feat: implement heal subsystem for automatic data repair

- Add heal module with core types (HealType, HealRequest, HealTask)
- Implement HealManager for task scheduling and execution
- Add HealStorageAPI trait and ECStoreHealStorage implementation
- Integrate heal capabilities into scanner for automatic repair
- Support multiple heal types: object, bucket, disk, metadata, MRF, EC decode
- Add progress tracking and event system for heal operations
- Merge heal and scanner error types for unified error handling
- Include comprehensive logging and metrics for heal operations

Signed-off-by: junxiang Mu <1948535941@qq.com>
This commit is contained in:
junxiang Mu
2025-07-10 18:11:42 +08:00
parent 0aff736efd
commit 4fb3d187d0
10 changed files with 1774 additions and 12 deletions
+50 -4
View File
@@ -14,8 +14,10 @@
use thiserror::Error;
/// RustFS AHM/Heal/Scanner 统一错误类型
#[derive(Debug, Error)]
pub enum Error {
// 通用
#[error("I/O error: {0}")]
Io(#[from] std::io::Error),
@@ -25,21 +27,65 @@ pub enum Error {
#[error("Configuration error: {0}")]
Config(String),
#[error("Heal configuration error: {message}")]
ConfigurationError { message: String },
#[error("Other error: {0}")]
Other(String),
#[error(transparent)]
Anyhow(#[from] anyhow::Error),
// Scanner相关
#[error("Scanner error: {0}")]
Scanner(String),
#[error("Metrics error: {0}")]
Metrics(String),
#[error(transparent)]
Other(#[from] anyhow::Error),
// Heal相关
#[error("Heal task not found: {task_id}")]
TaskNotFound { task_id: String },
#[error("Heal task already exists: {task_id}")]
TaskAlreadyExists { task_id: String },
#[error("Heal manager is not running")]
ManagerNotRunning,
#[error("Heal task execution failed: {message}")]
TaskExecutionFailed { message: String },
#[error("Invalid heal type: {heal_type}")]
InvalidHealType { heal_type: String },
#[error("Heal task cancelled")]
TaskCancelled,
#[error("Heal task timeout")]
TaskTimeout,
#[error("Heal event processing failed: {message}")]
EventProcessingFailed { message: String },
#[error("Heal progress tracking failed: {message}")]
ProgressTrackingFailed { message: String },
}
pub type Result<T, E = Error> = std::result::Result<T, E>;
// Implement conversion from ahm::Error to std::io::Error for use in main.rs
impl Error {
pub fn other<E>(error: E) -> Self
where
E: Into<Box<dyn std::error::Error + Send + Sync>>,
{
Error::Other(error.into().to_string())
}
}
// 可选:实现与 std::io::Error 的互转
impl From<Error> for std::io::Error {
fn from(err: Error) -> Self {
std::io::Error::other(err)
}
}
}