
},
Settings {
uuid: Uuid,
ret: oneshot::Sender<Result<Settings>>,
},
Documents {
uuid: Uuid,
attributes_to_retrieve: Option<Vec<String>>,
offset: usize,
limit: usize,
ret: oneshot::Sender<Result<Vec<Document>>>,
},
Document {
uuid: Uuid,
attributes_to_retrieve: Option<Vec<String>>,
doc_id: String,
ret: oneshot::Sender<Result<Document>>,
},
Delete {
uuid: Uuid,
ret: oneshot::Sender<Result<()>>,
},
GetMeta {
uuid: Uuid,
ret: oneshot::Sender<Result<IndexMeta>>,
},
UpdateIndex {
uuid: Uuid,
index_settings: IndexSettings,
ret: oneshot::Sender<Result<IndexMeta>>,
},
Snapshot {
uuid: Uuid,
path: PathBuf,
ret: oneshot::Sender<Result<()>>,
},
}

View File

@ -0,0 +1,103 @@
mod actor;
mod handle_impl;
mod message;
mod store;
use std::path::PathBuf;
use chrono::{DateTime, Utc};
use serde::{Deserialize, Serialize};
use thiserror::Error;
use uuid::Uuid;
use super::IndexSettings;
use crate::index::UpdateResult as UResult;
use crate::index::{Document, Index, SearchQuery, SearchResult, Settings};
use crate::index_controller::{
updates::{Failed, Processed, Processing},
UpdateMeta,
};
use message::IndexMsg;
use store::{IndexStore, MapIndexStore};
use actor::IndexActor;
pub use handle_impl::IndexActorHandleImpl;
pub type Result<T> = std::result::Result<T, IndexError>;
type UpdateResult = std::result::Result<Processed<UpdateMeta, UResult>, Failed<UpdateMeta, String>>;
#[derive(Debug, Serialize, Deserialize, Clone)]
#[serde(rename_all = "camelCase")]
pub struct IndexMeta {
created_at: DateTime<Utc>,
updated_at: DateTime<Utc>,
primary_key: Option<String>,
}
impl IndexMeta {
fn new(index: &Index) -> Result<Self> {
let txn = index.read_txn()?;
Self::new_txn(index, &txn)
}
fn new_txn(index: &Index, txn: &heed::RoTxn) -> Result<Self> {
let created_at = index.created_at(&txn)?;
let updated_at = index.updated_at(&txn)?;
let primary_key = index.primary_key(&txn)?.map(String::from);
Ok(Self {
primary_key,
updated_at,
created_at,
})
}
}
#[derive(Error, Debug)]
pub enum IndexError {
#[error("error with index: {0}")]
Error(#[from] anyhow::Error),
#[error("index already exists")]
IndexAlreadyExists,
#[error("Index doesn't exists")]
UnexistingIndex,
#[error("Heed error: {0}")]
HeedError(#[from] heed::Error),
#[error("Existing primary key")]
ExistingPrimaryKey,
}
#[async_trait::async_trait]
pub trait IndexActorHandle: Sync + Send + Clone {
async fn create_index(&self, uuid: Uuid, primary_key: Option<String>) -> Result<IndexMeta>;
async fn update(
&self,
meta: Processing<UpdateMeta>,
data: std::fs::File,
) -> anyhow::Result<UpdateResult>;
async fn search(&self, uuid: Uuid, query: SearchQuery) -> Result<SearchResult>;
async fn settings(&self, uuid: Uuid) -> Result<Settings>;
async fn documents(
&self,
uuid: Uuid,
offset: usize,
limit: usize,
attributes_to_retrieve: Option<Vec<String>>,
) -> Result<Vec<Document>>;
async fn document(
&self,
uuid: Uuid,
doc_id: String,
attributes_to_retrieve: Option<Vec<String>>,
) -> Result<Document>;
async fn delete(&self, uuid: Uuid) -> Result<()>;
async fn get_index_meta(&self, uuid: Uuid) -> Result<IndexMeta>;
async fn update_index(
&self,
uuid: Uuid,
index_settings: IndexSettings,
) -> Result<IndexMeta>;
async fn snapshot(&self, uuid: Uuid, path: PathBuf) -> Result<()>;
}

View File

@ -0,0 +1,105 @@
use std::path::{PathBuf, Path};
use std::sync::Arc;
use std::collections::HashMap;
use uuid::Uuid;
use tokio::sync::RwLock;
use tokio::task::spawn_blocking;
use tokio::fs;
use heed::EnvOpenOptions;
use super::{IndexError, Result};
use crate::index::Index;
type AsyncMap<K, V> = Arc<RwLock<HashMap<K, V>>>;
#[async_trait::async_trait]
pub trait IndexStore {
async fn create(&self, uuid: Uuid, primary_key: Option<String>) -> Result<Index>;
async fn get(&self, uuid: Uuid) -> Result<Option<Index>>;
async fn delete(&self, uuid: Uuid) -> Result<Option<Index>>;
}
pub struct MapIndexStore {
index_store: AsyncMap<Uuid, Index>,
path: PathBuf,
index_size: usize,
}
impl MapIndexStore {
pub fn new(path: impl AsRef<Path>, index_size: usize) -> Self {
let path = path.as_ref().join("indexes/");
let index_store = Arc::new(RwLock::new(HashMap::new()));
Self {
index_store,
path,
index_size,
}
}
}
#[async_trait::async_trait]
impl IndexStore for MapIndexStore {
async fn create(&self, uuid: Uuid, primary_key: Option<String>) -> Result<Index> {
let path = self.path.join(format!("index-{}", uuid));
if path.exists() {
return Err(IndexError::IndexAlreadyExists);
}
let index_size = self.index_size;
let index = spawn_blocking(move || -> Result<Index> {
let index = open_index(&path, index_size)?;
if let Some(primary_key) = primary_key {
let mut txn = index.write_txn()?;
index.put_primary_key(&mut txn, &primary_key)?;
txn.commit()?;
}
Ok(index)
})
.await
.map_err(|e| IndexError::Error(e.into()))??;
self.index_store.write().await.insert(uuid, index.clone());
Ok(index)
}
async fn get(&self, uuid: Uuid) -> Result<Option<Index>> {
let guard = self.index_store.read().await;
match guard.get(&uuid) {
Some(index) => Ok(Some(index.clone())),
None => {
// drop the guard here so we can perform the write after without deadlocking;
drop(guard);
let path = self.path.join(format!("index-{}", uuid));
if !path.exists() {
return Ok(None);
}
let index_size = self.index_size;
let index = spawn_blocking(move || open_index(path, index_size))
.await
.map_err(|e| IndexError::Error(e.into()))??;
self.index_store.write().await.insert(uuid, index.clone());
Ok(Some(index))
}
}
}
async fn delete(&self, uuid: Uuid) -> Result<Option<Index>> {
let db_path = self.path.join(format!("index-{}", uuid));
fs::remove_dir_all(db_path)
.await
.map_err(|e| IndexError::Error(e.into()))?;
let index = self.index_store.write().await.remove(&uuid);
Ok(index)
}
}
fn open_index(path: impl AsRef<Path>, size: usize) -> Result<Index> {
std::fs::create_dir_all(&path).map_err(|e| IndexError::Error(e.into()))?;
let mut options = EnvOpenOptions::new();
options.map_size(size);
let index = milli::Index::new(options, &path).map_err(IndexError::Error)?;
Ok(Index(Arc::new(index)))
}

