mirror of
https://github.com/GreptimeTeam/greptimedb.git
synced 2026-05-21 15:30:40 +00:00
feat: start LocalManager in Metasrv (#1279)
* feat: procedure store in Metasrv, backed by Etcd; start `LocalManager` in Metasrv leader * fix: resolve PR comments * fix: resolve PR comments
This commit is contained in:
@@ -13,6 +13,7 @@
|
||||
// limitations under the License.
|
||||
|
||||
use std::any::Any;
|
||||
use std::string::FromUtf8Error;
|
||||
use std::sync::Arc;
|
||||
|
||||
use common_error::prelude::*;
|
||||
@@ -47,10 +48,11 @@ pub enum Error {
|
||||
backtrace: Backtrace,
|
||||
},
|
||||
|
||||
#[snafu(display("Failed to put {}, source: {}", key, source))]
|
||||
#[snafu(display("Failed to put state, key: '{key}', source: {source}"))]
|
||||
PutState {
|
||||
key: String,
|
||||
source: object_store::Error,
|
||||
#[snafu(backtrace)]
|
||||
source: BoxedError,
|
||||
},
|
||||
|
||||
#[snafu(display("Failed to delete {}, source: {}", key, source))]
|
||||
@@ -59,10 +61,18 @@ pub enum Error {
|
||||
source: object_store::Error,
|
||||
},
|
||||
|
||||
#[snafu(display("Failed to list {}, source: {}", path, source))]
|
||||
#[snafu(display("Failed to delete keys: '{keys}', source: {source}"))]
|
||||
DeleteStates {
|
||||
keys: String,
|
||||
#[snafu(backtrace)]
|
||||
source: BoxedError,
|
||||
},
|
||||
|
||||
#[snafu(display("Failed to list state, path: '{path}', source: {source}"))]
|
||||
ListState {
|
||||
path: String,
|
||||
source: object_store::Error,
|
||||
#[snafu(backtrace)]
|
||||
source: BoxedError,
|
||||
},
|
||||
|
||||
#[snafu(display("Failed to read {}, source: {}", key, source))]
|
||||
@@ -107,6 +117,9 @@ pub enum Error {
|
||||
source: Arc<Error>,
|
||||
procedure_id: ProcedureId,
|
||||
},
|
||||
|
||||
#[snafu(display("Corrupted data, error: {source}"))]
|
||||
CorruptedData { source: FromUtf8Error },
|
||||
}
|
||||
|
||||
pub type Result<T> = std::result::Result<T, Error>;
|
||||
@@ -114,11 +127,13 @@ pub type Result<T> = std::result::Result<T, Error>;
|
||||
impl ErrorExt for Error {
|
||||
fn status_code(&self) -> StatusCode {
|
||||
match self {
|
||||
Error::External { source } => source.status_code(),
|
||||
Error::External { source }
|
||||
| Error::PutState { source, .. }
|
||||
| Error::DeleteStates { source, .. }
|
||||
| Error::ListState { source, .. } => source.status_code(),
|
||||
|
||||
Error::ToJson { .. }
|
||||
| Error::PutState { .. }
|
||||
| Error::DeleteState { .. }
|
||||
| Error::ListState { .. }
|
||||
| Error::ReadState { .. }
|
||||
| Error::FromJson { .. }
|
||||
| Error::RetryTimesExceeded { .. }
|
||||
@@ -127,7 +142,7 @@ impl ErrorExt for Error {
|
||||
Error::LoaderConflict { .. } | Error::DuplicateProcedure { .. } => {
|
||||
StatusCode::InvalidArguments
|
||||
}
|
||||
Error::ProcedurePanic { .. } => StatusCode::Unexpected,
|
||||
Error::ProcedurePanic { .. } | Error::CorruptedData { .. } => StatusCode::Unexpected,
|
||||
Error::ProcedureExec { source, .. } => source.status_code(),
|
||||
}
|
||||
}
|
||||
|
||||
@@ -17,7 +17,7 @@
|
||||
pub mod error;
|
||||
