gsadmin/rustfs
1
0
Fork 0
mirror of https://github.com/rustfs/rustfs.git synced 2026-08-16 18:08:21 +00:00
Code Issues Packages Projects Releases Wiki Activity

perf(ecstore): reduce inline PUT commit overhead (#6033)

Browse Source 
* perf(metrics): attribute PUT stage costs

Co-Authored-By: heihutu <heihutu@gmail.com>

* perf(ecstore): move PUT metadata during shuffle

Co-Authored-By: heihutu <heihutu@gmail.com>

* perf(s3): reuse PUT object lock state

Co-Authored-By: heihutu <heihutu@gmail.com>

* perf(ecstore): trim PUT metadata fanout clones

Build per-disk PUT metadata only for committed writer slots, move the response metadata out of the fanout vector, and preserve fresh FileInfo shuffle semantics.

Co-Authored-By: heihutu <heihutu@gmail.com>

* perf(metrics): make PUT stage attribution opt-in

Co-Authored-By: heihutu <heihutu@gmail.com>

* perf(ecstore): commit inline PUT shards directly

Co-Authored-By: heihutu <heihutu@gmail.com>

* perf(ecstore): streamline rename staging cleanup

Use the directory-specific removal operation for rename_data staging parents. This avoids a guaranteed failed file-removal probe on Unix-like hosts and lets Windows remove the empty directory directly while preserving best-effort non-empty handling.

Co-Authored-By: heihutu <heihutu@gmail.com>

* test(ecstore): cover inline PUT rename failures

Cache the detailed stage metrics gate once per PUT and exercise exact-quorum and quorum-minus-one failures after inline shard encoding.

Co-Authored-By: heihutu <heihutu@gmail.com>

---------

Co-authored-by: heihutu <heihutu@gmail.com>
This commit is contained in:
houseme
2026-08-13 10:04:20 +08:00
committed by GitHub
parent 846517625b
commit 3a0dbccc2e
9 changed files with 785 additions and 108 deletions
Download Patch File Download Diff File Expand all files Collapse all files
crates/ecstore/src/disk/local.rs
+46 -3
View File
@@ -9174,7 +9174,7 @@ impl DiskAPI for LocalDisk {
if let Some(src_file_path_parent) = src_file_path.parent() {
if src_volume != super::RUSTFS_META_MULTIPART_BUCKET {
let _ = remove_std(src_file_path_parent);
let _ = std::fs::remove_dir(src_file_path_parent);
} else {
let _ = self
.delete_file(&dst_volume_dir, &src_file_path_parent.to_path_buf(), true, false)
@@ -9499,7 +9499,7 @@ impl DiskAPI for LocalDisk {
if let Some(ref cleanup) = cleanup_path {
let _ = self.delete_file(&dst_volume_dir, cleanup, true, false).await;
} else if let Some(parent) = src_file_path.parent() {
let _ = remove_std(parent);
let _ = std::fs::remove_dir(parent);
}
// Heal reuses a version's `data_dir` and lands the rebuilt shard on
@@ -12439,6 +12439,10 @@ mod test {
.join(RUSTFS_META_TMP_BUCKET)
.join(tmp_object)
.join(new_data_dir.to_string());
let tmp_parent = tmp_data_dir
.parent()
.expect("tmp data dir should have a parent")
.to_path_buf();
fs::create_dir_all(&tmp_data_dir)
.await
.expect("new tmp data dir should be created");
@@ -12450,6 +12454,10 @@ mod test {
disk.rename_data(RUSTFS_META_TMP_BUCKET, tmp_object, new_fi, bucket, object)
.await
.expect("rename_data should commit");
assert!(
!tmp_parent.exists(),
"successful non-inline commit should remove the empty staging parent"
);
// The tmp xl.meta write point uses SyncMode::FileOnly: its parent dir
// ({tmp}/{tmp_object}) must not be fsynced.
@@ -12654,6 +12662,9 @@ mod test {
let tmp_object = "tmp-new-inline";
ensure_test_volume(&disk, bucket).await;
ensure_test_volume(&disk, RUSTFS_META_TMP_BUCKET).await;
let tmp_parent = disk
.get_object_path(RUSTFS_META_TMP_BUCKET, tmp_object)
.expect("tmp parent should resolve");
let _mode = durability_mode_override::set(DurabilityMode::Strict);
let version_id = Uuid::parse_str("99999999-9999-9999-9999-999999999999").expect("version id should parse");
@@ -12662,6 +12673,7 @@ mod test {
disk.rename_data(RUSTFS_META_TMP_BUCKET, tmp_object, new_fi, bucket, object)
.await
.expect("inline rename_data should commit the new object");
assert!(!tmp_parent.exists(), "successful inline commit should remove the empty staging parent");
let bucket_dir = disk.get_bucket_path(bucket).expect("bucket path should resolve");
let prefix_dir = disk.get_object_path(bucket, "prefix").expect("prefix path should resolve");
@@ -12685,6 +12697,34 @@ mod test {
);
}
#[tokio::test]
async fn rename_data_inline_preserves_non_empty_staging_parent() {
use tempfile::tempdir;
let dir = tempdir().expect("temp dir should be created");
let endpoint = Endpoint::try_from(dir.path().to_str().expect("temp dir should be utf8")).expect("endpoint should parse");
let disk = LocalDisk::new(&endpoint, false).await.expect("local disk should be created");
let bucket = "inline-staging-sentinel-bucket";
let object = "inline-object";
let tmp_object = "inline-stage-with-sentinel";
ensure_test_volume(&disk, bucket).await;
ensure_test_volume(&disk, RUSTFS_META_TMP_BUCKET).await;
let tmp_parent = disk
.get_object_path(RUSTFS_META_TMP_BUCKET, tmp_object)
.expect("tmp parent should resolve");
fs::create_dir_all(&tmp_parent).await.expect("tmp parent should be created");
let sentinel = tmp_parent.join("sentinel");
fs::write(&sentinel, b"keep").await.expect("sentinel should be written");
let fi = test_file_info(object, Uuid::new_v4(), None, Some(Bytes::from_static(b"inline-payload")));
disk.rename_data(RUSTFS_META_TMP_BUCKET, tmp_object, fi, bucket, object)
.await
.expect("non-empty staging cleanup must not negate the committed object");
assert_eq!(fs::read(&sentinel).await.expect("sentinel should remain"), b"keep");
}
#[cfg(unix)]
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
#[allow(clippy::await_holding_lock)]
@@ -12859,7 +12899,10 @@ mod test {
.expect("non-inline rename_data should commit");
assert!(!replacement_dir.exists(), "the destination object directory must not be replaced");
assert!(staging_parent.exists(), "the guarded staging parent must retain its identity");
assert!(
!staging_parent.exists(),
"successful commit should remove the empty staging parent after releasing its guard"
);
assert!(
!replacement_staging_parent.exists(),
"the staging parent must not be replaced between data and metadata publication"