View File

@ -2,7 +2,6 @@ mod index_actor;
mod snapshot; mod snapshot;
mod update_actor; mod update_actor;
mod update_handler; mod update_handler;
mod update_store;
mod updates; mod updates;
mod uuid_resolver; mod uuid_resolver;
@ -22,6 +21,9 @@ use crate::index::{Document, SearchQuery, SearchResult};
use crate::index::{Facets, Settings, UpdateResult}; use crate::index::{Facets, Settings, UpdateResult};
use crate::option::Opt; use crate::option::Opt;
use crate::helpers::compression; use crate::helpers::compression;
use index_actor::IndexActorHandle;
use update_actor::UpdateActorHandle;
use uuid_resolver::UuidResolverHandle;
use snapshot::SnapshotService; use snapshot::SnapshotService;
pub use updates::{Failed, Processed, Processing}; pub use updates::{Failed, Processed, Processing};
@ -58,9 +60,9 @@ pub struct IndexSettings {
} }
pub struct IndexController { pub struct IndexController {
uuid_resolver: uuid_resolver::UuidResolverHandle, uuid_resolver: uuid_resolver::UuidResolverHandleImpl,
index_handle: index_actor::IndexActorHandle, index_handle: index_actor::IndexActorHandleImpl,
update_handle: update_actor::UpdateActorHandle<Bytes>, update_handle: update_actor::UpdateActorHandleImpl<Bytes>,
} }
impl IndexController { impl IndexController {
@ -72,9 +74,9 @@ impl IndexController {
compression::from_tar_gz(path, &options.db_path)?; compression::from_tar_gz(path, &options.db_path)?;
} }
let uuid_resolver = uuid_resolver::UuidResolverHandle::new(&path)?; let uuid_resolver = uuid_resolver::UuidResolverHandleImpl::new(&path)?;
let index_handle = index_actor::IndexActorHandle::new(&path, index_size)?; let index_handle = index_actor::IndexActorHandleImpl::new(&path, index_size)?;
let update_handle = update_actor::UpdateActorHandle::new( let update_handle = update_actor::UpdateActorHandleImpl::new(
index_handle.clone(), index_handle.clone(),
&path, &path,
update_store_size, update_store_size,
@ -82,7 +84,6 @@ impl IndexController {
if options.schedule_snapshot { if options.schedule_snapshot {
let snapshot_service = SnapshotService::new( let snapshot_service = SnapshotService::new(
index_handle.clone(),
uuid_resolver.clone(), uuid_resolver.clone(),
update_handle.clone(), update_handle.clone(),
Duration::from_secs(options.snapshot_interval_sec), Duration::from_secs(options.snapshot_interval_sec),

View File

@ -8,29 +8,29 @@ use tokio::task::spawn_blocking;
use tokio::time::sleep; use tokio::time::sleep;
use crate::helpers::compression; use crate::helpers::compression;
use super::index_actor::IndexActorHandle;
use super::update_actor::UpdateActorHandle; use super::update_actor::UpdateActorHandle;
use super::uuid_resolver::UuidResolverHandle; use super::uuid_resolver::UuidResolverHandle;
#[allow(dead_code)] #[allow(dead_code)]
pub struct SnapshotService<B> { pub struct SnapshotService<U, R> {
index_handle: IndexActorHandle, uuid_resolver_handle: R,
uuid_resolver_handle: UuidResolverHandle, update_handle: U,
update_handle: UpdateActorHandle<B>,
snapshot_period: Duration, snapshot_period: Duration,
snapshot_path: PathBuf, snapshot_path: PathBuf,
} }
impl<B> SnapshotService<B> { impl<U, R> SnapshotService<U, R>
where
U: UpdateActorHandle,
R: UuidResolverHandle
{
pub fn new( pub fn new(
index_handle: IndexActorHandle, uuid_resolver_handle: R,
uuid_resolver_handle: UuidResolverHandle, update_handle: U,
update_handle: UpdateActorHandle<B>,
snapshot_period: Duration, snapshot_period: Duration,
snapshot_path: PathBuf, snapshot_path: PathBuf,
) -> Self { ) -> Self {
Self { Self {
index_handle,
uuid_resolver_handle, uuid_resolver_handle,
update_handle, update_handle,
snapshot_period, snapshot_period,