pub mod local;
|
||||
mod procedure;
|
||||
mod store;
|
||||
pub mod store;
|
||||
pub mod watcher;
|
||||
|
||||
pub use crate::error::{Error, Result};
|
||||
|
||||
@@ -22,7 +22,6 @@ use std::time::Duration;
|
||||
use async_trait::async_trait;
|
||||
use backon::ExponentialBuilder;
|
||||
use common_telemetry::logging;
|
||||
use object_store::ObjectStore;
|
||||
use snafu::ensure;
|
||||
use tokio::sync::watch::{self, Receiver, Sender};
|
||||
use tokio::sync::Notify;
|
||||
@@ -31,7 +30,7 @@ use crate::error::{DuplicateProcedureSnafu, LoaderConflictSnafu, Result};
|
||||
use crate::local::lock::LockMap;
|
||||
use crate::local::runner::Runner;
|
||||
use crate::procedure::BoxedProcedureLoader;
|
||||
use crate::store::{ObjectStateStore, ProcedureMessage, ProcedureStore, StateStoreRef};
|
||||
use crate::store::{ProcedureMessage, ProcedureStore, StateStoreRef};
|
||||
use crate::{
|
||||
BoxedProcedure, ContextProvider, LockKey, ProcedureId, ProcedureManager, ProcedureState,
|
||||
ProcedureWithId, Watcher,
|
||||
@@ -291,12 +290,19 @@ impl ManagerContext {
|
||||
/// Config for [LocalManager].
|
||||
#[derive(Debug)]
|
||||
pub struct ManagerConfig {
|
||||
/// Object store
|
||||
pub object_store: ObjectStore,
|
||||
pub max_retry_times: usize,
|
||||
pub retry_delay: Duration,
|
||||
}
|
||||
|
||||
impl Default for ManagerConfig {
|
||||
fn default() -> Self {
|
||||
Self {
|
||||
max_retry_times: 3,
|
||||
retry_delay: Duration::from_millis(500),
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/// A [ProcedureManager] that maintains procedure states locally.
|
||||
pub struct LocalManager {
|
||||
manager_ctx: Arc<ManagerContext>,
|
||||
@@ -307,10 +313,10 @@ pub struct LocalManager {
|
||||
|
||||
impl LocalManager {
|
||||
/// Create a new [LocalManager] with specific `config`.
|
||||
pub fn new(config: ManagerConfig) -> LocalManager {
|
||||
pub fn new(config: ManagerConfig, state_store: StateStoreRef) -> LocalManager {
|
||||
LocalManager {
|
||||
manager_ctx: Arc::new(ManagerContext::new()),
|
||||
state_store: Arc::new(ObjectStateStore::new(config.object_store)),
|
||||
state_store,
|
||||
max_retry_times: config.max_retry_times,
|
||||
retry_delay: config.retry_delay,
|
||||
}
|
||||
@@ -423,6 +429,7 @@ impl ProcedureManager for LocalManager {
|
||||
mod test_util {
|
||||
use common_test_util::temp_dir::TempDir;
|
||||
use object_store::services::Fs as Builder;
|
||||
use object_store::ObjectStore;
|
||||
|
||||
use super::*;
|
||||
|
||||
@@ -446,6 +453,7 @@ mod tests {
|
||||
|
||||
use super::*;
|
||||
use crate::error::Error;
|
||||
use crate::store::ObjectStateStore;
|
||||
use crate::{Context, Procedure, Status};
|
||||
|
||||
#[test]
|
||||
@@ -554,11 +562,11 @@ mod tests {
|
||||
fn test_register_loader() {
|
||||
let dir = create_temp_dir("register");
|
||||
let config = ManagerConfig {
|
||||
object_store: test_util::new_object_store(&dir),
|
||||
max_retry_times: 3,
|
||||
retry_delay: Duration::from_millis(500),
|
||||
};
|
||||
let manager = LocalManager::new(config);
|
||||
