fix(lock): prevent stale distributed object locks (#2633)

This commit is contained in:
weisd
2026-04-22 10:12:33 +08:00
committed by GitHub
parent 3ac1d2ab0b
commit a0f1bb4ff0
7 changed files with 456 additions and 133 deletions
@@ -120,11 +120,6 @@ impl LockClient for GrpcLockClient {
.map_err(|e| LockError::internal(e.to_string()))?
.into_inner();
// Check for explicit error first
if let Some(error_info) = resp.error_info {
return Err(LockError::internal(error_info));
}
// Check if the lock acquisition was successful
if resp.success {
Ok(LockResponse::success(
@@ -134,7 +129,8 @@ impl LockClient for GrpcLockClient {
} else {
// Lock acquisition failed
Ok(LockResponse::failure(
"Lock acquisition failed on remote server".to_string(),
resp.error_info
.unwrap_or_else(|| "Lock acquisition failed on remote server".to_string()),
std::time::Duration::ZERO,
))
}
+32 -15
View File
@@ -50,6 +50,18 @@ fn lock_result_from_error(error: impl Into<String>) -> GenerallyLockResult {
}
}
fn lock_result_from_release(lock_id: &rustfs_lock::LockId, success: bool) -> GenerallyLockResult {
if success {
GenerallyLockResult {
success: true,
error_info: None,
lock_info: None,
}
} else {
lock_result_from_error(format!("lock not found for release: {lock_id}"))
}
}
/// Minimal NodeService implementation that only supports Lock RPCs
/// Used for testing distributed lock scenarios with real gRPC
#[derive(Debug)]
@@ -102,7 +114,7 @@ impl NodeService for MinimalLockNodeService {
let lock_info_json = result.lock_info.as_ref().and_then(|info| serde_json::to_string(info).ok());
Ok(Response::new(GenerallyLockResponse {
success: result.success,
error_info: None,
error_info: result.error,
lock_info: lock_info_json,
}))
}
@@ -131,11 +143,14 @@ impl NodeService for MinimalLockNodeService {
};
match self.lock_client.release(&args.lock_id).await {
Ok(success) => Ok(Response::new(GenerallyLockResponse {
success,
error_info: None,
lock_info: None,
})),
Ok(success) => {
let result = lock_result_from_release(&args.lock_id, success);
Ok(Response::new(GenerallyLockResponse {
success: result.success,
error_info: result.error_info,
lock_info: None,
}))
}
Err(err) => Ok(Response::new(GenerallyLockResponse {
success: false,
error_info: Some(format!(
@@ -161,11 +176,14 @@ impl NodeService for MinimalLockNodeService {
};
match self.lock_client.force_release(&args.lock_id).await {
Ok(success) => Ok(Response::new(GenerallyLockResponse {
success,
error_info: None,
lock_info: None,
})),
Ok(success) => {
let result = lock_result_from_release(&args.lock_id, success);
Ok(Response::new(GenerallyLockResponse {
success: result.success,
error_info: result.error_info,
lock_info: None,
}))
}
Err(err) => Ok(Response::new(GenerallyLockResponse {
success: false,
error_info: Some(format!(
@@ -271,10 +289,9 @@ impl NodeService for MinimalLockNodeService {
Ok(batch_results) => {
for (result_idx, success) in batch_results.into_iter().enumerate() {
if let Some(request_idx) = valid_indices.get(result_idx) {
results[*request_idx] = GenerallyLockResult {
success,
error_info: None,
lock_info: None,
results[*request_idx] = match lock_ids.get(result_idx) {
Some(lock_id) => lock_result_from_release(lock_id, success),
None => lock_result_from_error(format!("unlock response index out of range: {result_idx}")),
};
}
}
+35
View File
@@ -275,6 +275,41 @@ async fn test_grpc_lock_client_batch_acquire_and_release() {
handle.abort();
}
#[tokio::test]
async fn test_grpc_lock_client_uses_request_lock_id_and_reports_missing_unlock() {
let manager = Arc::new(GlobalLockManager::new());
let local_client: Arc<dyn rustfs_lock::LockClient> = Arc::new(LocalClient::with_manager(manager));
let (addr, handle) = spawn_lock_server(local_client).await.expect("Failed to spawn server");
tokio::time::sleep(Duration::from_millis(100)).await;
let grpc_client = GrpcLockClient::new(addr);
let request = LockRequest::new(test_resource(), LockType::Exclusive, "owner-a").with_acquire_timeout(Duration::from_secs(2));
let response = grpc_client.acquire_lock(&request).await.expect("gRPC acquire should succeed");
let lock_info = response.lock_info.expect("gRPC acquire should include lock info");
assert_eq!(lock_info.id, request.lock_id);
assert!(
grpc_client
.release(&request.lock_id)
.await
.expect("gRPC release should succeed"),
"release should find the request lock id"
);
let missing_release = grpc_client
.release(&request.lock_id)
.await
.expect_err("second release should report missing lock");
assert!(
missing_release.to_string().contains("lock not found for release"),
"missing release should preserve server error, got: {missing_release}"
);
handle.abort();
}
#[tokio::test]
async fn test_distributed_lock_4_nodes_grpc_read_write_quorum_split_with_two_failed_nodes() {
let manager1 = Arc::new(GlobalLockManager::new());