Compare commits

...

6 Commits

Author SHA1 Message Date
Zhengchao An a82daa7b71 Merge branch 'main' into codex/fix-5166-metacache-timeout 2026-08-01 10:44:38 +08:00
houseme 4b6b6f14bd feat(hotpath): use mimalloc in inner counting allocator (#5523)
* feat(hotpath): use mimalloc in counting allocator

* fix(hotpath): adapt mimalloc for allocation counting

---------

Co-authored-by: overtrue <anzhengchao@gmail.com>
2026-08-01 10:43:52 +08:00
overtrue 7312c99853 test(ilm): stabilize async scope conflict coverage 2026-08-01 10:33:27 +08:00
Zhengchao An 9080ea8ea0 test(object-lock): cover unretained version cleanup (#5504)
Co-authored-by: cxymds <cxymds@gmail.com>
2026-08-01 10:14:32 +08:00
cxymds f5c4627057 Merge branch 'main' into codex/fix-5166-metacache-timeout 2026-08-01 08:05:00 +08:00
overtrue aa1a8c1b3f fix(scanner): remove total timeout from heal walks 2026-07-31 10:57:32 +08:00
4 changed files with 275 additions and 115 deletions
@@ -26,6 +26,7 @@
use super::common::*;
use aws_sdk_s3::Client;
use aws_sdk_s3::error::ProvideErrorMetadata;
use aws_sdk_s3::primitives::{ByteStream, DateTimeFormat};
use aws_sdk_s3::types::{
CompletedMultipartUpload, CompletedPart, Delete, MetadataDirective, ObjectIdentifier, ObjectLockLegalHoldStatus,
@@ -2120,6 +2121,127 @@ async fn test_multipart_default_retention_fixed_at_create() {
// Versioning Auto-Enable Tests
// ============================================================================
#[tokio::test]
#[serial]
async fn test_unretained_object_lock_object_delete_and_bucket_cleanup() {
init_logging();
info!("🧪 Test: Unretained Object Lock object delete and bucket cleanup (Issue #5339)");
let mut env = ObjectLockTestEnvironment::new()
.await
.expect("failed to create Object Lock test environment");
env.start_rustfs().await.expect("failed to start RustFS");
let bucket = "test-object-lock-delete-cleanup";
let key = "unretained-object";
env.create_object_lock_bucket(bucket)
.await
.expect("failed to create Object Lock bucket");
let client = env.s3_client();
let put_response = client
.put_object()
.bucket(bucket)
.key(key)
.body(ByteStream::from_static(b"unretained data"))
.send()
.await
.expect("failed to upload unretained object");
let object_version_id = put_response
.version_id()
.expect("Object Lock buckets must create versioned objects")
.to_string();
let delete_response = client
.delete_object()
.bucket(bucket)
.key(key)
.send()
.await
.expect("failed to create delete marker");
assert_eq!(delete_response.delete_marker(), Some(true));
let delete_marker_version_id = delete_response
.version_id()
.expect("Deleting without a version ID must create a delete marker")
.to_string();
let get_error = client
.get_object()
.bucket(bucket)
.key(key)
.send()
.await
.expect_err("GET must not return an object hidden by a delete marker");
assert_eq!(get_error.raw_response().map(|response| response.status().as_u16()), Some(404));
assert_eq!(get_error.as_service_error().and_then(|error| error.code()), Some("NoSuchKey"));
let listed_objects = client
.list_objects_v2()
.bucket(bucket)
.send()
.await
.expect("failed to list current objects");
assert!(
listed_objects.contents().iter().all(|object| object.key() != Some(key)),