View File

@ -1,455 +0,0 @@
use std::collections::{hash_map::Entry, HashMap};
use std::fs;
use std::io::SeekFrom;
use std::path::{Path, PathBuf};
use std::sync::Arc;
use super::index_actor::IndexActorHandle;
use log::info;
use oxidized_json_checker::JsonChecker;
use thiserror::Error;
use tokio::fs::OpenOptions;
use tokio::io::{AsyncSeekExt, AsyncWriteExt};
use tokio::sync::{mpsc, oneshot, RwLock};
use uuid::Uuid;
use super::get_arc_ownership_blocking;
use crate::index::UpdateResult;
use crate::index_controller::{UpdateMeta, UpdateStatus};
pub type Result<T> = std::result::Result<T, UpdateError>;
type UpdateStore = super::update_store::UpdateStore<UpdateMeta, UpdateResult, String>;
type PayloadData<D> = std::result::Result<D, Box<dyn std::error::Error + Sync + Send + 'static>>;
#[derive(Debug, Error)]
pub enum UpdateError {
#[error("error with update: {0}")]
Error(Box<dyn std::error::Error + Sync + Send + 'static>),
#[error("Index {0} doesn't exist.")]
UnexistingIndex(Uuid),
#[error("Update {0} doesn't exist.")]
UnexistingUpdate(u64),
}
enum UpdateMsg<D> {
Update {
uuid: Uuid,
meta: UpdateMeta,
data: mpsc::Receiver<PayloadData<D>>,
ret: oneshot::Sender<Result<UpdateStatus>>,
},
ListUpdates {
uuid: Uuid,
ret: oneshot::Sender<Result<Vec<UpdateStatus>>>,
},
GetUpdate {
uuid: Uuid,
ret: oneshot::Sender<Result<UpdateStatus>>,
id: u64,
},
Delete {
uuid: Uuid,
ret: oneshot::Sender<Result<()>>,
},
Create {
uuid: Uuid,
ret: oneshot::Sender<Result<()>>,
},
Snapshot {
uuid: Uuid,
path: PathBuf,
ret: oneshot::Sender<Result<()>>,
},
}
struct UpdateActor<D, S> {
path: PathBuf,
store: S,
inbox: mpsc::Receiver<UpdateMsg<D>>,
index_handle: IndexActorHandle,
}
#[async_trait::async_trait]
trait UpdateStoreStore {
async fn get_or_create(&self, uuid: Uuid) -> Result<Arc<UpdateStore>>;
async fn delete(&self, uuid: Uuid) -> Result<Option<Arc<UpdateStore>>>;
async fn get(&self, uuid: Uuid) -> Result<Option<Arc<UpdateStore>>>;
}
impl<D, S> UpdateActor<D, S>
where
D: AsRef<[u8]> + Sized + 'static,
S: UpdateStoreStore,
{
fn new(
store: S,
inbox: mpsc::Receiver<UpdateMsg<D>>,
path: impl AsRef<Path>,
index_handle: IndexActorHandle,
) -> anyhow::Result<Self> {
let path = path.as_ref().to_owned();
fs::create_dir_all(path.join("update_files"))?;
assert!(path.exists());
Ok(Self {
store,
inbox,
path,
index_handle,
})
}
async fn run(mut self) {
use UpdateMsg::*;
info!("Started update actor.");
loop {
match self.inbox.recv().await {
Some(Update {
uuid,
meta,
data,
ret,
}) => {
let _ = ret.send(self.handle_update(uuid, meta, data).await);
}
Some(ListUpdates { uuid, ret }) => {
let _ = ret.send(self.handle_list_updates(uuid).await);
}
Some(GetUpdate { uuid, ret, id }) => {
let _ = ret.send(self.handle_get_update(uuid, id).await);
}
Some(Delete { uuid, ret }) => {
let _ = ret.send(self.handle_delete(uuid).await);
}
Some(Create { uuid, ret }) => {
let _ = ret.send(self.handle_create(uuid).await);
}
Some(Snapshot { uuid, path, ret }) => {
let _ = ret.send(self.handle_snapshot(uuid, path).await);
}
None => break,
}
}
}
async fn handle_update(
&self,
uuid: Uuid,
meta: UpdateMeta,
mut payload: mpsc::Receiver<PayloadData<D>>,
) -> Result<UpdateStatus> {
let update_store = self.store.get_or_create(uuid).await?;
let update_file_id = uuid::Uuid::new_v4();
let path = self
.path
.join(format!("update_files/update_{}", update_file_id));
let mut file = OpenOptions::new()
.read(true)
.write(true)
.create(true)
.open(&path)
.await
.map_err(|e| UpdateError::Error(Box::new(e)))?;
while let Some(bytes) = payload.recv().await {
match bytes {
Ok(bytes) => {
file.write_all(bytes.as_ref())
.await
.map_err(|e| UpdateError::Error(Box::new(e)))?;
}
Err(e) => {
return Err(UpdateError::Error(e));
}
}
}
file.flush()
.await
.map_err(|e| UpdateError::Error(Box::new(e)))?;
file.seek(SeekFrom::Start(0))
.await
.map_err(|e| UpdateError::Error(Box::new(e)))?;
let mut file = file.into_std().await;
tokio::task::spawn_blocking(move || {
use std::io::{copy, sink, BufReader, Seek};
// If the payload is empty, ignore the check.
if file
.metadata()
.map_err(|e| UpdateError::Error(Box::new(e)))?
.len()
> 0
{
// Check that the json payload is valid:
let reader = BufReader::new(&mut file);
let mut checker = JsonChecker::new(reader);
if copy(&mut checker, &mut sink()).is_err() || checker.finish().is_err() {
// The json file is invalid, we use Serde to get a nice error message:
file.seek(SeekFrom::Start(0))
.map_err(|e| UpdateError::Error(Box::new(e)))?;
let _: serde_json::Value = serde_json::from_reader(file)
.map_err(|e| UpdateError::Error(Box::new(e)))?;
}
}
// The payload is valid, we can register it to the update store.
update_store
.register_update(meta, path, uuid)
.map(UpdateStatus::Pending)
.map_err(|e| UpdateError::Error(Box::new(e)))
})
.await
.map_err(|e| UpdateError::Error(Box::new(e)))?
}
async fn handle_list_updates(&self, uuid: Uuid) -> Result<Vec<UpdateStatus>> {
let update_store = self.store.get(uuid).await?;
tokio::task::spawn_blocking(move || {
let result = update_store
.ok_or(UpdateError::UnexistingIndex(uuid))?
.list()
.map_err(|e| UpdateError::Error(e.into()))?;
Ok(result)
})
.await
.map_err(|e| UpdateError::Error(Box::new(e)))?
}
async fn handle_get_update(&self, uuid: Uuid, id: u64) -> Result<UpdateStatus> {
let store = self
.store
.get(uuid)
.await?
.ok_or(UpdateError::UnexistingIndex(uuid))?;
let result = store
.meta(id)
.map_err(|e| UpdateError::Error(Box::new(e)))?
.ok_or(UpdateError::UnexistingUpdate(id))?;
Ok(result)
}
async fn handle_delete(&self, uuid: Uuid) -> Result<()> {
let store = self.store.delete(uuid).await?;
if let Some(store) = store {
tokio::task::spawn(async move {
let store = get_arc_ownership_blocking(store).await;
tokio::task::spawn_blocking(move || {
store.prepare_for_closing().wait();
info!("Update store {} was closed.", uuid);
});
});
}
Ok(())
}
async fn handle_create(&self, uuid: Uuid) -> Result<()> {
let _ = self.store.get_or_create(uuid).await?;
Ok(())
}
async fn handle_snapshot(&self, uuid: Uuid, path: PathBuf) -> Result<()> {
let index_handle = self.index_handle.clone();
if let Some(update_store) = self.store.get(uuid).await? {
tokio::task::spawn_blocking(move || -> anyhow::Result<()> {
// acquire write lock to prevent further writes during snapshot
// the update lock must be acquired BEFORE the write lock to prevent dead lock
let _lock = update_store.update_lock.lock();
let mut txn = update_store.env.write_txn()?;
// create db snapshot
update_store.snapshot(&mut txn, &path, uuid)?;
futures::executor::block_on(
async move { index_handle.snapshot(uuid, path).await },
)?;
Ok(())
})
.await
.map_err(|e| UpdateError::Error(e.into()))?
.map_err(|e| UpdateError::Error(e.into()))?;
}
Ok(())
}
}
#[derive(Clone)]
pub struct UpdateActorHandle<D> {
sender: mpsc::Sender<UpdateMsg<D>>,
}
impl<D> UpdateActorHandle<D>
where
D: AsRef<[u8]> + Sized + 'static + Sync + Send,
{
pub fn new(
index_handle: IndexActorHandle,
path: impl AsRef<Path>,
update_store_size: usize,
) -> anyhow::Result<Self> {
let path = path.as_ref().to_owned().join("updates");
let (sender, receiver) = mpsc::channel(100);
let store = MapUpdateStoreStore::new(index_handle.clone(), &path, update_store_size);
let actor = UpdateActor::new(store, receiver, path, index_handle)?;
tokio::task::spawn(actor.run());
Ok(Self { sender })
}
pub async fn update(
&self,
meta: UpdateMeta,
data: mpsc::Receiver<PayloadData<D>>,
uuid: Uuid,
) -> Result<UpdateStatus> {
let (ret, receiver) = oneshot::channel();
let msg = UpdateMsg::Update {
uuid,
data,
meta,
ret,
};
let _ = self.sender.send(msg).await;
receiver.await.expect("update actor killed.")
}
}
impl<D> UpdateActorHandle<D> {
pub async fn get_all_updates_status(&self, uuid: Uuid) -> Result<Vec<UpdateStatus>> {
let (ret, receiver) = oneshot::channel();
let msg = UpdateMsg::ListUpdates { uuid, ret };
let _ = self.sender.send(msg).await;
receiver.await.expect("update actor killed.")
}
pub async fn update_status(&self, uuid: Uuid, id: u64) -> Result<UpdateStatus> {
let (ret, receiver) = oneshot::channel();
let msg = UpdateMsg::GetUpdate { uuid, id, ret };
let _ = self.sender.send(msg).await;
receiver.await.expect("update actor killed.")
}
pub async fn delete(&self, uuid: Uuid) -> Result<()> {
let (ret, receiver) = oneshot::channel();
let msg = UpdateMsg::Delete { uuid, ret };
let _ = self.sender.send(msg).await;
receiver.await.expect("update actor killed.")
}
pub async fn create(&self, uuid: Uuid) -> Result<()> {
let (ret, receiver) = oneshot::channel();
let msg = UpdateMsg::Create { uuid, ret };
let _ = self.sender.send(msg).await;
receiver.await.expect("update actor killed.")
}
pub async fn snapshot(&self, uuid: Uuid, path: PathBuf) -> Result<()> {
let (ret, receiver) = oneshot::channel();
let msg = UpdateMsg::Snapshot { uuid, path, ret };
let _ = self.sender.send(msg).await;
receiver.await.expect("update actor killed.")
}
}
struct MapUpdateStoreStore {
db: Arc<RwLock<HashMap<Uuid, Arc<UpdateStore>>>>,
index_handle: IndexActorHandle,
path: PathBuf,
update_store_size: usize,
}
impl MapUpdateStoreStore {
fn new(
index_handle: IndexActorHandle,
path: impl AsRef<Path>,
update_store_size: usize,
) -> Self {
let db = Arc::new(RwLock::new(HashMap::new()));
let path = path.as_ref().to_owned();
Self {
db,
index_handle,
path,
update_store_size,
}
}
}
#[async_trait::async_trait]
impl UpdateStoreStore for MapUpdateStoreStore {
async fn get_or_create(&self, uuid: Uuid) -> Result<Arc<UpdateStore>> {
match self.db.write().await.entry(uuid) {
Entry::Vacant(e) => {
let mut options = heed::EnvOpenOptions::new();
let update_store_size = self.update_store_size;
options.map_size(update_store_size);
let path = self.path.clone().join(format!("updates-{}", e.key()));
fs::create_dir_all(&path).unwrap();
let index_handle = self.index_handle.clone();
let store = UpdateStore::open(options, &path, move |meta, file| {
futures::executor::block_on(index_handle.update(meta, file))
})
.map_err(|e| UpdateError::Error(e.into()))?;
let store = e.insert(store);
Ok(store.clone())
}
Entry::Occupied(e) => Ok(e.get().clone()),
}
}
async fn get(&self, uuid: Uuid) -> Result<Option<Arc<UpdateStore>>> {
let guard = self.db.read().await;
match guard.get(&uuid) {
Some(uuid) => Ok(Some(uuid.clone())),
None => {
// The index is not found in the found in the loaded indexes, so we attempt to load
// it from disk. We need to acquire a write lock **before** attempting to open the
// index, because someone could be trying to open it at the same time as us.
drop(guard);
let path = self.path.clone().join(format!("updates-{}", uuid));
if path.exists() {
let mut guard = self.db.write().await;
match guard.entry(uuid) {
Entry::Vacant(entry) => {
// We can safely load the index
let index_handle = self.index_handle.clone();
let mut options = heed::EnvOpenOptions::new();
let update_store_size = self.update_store_size;
options.map_size(update_store_size);
let store = UpdateStore::open(options, &path, move |meta, file| {
futures::executor::block_on(index_handle.update(meta, file))
})
.map_err(|e| UpdateError::Error(e.into()))?;
let store = entry.insert(store);
Ok(Some(store.clone()))
}
Entry::Occupied(entry) => {
// The index was loaded while we attempted to to iter
Ok(Some(entry.get().clone()))
}
}
} else {
Ok(None)
}
}
}
}
async fn delete(&self, uuid: Uuid) -> Result<Option<Arc<UpdateStore>>> {
let store = self.db.write().await.remove(&uuid);
let path = self.path.clone().join(format!("updates-{}", uuid));
if store.is_some() || path.exists() {
fs::remove_dir_all(path).unwrap();
}
Ok(store)
}
}