let state_store = Arc::new(ObjectStateStore::new(test_util::new_object_store(&dir)));
|
||||
let manager = LocalManager::new(config, state_store);
|
||||
|
||||
manager
|
||||
.register_loader("ProcedureToLoad", ProcedureToLoad::loader())
|
||||
@@ -575,11 +583,11 @@ mod tests {
|
||||
let dir = create_temp_dir("recover");
|
||||
let object_store = test_util::new_object_store(&dir);
|
||||
let config = ManagerConfig {
|
||||
object_store: object_store.clone(),
|
||||
max_retry_times: 3,
|
||||
retry_delay: Duration::from_millis(500),
|
||||
};
|
||||
let manager = LocalManager::new(config);
|
||||
let state_store = Arc::new(ObjectStateStore::new(object_store.clone()));
|
||||
let manager = LocalManager::new(config, state_store);
|
||||
|
||||
manager
|
||||
.register_loader("ProcedureToLoad", ProcedureToLoad::loader())
|
||||
@@ -621,11 +629,11 @@ mod tests {
|
||||
async fn test_submit_procedure() {
|
||||
let dir = create_temp_dir("submit");
|
||||
let config = ManagerConfig {
|
||||
object_store: test_util::new_object_store(&dir),
|
||||
max_retry_times: 3,
|
||||
retry_delay: Duration::from_millis(500),
|
||||
};
|
||||
let manager = LocalManager::new(config);
|
||||
let state_store = Arc::new(ObjectStateStore::new(test_util::new_object_store(&dir)));
|
||||
let manager = LocalManager::new(config, state_store);
|
||||
|
||||
let procedure_id = ProcedureId::random();
|
||||
assert!(manager
|
||||
@@ -669,11 +677,11 @@ mod tests {
|
||||
async fn test_state_changed_on_err() {
|
||||
let dir = create_temp_dir("on_err");
|
||||
let config = ManagerConfig {
|
||||
object_store: test_util::new_object_store(&dir),
|
||||
max_retry_times: 3,
|
||||
retry_delay: Duration::from_millis(500),
|
||||
};
|
||||
let manager = LocalManager::new(config);
|
||||
let state_store = Arc::new(ObjectStateStore::new(test_util::new_object_store(&dir)));
|
||||
let manager = LocalManager::new(config, state_store);
|
||||
|
||||
#[derive(Debug)]
|
||||
struct MockProcedure {
|
||||
|
||||
@@ -26,7 +26,7 @@ use crate::error::{Result, ToJsonSnafu};
|
||||
pub(crate) use crate::store::state_store::{ObjectStateStore, StateStoreRef};
|
||||
use crate::{BoxedProcedure, ProcedureId};
|
||||
|
||||
mod state_store;
|
||||
pub mod state_store;
|
||||
|
||||
/// Serialized data of a procedure.
|
||||
#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
|
||||
|
||||
@@ -17,21 +17,23 @@ use std::sync::Arc;
|
||||
|
||||
use async_stream::try_stream;
|
||||
use async_trait::async_trait;
|
||||
use common_error::ext::PlainError;
|
||||
use common_error::prelude::{BoxedError, StatusCode};
|
||||
use futures::{Stream, StreamExt};
|
||||
use object_store::{EntryMode, Metakey, ObjectStore};
|
||||
use snafu::ResultExt;
|
||||
|
||||
use crate::error::{DeleteStateSnafu, Error, ListStateSnafu, PutStateSnafu, Result};
|
||||
use crate::error::{DeleteStateSnafu, ListStateSnafu, PutStateSnafu, Result};
|
||||
|
||||
/// Key value from state store.
|
||||
type KeyValue = (String, Vec<u8>);
|
||||
pub type KeyValue = (String, Vec<u8>);
|
||||
|
||||
/// Stream that yields [KeyValue].