"ListObjectsV2 must hide objects whose latest version is a delete marker"
);
let listed_versions = client
.list_object_versions()
.bucket(bucket)
.send()
.await
.expect("failed to list object versions");
assert!(
listed_versions
.versions()
.iter()
.any(|version| version.key() == Some(key) && version.version_id() == Some(object_version_id.as_str())),
"The data version must remain until it is explicitly deleted"
);
assert!(
listed_versions
.delete_markers()
.iter()
.any(|marker| marker.key() == Some(key) && marker.version_id() == Some(delete_marker_version_id.as_str())),
"ListObjectVersions must expose the delete marker"
);
client
.delete_object()
.bucket(bucket)
.key(key)
.version_id(object_version_id)
.send()
.await
.expect("failed to delete the data version");
client
.delete_object()
.bucket(bucket)
.key(key)
.version_id(delete_marker_version_id)
.send()
.await
.expect("failed to delete the delete marker");
let remaining_versions = client
.list_object_versions()
.bucket(bucket)
.send()
.await
.expect("failed to list versions after cleanup");
assert!(remaining_versions.versions().is_empty());
assert!(remaining_versions.delete_markers().is_empty());
client
.delete_bucket()
.bucket(bucket)
.send()
.await
.expect("Deleting every version must remove xl.meta so the bucket can be deleted normally");
}
#[tokio::test]
#[serial]
async fn test_versioning_auto_enabled_with_object_lock() {
+18 -46
View File
@@ -382,7 +382,6 @@ struct ManualTransitionRunReport {
skipped_delete_marker: u64,
skipped_directory: u64,
skipped_replication: u64,
skipped_already_transitioned: u64,
skipped_already_in_flight: u64,
skipped_queue_full: u64,
skipped_queue_closed: u64,
@@ -408,41 +407,6 @@ fn assert_completed_or_in_flight_partial(state: &str, report: &ManualTransitionR
}
}
fn assert_conflict_winner_report(state: &str, report: &ManualTransitionRunReport, expected_objects: u64, context: &str) {
assert_completed_or_in_flight_partial(state, report, context);
if report.skipped_already_in_flight > 0 {
assert!(
report.scanned <= expected_objects,
"{context}: scanned more objects than the conflict scope contains: {report:#?}"
);
assert!(
report.eligible <= expected_objects,
"{context}: marked more objects eligible than the conflict scope contains: {report:#?}"
);
assert_eq!(
report.enqueued + report.skipped_already_in_flight,
report.eligible,
"{context}: partial in-flight accounting must cover every eligible object: {report:#?}"
);
} else {
assert_eq!(report.scanned, expected_objects, "{context}: {report:#?}");
assert_eq!(
report.eligible + report.skipped_already_transitioned,
expected_objects,
"{context}: {report:#?}"
);
assert_eq!(
report.enqueued + report.skipped_already_in_flight,
expected_objects,
"{context}: {report:#?}"
);
}
assert_eq!(
report.transition_completed, report.enqueued,
"{context}: winner must wait for all queued transitions: {report:#?}"
);
}
#[derive(Debug, Deserialize)]
struct ManualTransitionQueueSnapshot {
queue_capacity: u64,
@@ -1276,8 +1240,15 @@ async fn test_manual_transition_async_scope_conflicts_report_active_job() -> Tes
cold_client.create_bucket().bucket(TIER_BUCKET).send().await?;
let mut hot = RustFSTestEnvironment::new().await?;
hot.start_rustfs_server_with_env(vec![], &[("RUSTFS_SCANNER_ENABLED", "false"), ("RUSTFS_SCANNER_CYCLE", "3600")])