View File

@ -0,0 +1,226 @@
use std::io::SeekFrom;
use std::path::{Path, PathBuf};
use log::info;
use tokio::sync::mpsc;
use uuid::Uuid;
use oxidized_json_checker::JsonChecker;
use tokio::fs;
use tokio::io::{AsyncSeekExt, AsyncWriteExt};
use super::{PayloadData, UpdateError, UpdateMsg, UpdateStoreStore, Result};
use crate::index_controller::index_actor::IndexActorHandle;
use crate::index_controller::{UpdateMeta, UpdateStatus, get_arc_ownership_blocking};
pub struct UpdateActor<D, S, I> {
path: PathBuf,
store: S,
inbox: mpsc::Receiver<UpdateMsg<D>>,
index_handle: I,
}
impl<D, S, I> UpdateActor<D, S, I>
where
D: AsRef<[u8]> + Sized + 'static,
S: UpdateStoreStore,
I: IndexActorHandle + 'static,
{
pub fn new(
store: S,
inbox: mpsc::Receiver<UpdateMsg<D>>,
path: impl AsRef<Path>,
index_handle: I,
) -> anyhow::Result<Self> {
let path = path.as_ref().to_owned();
std::fs::create_dir_all(path.join("update_files"))?;
assert!(path.exists());
Ok(Self {
store,
inbox,
path,
index_handle,
})
}
pub async fn run(mut self) {
use UpdateMsg::*;
info!("Started update actor.");
loop {
match self.inbox.recv().await {
Some(Update {
uuid,
meta,
data,
ret,
}) => {
let _ = ret.send(self.handle_update(uuid, meta, data).await);
}
Some(ListUpdates { uuid, ret }) => {
let _ = ret.send(self.handle_list_updates(uuid).await);
}
Some(GetUpdate { uuid, ret, id }) => {
let _ = ret.send(self.handle_get_update(uuid, id).await);
}
Some(Delete { uuid, ret }) => {
let _ = ret.send(self.handle_delete(uuid).await);
}
Some(Create { uuid, ret }) => {
let _ = ret.send(self.handle_create(uuid).await);
}
Some(Snapshot { uuid, path, ret }) => {
let _ = ret.send(self.handle_snapshot(uuid, path).await);
}
None => break,
}
}
}
async fn handle_update(
&self,
uuid: Uuid,
meta: UpdateMeta,
mut payload: mpsc::Receiver<PayloadData<D>>,
) -> Result<UpdateStatus> {
let update_store = self.store.get_or_create(uuid).await?;
let update_file_id = uuid::Uuid::new_v4();
let path = self
.path
.join(format!("update_files/update_{}", update_file_id));
let mut file = fs::OpenOptions::new()
.read(true)
.write(true)
.create(true)
.open(&path)
.await
.map_err(|e| UpdateError::Error(Box::new(e)))?;
while let Some(bytes) = payload.recv().await {
match bytes {
Ok(bytes) => {
file.write_all(bytes.as_ref())
.await
.map_err(|e| UpdateError::Error(Box::new(e)))?;
}
Err(e) => {
return Err(UpdateError::Error(e));
}
}
}
file.flush()
.await
.map_err(|e| UpdateError::Error(Box::new(e)))?;
file.seek(SeekFrom::Start(0))
.await
.map_err(|e| UpdateError::Error(Box::new(e)))?;
let mut file = file.into_std().await;
tokio::task::spawn_blocking(move || {
use std::io::{copy, sink, BufReader, Seek};
// If the payload is empty, ignore the check.
if file
.metadata()
.map_err(|e| UpdateError::Error(Box::new(e)))?
.len()
> 0
{
// Check that the json payload is valid:
let reader = BufReader::new(&mut file);
let mut checker = JsonChecker::new(reader);
if copy(&mut checker, &mut sink()).is_err() || checker.finish().is_err() {
// The json file is invalid, we use Serde to get a nice error message:
file.seek(SeekFrom::Start(0))
.map_err(|e| UpdateError::Error(Box::new(e)))?;
let _: serde_json::Value = serde_json::from_reader(file)
.map_err(|e| UpdateError::Error(Box::new(e)))?;
}
}
// The payload is valid, we can register it to the update store.
update_store
.register_update(meta, path, uuid)
.map(UpdateStatus::Pending)
.map_err(|e| UpdateError::Error(Box::new(e)))
})
.await
.map_err(|e| UpdateError::Error(Box::new(e)))?
}
async fn handle_list_updates(&self, uuid: Uuid) -> Result<Vec<UpdateStatus>> {
let update_store = self.store.get(uuid).await?;
tokio::task::spawn_blocking(move || {
let result = update_store
.ok_or(UpdateError::UnexistingIndex(uuid))?
.list()
.map_err(|e| UpdateError::Error(e.into()))?;
Ok(result)
})
.await
.map_err(|e| UpdateError::Error(Box::new(e)))?
}
async fn handle_get_update(&self, uuid: Uuid, id: u64) -> Result<UpdateStatus> {
let store = self
.store
.get(uuid)
.await?
.ok_or(UpdateError::UnexistingIndex(uuid))?;
let result = store
.meta(id)
.map_err(|e| UpdateError::Error(Box::new(e)))?
.ok_or(UpdateError::UnexistingUpdate(id))?;
Ok(result)
}
async fn handle_delete(&self, uuid: Uuid) -> Result<()> {
let store = self.store.delete(uuid).await?;
if let Some(store) = store {
tokio::task::spawn(async move {
let store = get_arc_ownership_blocking(store).await;
tokio::task::spawn_blocking(move || {
store.prepare_for_closing().wait();
info!("Update store {} was closed.", uuid);
});
});
}
Ok(())
}
async fn handle_create(&self, uuid: Uuid) -> Result<()> {
let _ = self.store.get_or_create(uuid).await?;
Ok(())
}
async fn handle_snapshot(&self, uuid: Uuid, path: PathBuf) -> Result<()> {
let index_handle = self.index_handle.clone();
if let Some(update_store) = self.store.get(uuid).await? {
tokio::task::spawn_blocking(move || -> anyhow::Result<()> {
// acquire write lock to prevent further writes during snapshot
// the update lock must be acquired BEFORE the write lock to prevent dead lock
let _lock = update_store.update_lock.lock();
let mut txn = update_store.env.write_txn()?;
// create db snapshot
update_store.snapshot(&mut txn, &path, uuid)?;
futures::executor::block_on(
async move { index_handle.snapshot(uuid, path).await },
)?;
Ok(())
})
.await
.map_err(|e| UpdateError::Error(e.into()))?
.map_err(|e| UpdateError::Error(e.into()))?;
}
Ok(())
}
}