|
||||
type KeyValueStream = Pin<Box<dyn Stream<Item = Result<KeyValue>> + Send>>;
|
||||
pub type KeyValueStream = Pin<Box<dyn Stream<Item = Result<KeyValue>> + Send>>;
|
||||
|
||||
/// Storage layer for persisting procedure's state.
|
||||
#[async_trait]
|
||||
pub(crate) trait StateStore: Send + Sync {
|
||||
pub trait StateStore: Send + Sync {
|
||||
/// Puts `key` and `value` into the store.
|
||||
async fn put(&self, key: &str, value: Vec<u8>) -> Result<()>;
|
||||
|
||||
@@ -51,13 +53,13 @@ pub(crate) type StateStoreRef = Arc<dyn StateStore>;
|
||||
|
||||
/// [StateStore] based on [ObjectStore].
|
||||
#[derive(Debug)]
|
||||
pub(crate) struct ObjectStateStore {
|
||||
pub struct ObjectStateStore {
|
||||
store: ObjectStore,
|
||||
}
|
||||
|
||||
impl ObjectStateStore {
|
||||
/// Returns a new [ObjectStateStore] with specific `store`.
|
||||
pub(crate) fn new(store: ObjectStore) -> ObjectStateStore {
|
||||
pub fn new(store: ObjectStore) -> ObjectStateStore {
|
||||
ObjectStateStore { store }
|
||||
}
|
||||
}
|
||||
@@ -68,31 +70,65 @@ impl StateStore for ObjectStateStore {
|
||||
self.store
|
||||
.write(key, value)
|
||||
.await
|
||||
.map_err(|e| {
|
||||
BoxedError::new(PlainError::new(
|
||||
e.to_string(),
|
||||
StatusCode::StorageUnavailable,
|
||||
))
|
||||
})
|
||||
.context(PutStateSnafu { key })
|
||||
}
|
||||
|
||||
async fn walk_top_down(&self, path: &str) -> Result<KeyValueStream> {
|
||||
let path_string = path.to_string();
|
||||
|
||||
let mut lister = self.store.scan(path).await.map_err(|e| Error::ListState {
|
||||
path: path_string.clone(),
|
||||
source: e,
|
||||
})?;
|
||||
let mut lister = self
|
||||
.store
|
||||
.scan(path)
|
||||
.await
|
||||
.map_err(|e| {
|
||||
BoxedError::new(PlainError::new(
|
||||
e.to_string(),
|
||||
StatusCode::StorageUnavailable,
|
||||
))
|
||||
})
|
||||
.with_context(|_| ListStateSnafu {
|
||||
path: path_string.clone(),
|
||||
})?;
|
||||
|
||||
let store = self.store.clone();
|
||||
|
||||
let stream = try_stream!({
|
||||
while let Some(res) = lister.next().await {
|
||||
let entry = res.context(ListStateSnafu { path: &path_string })?;
|
||||
let entry = res
|
||||
.map_err(|e| {
|
||||
BoxedError::new(PlainError::new(
|
||||
e.to_string(),
|
||||
StatusCode::StorageUnavailable,
|
||||
))
|
||||
})
|
||||
.context(ListStateSnafu { path: &path_string })?;
|
||||
let key = entry.path();
|
||||
let metadata = store
|
||||
.metadata(&entry, Metakey::Mode)
|
||||
.await
|
||||
.map_err(|e| {
|
||||
BoxedError::new(PlainError::new(
|
||||
e.to_string(),
|
||||
StatusCode::StorageUnavailable,
|
||||
))
|
||||
})
|
||||
.context(ListStateSnafu { path: key })?;
|
||||
if let EntryMode::FILE = metadata.mode() {
|
||||
let value = store
|
||||
.read(key)
|
||||
.await
|
||||
.map_err(|e| {
|
||||
BoxedError::new(PlainError::new(
|
||||
e.to_string(),
|
||||
StatusCode::StorageUnavailable,
|
||||
))
|
||||
})
|
||||
.context(ListStateSnafu { path: key })?;
|
||||
yield (key.to_string(), value);
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user