.await?;
hot.start_rustfs_server_with_env(
vec![],
&[
("RUSTFS_SCANNER_ENABLED", "false"),
("RUSTFS_SCANNER_CYCLE", "3600"),
(MANUAL_TRANSITION_CANCEL_BARRIER_ENV, "1"),
],
)
.await?;
let hot_client = hot.create_s3_client();
add_rustfs_tier(&hot, &cold).await?;
@@ -1347,20 +1318,21 @@ async fn test_manual_transition_async_scope_conflicts_report_active_job() -> Tes
assert_eq!(conflict.cancel_endpoint, status_endpoint);
assert!(!conflict.scope_key.is_empty());
manual_transition_job_cancel(&hot, cancel_endpoint).await?;
let terminal = wait_for_manual_transition_job_terminal(&hot, status_endpoint, MANUAL_ASYNC_CONFLICT_TERMINAL_TIMEOUT).await?;
assert_eq!(terminal.job_id, job_id);
assert_eq!(terminal.status, "cancelled", "terminal conflict winner response: {terminal:#?}");
assert!(!terminal.report.dry_run);
assert_eq!(terminal.report.bucket, MANUAL_ASYNC_CONFLICT_BUCKET);
assert_eq!(terminal.report.prefix, accepted.report.prefix);
assert_conflict_winner_report(
&terminal.status,
&terminal.report,
MANUAL_ASYNC_CONFLICT_OBJECTS as u64,
"terminal conflict winner response",
assert!(terminal.report.cancelled, "terminal conflict winner response: {terminal:#?}");
assert_eq!(terminal.report.scanned, 0, "terminal conflict winner response: {terminal:#?}");
assert_eq!(terminal.report.enqueued, 0, "terminal conflict winner response: {terminal:#?}");
assert_eq!(
terminal.report.transition_completed, 0,
"terminal conflict winner response: {terminal:#?}"
);
assert_eq!(terminal.report.dry_run_eligible, 0, "terminal conflict winner response: {terminal:#?}");
assert_eq!(terminal.report.transition_failed, 0, "terminal conflict winner response: {terminal:#?}");
assert_eq!(terminal.report.tier_failure, 0, "terminal conflict winner response: {terminal:#?}");
let after_remote_count = cold_tier_object_count(&cold_client).await?;
assert!(after_remote_count >= before_remote_count);
assert!(after_remote_count <= before_remote_count + MANUAL_ASYNC_CONFLICT_OBJECTS);
+99 -68
View File
@@ -75,7 +75,7 @@ const DATA_SCANNER_COMPACT_LEAST_OBJECT: usize = 500;
const DATA_SCANNER_COMPACT_AT_CHILDREN: usize = 10000;
const DATA_SCANNER_COMPACT_AT_FOLDERS: usize = DATA_SCANNER_COMPACT_AT_CHILDREN / 4;
const DATA_SCANNER_FORCE_COMPACT_AT_FOLDERS: usize = 250_000;
const SCANNER_LIST_PATH_RAW_TIMEOUT: Duration = Duration::from_secs(60);
const SCANNER_LIST_PATH_RAW_STALL_TIMEOUT: Duration = Duration::from_secs(60);
const SCANNER_ENTRY_PROGRESS_BATCH: u64 = 32;
const SCANNER_ENTRY_PROGRESS_INTERVAL: Duration = Duration::from_secs(30);
const DEFAULT_HEAL_OBJECT_SELECT_PROB: u32 = 1024;
@@ -100,6 +100,21 @@ static SCANNER_INLINE_HEAL_WARN_ONCE: Once = Once::new();
static SCANNER_INLINE_HEAL_METRICS_ONCE: Once = Once::new();
static SCANNER_ALERT_METRICS_ONCE: Once = Once::new();
#[cfg(test)]
type ListPathRawTimeoutSnapshot = (bool, Option<Duration>, Option<Duration>);
fn scanner_abandoned_child_list_options() -> ListPathRawOptions {
// A complete heal walk scales with bucket size and may legitimately take
// longer than a fixed wall-clock budget. Keep the total duration unbounded;
// Retain the scanner's per-read stall budget and keep cancellation controlled
// by the scanner cycle token.