View File

@ -0,0 +1,96 @@
use std::path::{Path, PathBuf};
use tokio::sync::{mpsc, oneshot};
use uuid::Uuid;
use super::{
MapUpdateStoreStore, PayloadData, Result, UpdateActor, UpdateActorHandle, UpdateMeta,
UpdateMsg, UpdateStatus,
};
use crate::index_controller::IndexActorHandle;
#[derive(Clone)]
pub struct UpdateActorHandleImpl<D> {
sender: mpsc::Sender<UpdateMsg<D>>,
}
impl<D> UpdateActorHandleImpl<D>
where
D: AsRef<[u8]> + Sized + 'static + Sync + Send,
{
pub fn new<I>(
index_handle: I,
path: impl AsRef<Path>,
update_store_size: usize,
) -> anyhow::Result<Self>
where
I: IndexActorHandle + 'static,
{
let path = path.as_ref().to_owned().join("updates");
let (sender, receiver) = mpsc::channel(100);
let store = MapUpdateStoreStore::new(index_handle.clone(), &path, update_store_size);
let actor = UpdateActor::new(store, receiver, path, index_handle)?;
tokio::task::spawn(actor.run());
Ok(Self { sender })
}
}
#[async_trait::async_trait]
impl<D> UpdateActorHandle for UpdateActorHandleImpl<D>
where
D: AsRef<[u8]> + Sized + 'static + Sync + Send,
{
type Data = D;
async fn update(
&self,
meta: UpdateMeta,
data: mpsc::Receiver<PayloadData<Self::Data>>,
uuid: Uuid,
) -> Result<UpdateStatus> {
let (ret, receiver) = oneshot::channel();
let msg = UpdateMsg::Update {
uuid,
data,
meta,
ret,
};
let _ = self.sender.send(msg).await;
receiver.await.expect("update actor killed.")
}
async fn get_all_updates_status(&self, uuid: Uuid) -> Result<Vec<UpdateStatus>> {
let (ret, receiver) = oneshot::channel();
let msg = UpdateMsg::ListUpdates { uuid, ret };
let _ = self.sender.send(msg).await;
receiver.await.expect("update actor killed.")
}
async fn update_status(&self, uuid: Uuid, id: u64) -> Result<UpdateStatus> {
let (ret, receiver) = oneshot::channel();
let msg = UpdateMsg::GetUpdate { uuid, id, ret };
let _ = self.sender.send(msg).await;
receiver.await.expect("update actor killed.")
}
async fn delete(&self, uuid: Uuid) -> Result<()> {
let (ret, receiver) = oneshot::channel();
let msg = UpdateMsg::Delete { uuid, ret };
let _ = self.sender.send(msg).await;
receiver.await.expect("update actor killed.")
}
async fn create(&self, uuid: Uuid) -> Result<()> {
let (ret, receiver) = oneshot::channel();
let msg = UpdateMsg::Create { uuid, ret };
let _ = self.sender.send(msg).await;
receiver.await.expect("update actor killed.")
}
async fn snapshot(&self, uuid: Uuid, path: PathBuf) -> Result<()> {
let (ret, receiver) = oneshot::channel();
let msg = UpdateMsg::Snapshot { uuid, path, ret };
let _ = self.sender.send(msg).await;
receiver.await.expect("update actor killed.")
}
}

View File

@ -0,0 +1,37 @@
use std::path::PathBuf;
use uuid::Uuid;
use tokio::sync::{oneshot, mpsc};
use super::{Result, PayloadData, UpdateStatus, UpdateMeta};
pub enum UpdateMsg<D> {
Update {
uuid: Uuid,
meta: UpdateMeta,
data: mpsc::Receiver<PayloadData<D>>,
ret: oneshot::Sender<Result<UpdateStatus>>,
},
ListUpdates {
uuid: Uuid,
ret: oneshot::Sender<Result<Vec<UpdateStatus>>>,
},
GetUpdate {
uuid: Uuid,
ret: oneshot::Sender<Result<UpdateStatus>>,
id: u64,
},
Delete {
uuid: Uuid,
ret: oneshot::Sender<Result<()>>,
},
Create {
uuid: Uuid,
ret: oneshot::Sender<Result<()>>,
},
Snapshot {
uuid: Uuid,
path: PathBuf,
ret: oneshot::Sender<Result<()>>,
},
}

View File

@ -0,0 +1,51 @@
mod actor;
mod store;
mod message;
mod handle_impl;
mod update_store;
use std::path::PathBuf;
use thiserror::Error;
use tokio::sync::mpsc;
use uuid::Uuid;
use crate::index::UpdateResult;
use crate::index_controller::{UpdateMeta, UpdateStatus};
use actor::UpdateActor;
use message::UpdateMsg;
use store::{UpdateStoreStore, MapUpdateStoreStore};
pub use handle_impl::UpdateActorHandleImpl;
pub type Result<T> = std::result::Result<T, UpdateError>;
type UpdateStore = update_store::UpdateStore<UpdateMeta, UpdateResult, String>;
type PayloadData<D> = std::result::Result<D, Box<dyn std::error::Error + Sync + Send + 'static>>;
#[derive(Debug, Error)]
pub enum UpdateError {
#[error("error with update: {0}")]
Error(Box<dyn std::error::Error + Sync + Send + 'static>),
#[error("Index {0} doesn't exist.")]
UnexistingIndex(Uuid),
#[error("Update {0} doesn't exist.")]
UnexistingUpdate(u64),
}
#[async_trait::async_trait]
pub trait UpdateActorHandle {
type Data: AsRef<[u8]> + Sized + 'static + Sync + Send;
async fn get_all_updates_status(&self, uuid: Uuid) -> Result<Vec<UpdateStatus>>;
async fn update_status(&self, uuid: Uuid, id: u64) -> Result<UpdateStatus>;
async fn delete(&self, uuid: Uuid) -> Result<()>;
async fn create(&self, uuid: Uuid) -> Result<()>;
async fn snapshot(&self, uuid: Uuid, path: PathBuf) -> Result<()>;
async fn update(
&self,
meta: UpdateMeta,
data: mpsc::Receiver<PayloadData<Self::Data>>,
uuid: Uuid,
) -> Result<UpdateStatus> ;
}

View File

@ -0,0 +1,112 @@
use std::collections::HashMap;
use std::collections::hash_map::Entry;
use std::path::{Path, PathBuf};
use std::sync::Arc;
use uuid::Uuid;
use tokio::sync::RwLock;
use tokio::fs;
use crate::index_controller::IndexActorHandle;
use super::{UpdateStore, UpdateError, Result};
#[async_trait::async_trait]
pub trait UpdateStoreStore {
async fn get_or_create(&self, uuid: Uuid) -> Result<Arc<UpdateStore>>;
async fn delete(&self, uuid: Uuid) -> Result<Option<Arc<UpdateStore>>>;
async fn get(&self, uuid: Uuid) -> Result<Option<Arc<UpdateStore>>>;
}
pub struct MapUpdateStoreStore<I> {
db: Arc<RwLock<HashMap<Uuid, Arc<UpdateStore>>>>,
index_handle: I,
path: PathBuf,
update_store_size: usize,
}
impl<I: IndexActorHandle> MapUpdateStoreStore<I> {
pub fn new(
index_handle: I,
path: impl AsRef<Path>,
update_store_size: usize,
) -> Self {
let db = Arc::new(RwLock::new(HashMap::new()));
let path = path.as_ref().to_owned();
Self {
db,
index_handle,
path,
update_store_size,
}
}
}
#[async_trait::async_trait]
impl<I: IndexActorHandle + 'static> UpdateStoreStore for MapUpdateStoreStore<I> {
async fn get_or_create(&self, uuid: Uuid) -> Result<Arc<UpdateStore>> {
match self.db.write().await.entry(uuid) {
Entry::Vacant(e) => {
let mut options = heed::EnvOpenOptions::new();
let update_store_size = self.update_store_size;
options.map_size(update_store_size);
let path = self.path.clone().join(format!("updates-{}", e.key()));
fs::create_dir_all(&path).await.unwrap();
let index_handle = self.index_handle.clone();
let store = UpdateStore::open(options, &path, move |meta, file| {
futures::executor::block_on(index_handle.update(meta, file))
})
.map_err(|e| UpdateError::Error(e.into()))?;
let store = e.insert(store);
Ok(store.clone())
}
Entry::Occupied(e) => Ok(e.get().clone()),
}
}
async fn get(&self, uuid: Uuid) -> Result<Option<Arc<UpdateStore>>> {
let guard = self.db.read().await;
match guard.get(&uuid) {
Some(uuid) => Ok(Some(uuid.clone())),
None => {
// The index is not found in the found in the loaded indexes, so we attempt to load
// it from disk. We need to acquire a write lock **before** attempting to open the
// index, because someone could be trying to open it at the same time as us.
drop(guard);
let path = self.path.clone().join(format!("updates-{}", uuid));
if path.exists() {
let mut guard = self.db.write().await;
match guard.entry(uuid) {
Entry::Vacant(entry) => {
// We can safely load the index
let index_handle = self.index_handle.clone();
let mut options = heed::EnvOpenOptions::new();
let update_store_size = self.update_store_size;
options.map_size(update_store_size);
let store = UpdateStore::open(options, &path, move |meta, file| {
futures::executor::block_on(index_handle.update(meta, file))
})
.map_err(|e| UpdateError::Error(e.into()))?;
let store = entry.insert(store);
Ok(Some(store.clone()))
}
Entry::Occupied(entry) => {
// The index was loaded while we attempted to to iter
Ok(Some(entry.get().clone()))
}
}
} else {
Ok(None)
}
}
}
}
async fn delete(&self, uuid: Uuid) -> Result<Option<Arc<UpdateStore>>> {
let store = self.db.write().await.remove(&uuid);
let path = self.path.clone().join(format!("updates-{}", uuid));
if store.is_some() || path.exists() {
fs::remove_dir_all(path).await.unwrap();
}
Ok(store)
}
}