ListPathRawOptions {
skip_walkdir_total_timeout: true,
walkdir_stall_timeout: Some(SCANNER_LIST_PATH_RAW_STALL_TIMEOUT),
..Default::default()
}
}
pub fn data_usage_update_dir_cycles() -> u32 {
rustfs_utils::get_env_u32(ENV_DATA_USAGE_UPDATE_DIR_CYCLES, DATA_USAGE_UPDATE_DIR_CYCLES)
}
@@ -1281,6 +1296,8 @@ pub struct FolderScanner {
skip_heal: Arc<std::sync::atomic::AtomicBool>,
local_disk: Arc<Disk>,
pending_heals_changed: bool,
#[cfg(test)]
list_path_raw_options_observer: Option<mpsc::UnboundedSender<ListPathRawTimeoutSnapshot>>,
}
impl FolderScanner {
@@ -2410,75 +2427,79 @@ impl FolderScanner {
let bucket_clone = bucket.clone();
let prefix_clone = prefix.clone();
let child_ctx_clone = child_ctx.clone();
#[cfg(test)]
let list_path_raw_options_observer = self.list_path_raw_options_observer.clone();
tokio::spawn(async move {
if let Err(e) = list_path_raw(
child_ctx_clone.clone(),
ListPathRawOptions {
disks,
bucket: bucket_clone.clone(),
path: prefix_clone.clone(),
recursive: true,
report_not_found: true,
min_disks: disks_quorum,
walkdir_timeout: Some(SCANNER_LIST_PATH_RAW_TIMEOUT),
walkdir_stall_timeout: Some(SCANNER_LIST_PATH_RAW_TIMEOUT),
agreed: Some(Box::new(move |entry: MetaCacheEntry| {
let entry_name = entry.name.clone();
let agreed_tx = agreed_tx.clone();
Box::pin(async move {
if let Err(e) = agreed_tx.send(entry_name).await {
error!(
target: "rustfs::scanner::folder",
event = EVENT_SCANNER_FOLDER_STATE,
component = LOG_COMPONENT_SCANNER,
subsystem = LOG_SUBSYSTEM_FOLDER,
entry = %entry.name,
state = "list_path_agreed_send_failed",
error = %e,
"Scanner list_path_raw agreed callback failed"
);
}
})
})),
partial: Some(Box::new(move |entries: MetaCacheEntries, _: &[Option<DiskError>]| {
let partial_tx = partial_tx.clone();
Box::pin(async move {
if let Err(e) = partial_tx.send(entries).await {
error!(
target: "rustfs::scanner::folder",
event = EVENT_SCANNER_FOLDER_STATE,
component = LOG_COMPONENT_SCANNER,
subsystem = LOG_SUBSYSTEM_FOLDER,
state = "list_path_partial_send_failed",
error = %e,
"Scanner list_path_raw partial callback failed"
);
}
})
})),
finished: Some(Box::new(move |errs: &[Option<DiskError>]| {
let finished_tx = finished_tx.clone();
let errs_clone = errs.to_vec();
Box::pin(async move {
if let Err(e) = finished_tx.send(errs_clone).await {
error!(
target: "rustfs::scanner::folder",
event = EVENT_SCANNER_FOLDER_STATE,
component = LOG_COMPONENT_SCANNER,
subsystem = LOG_SUBSYSTEM_FOLDER,
state = "list_path_finished_send_failed",
error = %e,
"Scanner list_path_raw finished callback failed"
);
}
})
})),
..Default::default()
},
)
.await
{
let options = ListPathRawOptions {
disks,
bucket: bucket_clone.clone(),
path: prefix_clone.clone(),
recursive: true,
report_not_found: true,
min_disks: disks_quorum,
agreed: Some(Box::new(move |entry: MetaCacheEntry| {
let entry_name = entry.name.clone();
let agreed_tx = agreed_tx.clone();
Box::pin(async move {
if let Err(e) = agreed_tx.send(entry_name).await {
error!(
target: "rustfs::scanner::folder",
event = EVENT_SCANNER_FOLDER_STATE,
component = LOG_COMPONENT_SCANNER,
subsystem = LOG_SUBSYSTEM_FOLDER,
entry = %entry.name,
state = "list_path_agreed_send_failed",
error = %e,
"Scanner list_path_raw agreed callback failed"
);
}
})
})),
partial: Some(Box::new(move |entries: MetaCacheEntries, _: &[Option<DiskError>]| {
let partial_tx = partial_tx.clone();
Box::pin(async move {
if let Err(e) = partial_tx.send(entries).await {
error!(
target: "rustfs::scanner::folder",
event = EVENT_SCANNER_FOLDER_STATE,
component = LOG_COMPONENT_SCANNER,
subsystem = LOG_SUBSYSTEM_FOLDER,
state = "list_path_partial_send_failed",
error = %e,
"Scanner list_path_raw partial callback failed"
);
}
})
})),
finished: Some(Box::new(move |errs: &[Option<DiskError>]| {
let finished_tx = finished_tx.clone();
let errs_clone = errs.to_vec();
Box::pin(async move {
if let Err(e) = finished_tx.send(errs_clone).await {
error!(
target: "rustfs::scanner::folder",
event = EVENT_SCANNER_FOLDER_STATE,
component = LOG_COMPONENT_SCANNER,
subsystem = LOG_SUBSYSTEM_FOLDER,
state = "list_path_finished_send_failed",
error = %e,
"Scanner list_path_raw finished callback failed"
);
}
})
})),
..scanner_abandoned_child_list_options()
};
#[cfg(test)]
if let Some(observer) = list_path_raw_options_observer {
let _ = observer.send((
options.skip_walkdir_total_timeout,
options.walkdir_timeout,
options.walkdir_stall_timeout,
));
}
if let Err(e) = list_path_raw(child_ctx_clone.clone(), options).await {
if is_missing_path_disk_error(&e) {
debug!(
target: "rustfs::scanner::folder",
@@ -2820,6 +2841,8 @@ pub async fn scan_data_folder(
skip_heal,
local_disk,
pending_heals_changed: false,
#[cfg(test)]
list_path_raw_options_observer: None,
};
// Check if context is cancelled
@@ -3026,6 +3049,7 @@ mod tests {
skip_heal: Arc::new(AtomicBool::new(false)),
local_disk: disk,
pending_heals_changed: false,
list_path_raw_options_observer: None,
};
(scanner, temp_dir)
@@ -4210,6 +4234,8 @@ mod tests {
scanner.heal_object_select = 1;
scanner.disks = disks;
scanner.disks_quorum = 2;
let (options_tx, mut options_rx) = mpsc::unbounded_channel();
scanner.list_path_raw_options_observer = Some(options_tx);
scanner.old_cache.replace(
&format!("{bucket}/{object}"),
bucket,
@@ -4231,6 +4257,11 @@ mod tests {
.expect("scan_folder should not hang after list_path_raw finishes")
.expect("scan_folder should finish successfully");
let observed_options = tokio::time::timeout(Duration::from_secs(1), options_rx.recv())
.await
.expect("abandoned-child listing options should be observed promptly")
.expect("abandoned-child listing options channel should remain open");
assert_eq!(observed_options, (true, None, Some(SCANNER_LIST_PATH_RAW_STALL_TIMEOUT)));
let root = scanner
.new_cache
.checked_flatten(bucket)
+36 -1
View File
@@ -12,9 +12,32 @@
// See the License for the specific language governing permissions and
// limitations under the License.
#[cfg(all(feature = "hotpath", feature = "hotpath-alloc"))]
use std::alloc::{GlobalAlloc, Layout};
#[cfg(all(feature = "hotpath", feature = "hotpath-alloc"))]
#[derive(Default)]
struct DefaultMiMalloc;
#[cfg(all(feature = "hotpath", feature = "hotpath-alloc"))]
// SAFETY: allocation and deallocation are forwarded unchanged to MiMalloc, so
// MiMalloc's GlobalAlloc guarantees apply to every returned pointer and layout.
#[allow(unsafe_code)]
unsafe impl GlobalAlloc for DefaultMiMalloc {
unsafe fn alloc(&self, layout: Layout) -> *mut u8 {
// SAFETY: the caller upholds GlobalAlloc's contract for layout.
unsafe { mimalloc::MiMalloc.alloc(layout) }
}
unsafe fn dealloc(&self, ptr: *mut u8, layout: Layout) {
// SAFETY: ptr and layout came from this allocator and are forwarded unchanged.
unsafe { mimalloc::MiMalloc.dealloc(ptr, layout) }
}
}
#[cfg(all(feature = "hotpath", feature = "hotpath-alloc"))]
#[global_allocator]
static GLOBAL: hotpath::CountingAllocator = hotpath::CountingAllocator::new();
static GLOBAL: hotpath::CountingAllocator<DefaultMiMalloc> = hotpath::CountingAllocator::new();
#[cfg(not(all(feature = "hotpath", feature = "hotpath-alloc")))]
#[global_allocator]
@@ -25,3 +48,15 @@ fn main() {
rustfs::startup_entrypoint::run_process();
}
#[cfg(all(test, feature = "hotpath", feature = "hotpath-alloc"))]
mod tests {
#[test]
#[allow(unsafe_code)]
fn hotpath_allocator_uses_mimalloc() {
let allocation = Box::new([0_u8; 64]);
// SAFETY: the live Box pointer is valid to inspect for heap ownership.
assert!(unsafe { libmimalloc_sys::mi_is_in_heap_region(allocation.as_ptr().cast()) });
}
}