View File

@ -1,343 +0,0 @@
use std::fs::create_dir_all;
use std::path::{Path, PathBuf};
use heed::{
types::{ByteSlice, Str},
Database, Env, EnvOpenOptions,CompactionOption
};
use log::{info, warn};
use thiserror::Error;
use tokio::sync::{mpsc, oneshot};
use uuid::Uuid;
const UUID_STORE_SIZE: usize = 1_073_741_824; //1GiB
pub type Result<T> = std::result::Result<T, UuidError>;
#[derive(Debug)]
enum UuidResolveMsg {
Resolve {
uid: String,
ret: oneshot::Sender<Result<Uuid>>,
},
GetOrCreate {
uid: String,
ret: oneshot::Sender<Result<Uuid>>,
},
Create {
uid: String,
ret: oneshot::Sender<Result<Uuid>>,
},
Delete {
uid: String,
ret: oneshot::Sender<Result<Uuid>>,
},
List {
ret: oneshot::Sender<Result<Vec<(String, Uuid)>>>,
},
SnapshotRequest {
path: PathBuf,
ret: oneshot::Sender<Result<Vec<Uuid>>>,
},
}
struct UuidResolverActor<S> {
inbox: mpsc::Receiver<UuidResolveMsg>,
store: S,
}
impl<S: UuidStore> UuidResolverActor<S> {
fn new(inbox: mpsc::Receiver<UuidResolveMsg>, store: S) -> Self {
Self { inbox, store }
}
async fn run(mut self) {
use UuidResolveMsg::*;
info!("uuid resolver started");
loop {
match self.inbox.recv().await {
Some(Create { uid: name, ret }) => {
let _ = ret.send(self.handle_create(name).await);
}
Some(GetOrCreate { uid: name, ret }) => {
let _ = ret.send(self.handle_get_or_create(name).await);
}
Some(Resolve { uid: name, ret }) => {
let _ = ret.send(self.handle_resolve(name).await);
}
Some(Delete { uid: name, ret }) => {
let _ = ret.send(self.handle_delete(name).await);
}
Some(List { ret }) => {
let _ = ret.send(self.handle_list().await);
}
Some(SnapshotRequest { path, ret }) => {
let _ = ret.send(self.handle_snapshot(path).await);
}
// all senders have been dropped, need to quit.
None => break,
}
}
warn!("exiting uuid resolver loop");
}
async fn handle_create(&self, uid: String) -> Result<Uuid> {
if !is_index_uid_valid(&uid) {
return Err(UuidError::BadlyFormatted(uid));
}
self.store.create_uuid(uid, true).await
}
async fn handle_get_or_create(&self, uid: String) -> Result<Uuid> {
if !is_index_uid_valid(&uid) {
return Err(UuidError::BadlyFormatted(uid));
}
self.store.create_uuid(uid, false).await
}
async fn handle_resolve(&self, uid: String) -> Result<Uuid> {
self.store
.get_uuid(uid.clone())
.await?
.ok_or(UuidError::UnexistingIndex(uid))
}
async fn handle_delete(&self, uid: String) -> Result<Uuid> {
self.store
.delete(uid.clone())
.await?
.ok_or(UuidError::UnexistingIndex(uid))
}
async fn handle_list(&self) -> Result<Vec<(String, Uuid)>> {
let result = self.store.list().await?;
Ok(result)
}
async fn handle_snapshot(&self, path: PathBuf) -> Result<Vec<Uuid>> {
self.store.snapshot(path).await
}
}
fn is_index_uid_valid(uid: &str) -> bool {
uid.chars()
.all(|x| x.is_ascii_alphanumeric() || x == '-' || x == '_')
}
#[derive(Clone)]
pub struct UuidResolverHandle {
sender: mpsc::Sender<UuidResolveMsg>,
}
impl UuidResolverHandle {
pub fn new(path: impl AsRef<Path>) -> anyhow::Result<Self> {
let (sender, reveiver) = mpsc::channel(100);
let store = HeedUuidStore::new(path)?;
let actor = UuidResolverActor::new(reveiver, store);
tokio::spawn(actor.run());
Ok(Self { sender })
}
pub async fn resolve(&self, name: String) -> anyhow::Result<Uuid> {
let (ret, receiver) = oneshot::channel();
let msg = UuidResolveMsg::Resolve { uid: name, ret };
let _ = self.sender.send(msg).await;
Ok(receiver
.await
.expect("Uuid resolver actor has been killed")?)
}
pub async fn get_or_create(&self, name: String) -> Result<Uuid> {
let (ret, receiver) = oneshot::channel();
let msg = UuidResolveMsg::GetOrCreate { uid: name, ret };
let _ = self.sender.send(msg).await;
Ok(receiver
.await
.expect("Uuid resolver actor has been killed")?)
}
pub async fn create(&self, name: String) -> anyhow::Result<Uuid> {
let (ret, receiver) = oneshot::channel();
let msg = UuidResolveMsg::Create { uid: name, ret };
let _ = self.sender.send(msg).await;
Ok(receiver
.await
.expect("Uuid resolver actor has been killed")?)
}
pub async fn delete(&self, name: String) -> anyhow::Result<Uuid> {
let (ret, receiver) = oneshot::channel();
let msg = UuidResolveMsg::Delete { uid: name, ret };
let _ = self.sender.send(msg).await;
Ok(receiver
.await
.expect("Uuid resolver actor has been killed")?)
}
pub async fn list(&self) -> anyhow::Result<Vec<(String, Uuid)>> {
let (ret, receiver) = oneshot::channel();
let msg = UuidResolveMsg::List { ret };
let _ = self.sender.send(msg).await;
Ok(receiver
.await
.expect("Uuid resolver actor has been killed")?)
}
pub async fn snapshot(&self, path: PathBuf) -> Result<Vec<Uuid>> {
let (ret, receiver) = oneshot::channel();
let msg = UuidResolveMsg::SnapshotRequest { path, ret };
let _ = self.sender.send(msg).await;
Ok(receiver
.await
.expect("Uuid resolver actor has been killed")?)
}
}
#[derive(Debug, Error)]
pub enum UuidError {
#[error("Name already exist.")]
NameAlreadyExist,
#[error("Index \"{0}\" doesn't exist.")]
UnexistingIndex(String),
#[error("Error performing task: {0}")]
TokioTask(#[from] tokio::task::JoinError),
#[error("Database error: {0}")]
Heed(#[from] heed::Error),
#[error("Uuid error: {0}")]
Uuid(#[from] uuid::Error),
#[error("Badly formatted index uid: {0}")]
BadlyFormatted(String),
}
#[async_trait::async_trait]
trait UuidStore {
// Create a new entry for `name`. Return an error if `err` and the entry already exists, return
// the uuid otherwise.
async fn create_uuid(&self, uid: String, err: bool) -> Result<Uuid>;
async fn get_uuid(&self, uid: String) -> Result<Option<Uuid>>;
async fn delete(&self, uid: String) -> Result<Option<Uuid>>;
async fn list(&self) -> Result<Vec<(String, Uuid)>>;
async fn snapshot(&self, path: PathBuf) -> Result<Vec<Uuid>>;
}
struct HeedUuidStore {
env: Env,
db: Database<Str, ByteSlice>,
}
impl HeedUuidStore {
fn new(path: impl AsRef<Path>) -> anyhow::Result<Self> {
let path = path.as_ref().join("index_uuids");
create_dir_all(&path)?;
let mut options = EnvOpenOptions::new();
options.map_size(UUID_STORE_SIZE); // 1GB
let env = options.open(path)?;
let db = env.create_database(None)?;
Ok(Self { env, db })
}
}
#[async_trait::async_trait]
impl UuidStore for HeedUuidStore {
async fn create_uuid(&self, name: String, err: bool) -> Result<Uuid> {
let env = self.env.clone();
let db = self.db;
tokio::task::spawn_blocking(move || {
let mut txn = env.write_txn()?;
match db.get(&txn, &name)? {
Some(uuid) => {
if err {
Err(UuidError::NameAlreadyExist)
} else {
let uuid = Uuid::from_slice(uuid)?;
Ok(uuid)
}
}
None => {
let uuid = Uuid::new_v4();
db.put(&mut txn, &name, uuid.as_bytes())?;
txn.commit()?;
Ok(uuid)
}
}
})
.await?
}
async fn get_uuid(&self, name: String) -> Result<Option<Uuid>> {
let env = self.env.clone();
let db = self.db;
tokio::task::spawn_blocking(move || {
let txn = env.read_txn()?;
match db.get(&txn, &name)? {
Some(uuid) => {
let uuid = Uuid::from_slice(uuid)?;
Ok(Some(uuid))
}
None => Ok(None),
}
})
.await?
}
async fn delete(&self, uid: String) -> Result<Option<Uuid>> {
let env = self.env.clone();
let db = self.db;
tokio::task::spawn_blocking(move || {
let mut txn = env.write_txn()?;
match db.get(&txn, &uid)? {
Some(uuid) => {
let uuid = Uuid::from_slice(uuid)?;
db.delete(&mut txn, &uid)?;
txn.commit()?;
Ok(Some(uuid))
}
None => Ok(None),
}
})
.await?
}
async fn list(&self) -> Result<Vec<(String, Uuid)>> {
let env = self.env.clone();
let db = self.db;
tokio::task::spawn_blocking(move || {
let txn = env.read_txn()?;
let mut entries = Vec::new();
for entry in db.iter(&txn)? {
let (name, uuid) = entry?;
let uuid = Uuid::from_slice(uuid)?;
entries.push((name.to_owned(), uuid))
}
Ok(entries)
})
.await?
}
async fn snapshot(&self, mut path: PathBuf) -> Result<Vec<Uuid>> {
let env = self.env.clone();
let db = self.db;
tokio::task::spawn_blocking(move || {
// Write transaction to acquire a lock on the database.
let txn = env.write_txn()?;
let mut entries = Vec::new();
for entry in db.iter(&txn)? {
let (_, uuid) = entry?;
let uuid = Uuid::from_slice(uuid)?;
entries.push(uuid)
}
// only perform snapshot if there are indexes
if !entries.is_empty() {
path.push("index_uuids");
create_dir_all(&path).unwrap();
path.push("data.mdb");
env.copy_to_path(path, CompactionOption::Enabled)?;
}
Ok(entries)
})
.await?
}
}

View File

@ -0,0 +1,94 @@
use std::path::PathBuf;
use log::{info, warn};
use tokio::sync::mpsc;
use uuid::Uuid;
use super::{UuidResolveMsg, UuidStore, Result, UuidError};
pub struct UuidResolverActor<S> {
inbox: mpsc::Receiver<UuidResolveMsg>,
store: S,
}
impl<S: UuidStore> UuidResolverActor<S> {
pub fn new(inbox: mpsc::Receiver<UuidResolveMsg>, store: S) -> Self {
Self { inbox, store }
}
pub async fn run(mut self) {
use UuidResolveMsg::*;
info!("uuid resolver started");
loop {
match self.inbox.recv().await {
Some(Create { uid: name, ret }) => {
let _ = ret.send(self.handle_create(name).await);
}
Some(GetOrCreate { uid: name, ret }) => {
let _ = ret.send(self.handle_get_or_create(name).await);
}
Some(Resolve { uid: name, ret }) => {
let _ = ret.send(self.handle_resolve(name).await);
}
Some(Delete { uid: name, ret }) => {
let _ = ret.send(self.handle_delete(name).await);
}
Some(List { ret }) => {
let _ = ret.send(self.handle_list().await);
}
Some(SnapshotRequest { path, ret }) => {
let _ = ret.send(self.handle_snapshot(path).await);
}
// all senders have been dropped, need to quit.
None => break,
}
}
warn!("exiting uuid resolver loop");
}
async fn handle_create(&self, uid: String) -> Result<Uuid> {
if !is_index_uid_valid(&uid) {
return Err(UuidError::BadlyFormatted(uid));
}
self.store.create_uuid(uid, true).await
}
async fn handle_get_or_create(&self, uid: String) -> Result<Uuid> {
if !is_index_uid_valid(&uid) {
return Err(UuidError::BadlyFormatted(uid));
}
self.store.create_uuid(uid, false).await
}
async fn handle_resolve(&self, uid: String) -> Result<Uuid> {
self.store
.get_uuid(uid.clone())
.await?
.ok_or(UuidError::UnexistingIndex(uid))
}
async fn handle_delete(&self, uid: String) -> Result<Uuid> {
self.store
.delete(uid.clone())
.await?
.ok_or(UuidError::UnexistingIndex(uid))
}
async fn handle_list(&self) -> Result<Vec<(String, Uuid)>> {
let result = self.store.list().await?;
Ok(result)
}
async fn handle_snapshot(&self, path: PathBuf) -> Result<Vec<Uuid>> {
self.store.snapshot(path).await
}
}
fn is_index_uid_valid(uid: &str) -> bool {
uid.chars()
.all(|x| x.is_ascii_alphanumeric() || x == '-' || x == '_')
}

View File

@ -0,0 +1,78 @@
use std::path::{Path, PathBuf};
use tokio::sync::{mpsc, oneshot};
use uuid::Uuid;
use super::{HeedUuidStore, UuidResolverActor, UuidResolveMsg, UuidResolverHandle, Result};
#[derive(Clone)]
pub struct UuidResolverHandleImpl {
sender: mpsc::Sender<UuidResolveMsg>,
}
impl UuidResolverHandleImpl {
pub fn new(path: impl AsRef<Path>) -> anyhow::Result<Self> {
let (sender, reveiver) = mpsc::channel(100);
let store = HeedUuidStore::new(path)?;
let actor = UuidResolverActor::new(reveiver, store);
tokio::spawn(actor.run());
Ok(Self { sender })
}
}
#[async_trait::async_trait]
impl UuidResolverHandle for UuidResolverHandleImpl {
async fn resolve(&self, name: String) -> anyhow::Result<Uuid> {
let (ret, receiver) = oneshot::channel();
let msg = UuidResolveMsg::Resolve { uid: name, ret };
let _ = self.sender.send(msg).await;
Ok(receiver
.await
.expect("Uuid resolver actor has been killed")?)
}
async fn get_or_create(&self, name: String) -> Result<Uuid> {
let (ret, receiver) = oneshot::channel();
let msg = UuidResolveMsg::GetOrCreate { uid: name, ret };
let _ = self.sender.send(msg).await;
Ok(receiver
.await
.expect("Uuid resolver actor has been killed")?)
}
async fn create(&self, name: String) -> anyhow::Result<Uuid> {
let (ret, receiver) = oneshot::channel();
let msg = UuidResolveMsg::Create { uid: name, ret };
let _ = self.sender.send(msg).await;
Ok(receiver
.await
.expect("Uuid resolver actor has been killed")?)
}
async fn delete(&self, name: String) -> anyhow::Result<Uuid> {
let (ret, receiver) = oneshot::channel();
let msg = UuidResolveMsg::Delete { uid: name, ret };
let _ = self.sender.send(msg).await;
Ok(receiver
.await
.expect("Uuid resolver actor has been killed")?)
}
async fn list(&self) -> anyhow::Result<Vec<(String, Uuid)>> {
let (ret, receiver) = oneshot::channel();
let msg = UuidResolveMsg::List { ret };
let _ = self.sender.send(msg).await;
Ok(receiver
.await
.expect("Uuid resolver actor has been killed")?)
}
async fn snapshot(&self, path: PathBuf) -> Result<Vec<Uuid>> {
let (ret, receiver) = oneshot::channel();
let msg = UuidResolveMsg::SnapshotRequest { path, ret };
let _ = self.sender.send(msg).await;
Ok(receiver
.await
.expect("Uuid resolver actor has been killed")?)
}
}

View File

@ -0,0 +1,33 @@
use std::path::PathBuf;
use tokio::sync::oneshot;
use uuid::Uuid;
use super::Result;
#[derive(Debug)]
pub enum UuidResolveMsg {
Resolve {
uid: String,
ret: oneshot::Sender<Result<Uuid>>,
},
GetOrCreate {
uid: String,
ret: oneshot::Sender<Result<Uuid>>,
},
Create {
uid: String,
ret: oneshot::Sender<Result<Uuid>>,
},
Delete {
uid: String,
ret: oneshot::Sender<Result<Uuid>>,
},
List {
ret: oneshot::Sender<Result<Vec<(String, Uuid)>>>,
},
SnapshotRequest {
path: PathBuf,
ret: oneshot::Sender<Result<Vec<Uuid>>>,
},
}

View File

@ -0,0 +1,45 @@
mod actor;
mod handle_impl;
mod message;
mod store;
use std::path::PathBuf;
use thiserror::Error;
use uuid::Uuid;
use actor::UuidResolverActor;
use message::UuidResolveMsg;
use store::{HeedUuidStore, UuidStore};
pub use handle_impl::UuidResolverHandleImpl;
const UUID_STORE_SIZE: usize = 1_073_741_824; //1GiB
pub type Result<T> = std::result::Result<T, UuidError>;
#[async_trait::async_trait]
pub trait UuidResolverHandle {
async fn resolve(&self, name: String) -> anyhow::Result<Uuid>;
async fn get_or_create(&self, name: String) -> Result<Uuid>;
async fn create(&self, name: String) -> anyhow::Result<Uuid>;
async fn delete(&self, name: String) -> anyhow::Result<Uuid>;
async fn list(&self) -> anyhow::Result<Vec<(String, Uuid)>>;
async fn snapshot(&self, path: PathBuf) -> Result<Vec<Uuid>>;
}
#[derive(Debug, Error)]
pub enum UuidError {
#[error("Name already exist.")]
NameAlreadyExist,
#[error("Index \"{0}\" doesn't exist.")]
UnexistingIndex(String),
#[error("Error performing task: {0}")]
TokioTask(#[from] tokio::task::JoinError),
#[error("Database error: {0}")]
Heed(#[from] heed::Error),
#[error("Uuid error: {0}")]
Uuid(#[from] uuid::Error),
#[error("Badly formatted index uid: {0}")]
BadlyFormatted(String),
}

View File

@ -0,0 +1,140 @@
use std::fs::create_dir_all;
use std::path::{Path, PathBuf};
use heed::{
types::{ByteSlice, Str},
Database, Env, EnvOpenOptions,CompactionOption
};
use uuid::Uuid;
use super::{UUID_STORE_SIZE, UuidError, Result};
#[async_trait::async_trait]
pub trait UuidStore {
// Create a new entry for `name`. Return an error if `err` and the entry already exists, return
// the uuid otherwise.
async fn create_uuid(&self, uid: String, err: bool) -> Result<Uuid>;
async fn get_uuid(&self, uid: String) -> Result<Option<Uuid>>;
async fn delete(&self, uid: String) -> Result<Option<Uuid>>;
async fn list(&self) -> Result<Vec<(String, Uuid)>>;
async fn snapshot(&self, path: PathBuf) -> Result<Vec<Uuid>>;
}
pub struct HeedUuidStore {
env: Env,
db: Database<Str, ByteSlice>,
}
impl HeedUuidStore {
pub fn new(path: impl AsRef<Path>) -> anyhow::Result<Self> {
let path = path.as_ref().join("index_uuids");
create_dir_all(&path)?;
let mut options = EnvOpenOptions::new();
options.map_size(UUID_STORE_SIZE); // 1GB
let env = options.open(path)?;
let db = env.create_database(None)?;
Ok(Self { env, db })
}
}
#[async_trait::async_trait]
impl UuidStore for HeedUuidStore {
async fn create_uuid(&self, name: String, err: bool) -> Result<Uuid> {
let env = self.env.clone();
let db = self.db;
tokio::task::spawn_blocking(move || {
let mut txn = env.write_txn()?;
match db.get(&txn, &name)? {
Some(uuid) => {
if err {
Err(UuidError::NameAlreadyExist)
} else {
let uuid = Uuid::from_slice(uuid)?;
Ok(uuid)
}
}
None => {
let uuid = Uuid::new_v4();
db.put(&mut txn, &name, uuid.as_bytes())?;
txn.commit()?;
Ok(uuid)
}
}
})
.await?
}
async fn get_uuid(&self, name: String) -> Result<Option<Uuid>> {
let env = self.env.clone();
let db = self.db;
tokio::task::spawn_blocking(move || {
let txn = env.read_txn()?;
match db.get(&txn, &name)? {
Some(uuid) => {
let uuid = Uuid::from_slice(uuid)?;
Ok(Some(uuid))
}
None => Ok(None),
}
})
.await?
}
async fn delete(&self, uid: String) -> Result<Option<Uuid>> {
let env = self.env.clone();
let db = self.db;
tokio::task::spawn_blocking(move || {
let mut txn = env.write_txn()?;
match db.get(&txn, &uid)? {
Some(uuid) => {
let uuid = Uuid::from_slice(uuid)?;
db.delete(&mut txn, &uid)?;
txn.commit()?;
Ok(Some(uuid))
}
None => Ok(None),
}
})
.await?
}
async fn list(&self) -> Result<Vec<(String, Uuid)>> {
let env = self.env.clone();
let db = self.db;
tokio::task::spawn_blocking(move || {
let txn = env.read_txn()?;
let mut entries = Vec::new();
for entry in db.iter(&txn)? {
let (name, uuid) = entry?;
let uuid = Uuid::from_slice(uuid)?;
entries.push((name.to_owned(), uuid))
}
Ok(entries)
})
.await?
}
async fn snapshot(&self, mut path: PathBuf) -> Result<Vec<Uuid>> {
let env = self.env.clone();
let db = self.db;
tokio::task::spawn_blocking(move || {
// Write transaction to acquire a lock on the database.
let txn = env.write_txn()?;
let mut entries = Vec::new();
for entry in db.iter(&txn)? {
let (_, uuid) = entry?;
let uuid = Uuid::from_slice(uuid)?;
entries.push(uuid)
}
// only perform snapshot if there are indexes
if !entries.is_empty() {
path.push("index_uuids");
create_dir_all(&path).unwrap();
path.push("data.mdb");
env.copy_to_path(path, CompactionOption::Enabled)?;
}
Ok(entries)
})
.await?
}
}

View File

@ -43,7 +43,7 @@ impl Server {
ignore_snapshot_if_db_exists: false, ignore_snapshot_if_db_exists: false,
snapshot_dir: ".".into(), snapshot_dir: ".".into(),
schedule_snapshot: false, schedule_snapshot: false,
snapshot_interval_sec: None, snapshot_interval_sec: 0,
import_dump: None, import_dump: None,
indexer_options: IndexerOpts::default(), indexer_options: IndexerOpts::default(),
#[cfg(all(not(debug_assertions), feature = "sentry"))] #[cfg(all(not(debug_assertions), feature = "sentry"))]