fix(admin): serve data usage endpoints from scanner snapshot instead of live listing (#4980)

This commit is contained in:
Zhengchao An
2026-07-18 00:09:02 +08:00
committed by GitHub
parent accd312465
commit 4589148f48
11 changed files with 370 additions and 133 deletions
+4 -2
View File
@@ -258,12 +258,14 @@ pub mod config {
pub mod data_usage {
pub use crate::data_usage::{
DATA_USAGE_CACHE_NAME, apply_bucket_usage_memory_overlay, init_compression_total_memory_from_backend,
load_compression_total_from_memory, load_data_usage_from_backend, record_bucket_delete_marker_memory,
DATA_USAGE_CACHE_NAME, apply_bucket_usage_memory_overlay, compute_bucket_usage,
init_compression_total_memory_from_backend, live_bucket_usage_computations, load_compression_total_from_memory,
load_data_usage_from_backend, load_data_usage_from_backend_cached, record_bucket_delete_marker_memory,
record_bucket_object_delete_memory, record_bucket_object_version_write_memory, record_bucket_object_write_memory,
record_bucket_object_write_unknown_previous_memory, record_compression_total_memory,
refresh_bucket_usage_from_object_layer, refresh_versioned_bucket_usage_from_object_layer,
remove_bucket_usage_from_backend, replace_bucket_usage_memory_from_info, store_compression_total_in_backend,
store_data_usage_in_backend,
};
}
+72 -1
View File
@@ -41,7 +41,10 @@ use rustfs_utils::path::SLASH_SEPARATOR;
use std::{
collections::{HashMap, HashSet, hash_map::Entry},
future::Future,
sync::{Arc, LazyLock, OnceLock},
sync::{
Arc, LazyLock, OnceLock,
atomic::{AtomicU64, Ordering},
},
time::{Duration, SystemTime},
};
use tokio::fs;
@@ -76,6 +79,28 @@ static USAGE_MEMORY_CACHE: OnceLock<UsageMemoryCache> = OnceLock::new();
static USAGE_CACHE_UPDATING: OnceLock<CacheUpdating> = OnceLock::new();
static LIVE_BUCKET_USAGE_CACHE: OnceLock<LiveBucketUsageCache> = OnceLock::new();
/// Cached copy of the last persisted data usage snapshot, served to admin
/// endpoints for up to `DATA_USAGE_CACHE_TTL_SECS` between backend reads.
#[derive(Debug, Clone)]
struct CachedDataUsageSnapshot {
info: DataUsageInfo,
loaded_at: SystemTime,
}
type DataUsageSnapshotCache = Arc<RwLock<Option<CachedDataUsageSnapshot>>>;
static DATA_USAGE_SNAPSHOT_CACHE: OnceLock<DataUsageSnapshotCache> = OnceLock::new();
// Always-on revert detector for rustfs/backlog#1306: one relaxed increment per
// full-bucket version listing is negligible and lets tests prove that admin
// request paths never trigger live listings.
static LIVE_BUCKET_USAGE_COMPUTATIONS: AtomicU64 = AtomicU64::new(0);
/// Number of live full-bucket usage computations performed by this process.
pub fn live_bucket_usage_computations() -> u64 {
LIVE_BUCKET_USAGE_COMPUTATIONS.load(Ordering::Relaxed)
}
/// Deferred persist thresholds for compression totals: persist after this many
/// operations recorded, but no more often than the min interval.
const COMPRESSION_PERSIST_BATCH_SIZE: u64 = 100;
@@ -115,6 +140,10 @@ fn cache_updating() -> &'static CacheUpdating {
USAGE_CACHE_UPDATING.get_or_init(|| Arc::new(RwLock::new(false)))
}
fn data_usage_snapshot_cache() -> &'static DataUsageSnapshotCache {
DATA_USAGE_SNAPSHOT_CACHE.get_or_init(|| Arc::new(RwLock::new(None)))
}
fn live_bucket_usage_cache() -> &'static LiveBucketUsageCache {
LIVE_BUCKET_USAGE_CACHE.get_or_init(|| {
moka::future::Cache::builder()
@@ -193,6 +222,12 @@ async fn save_data_usage_in_backend(data_usage_info: DataUsageInfo, store: Arc<E
.await
.map_err(Error::other)?;
// Invalidate the cached snapshot so readers observe the new save on their
// next request instead of waiting out the remaining TTL. The next cached
// read reloads through `load_data_usage_from_backend`, keeping its
// backward-compatibility post-processing.
*data_usage_snapshot_cache().write().await = None;
Ok(())
}
@@ -346,6 +381,41 @@ pub async fn load_data_usage_from_backend(store: Arc<ECStore>) -> Result<DataUsa
Ok(data_usage_info)
}
/// Load the persisted data usage snapshot through a small in-process cache.
///
/// Admin read endpoints call this on every request; the cache bounds backend
/// reads (and the associated JSON parse and INFO log) to once per
/// `DATA_USAGE_CACHE_TTL_SECS` per process. `save_data_usage_in_backend`
/// invalidates the cache so a fresh scanner save is visible immediately.
pub async fn load_data_usage_from_backend_cached(store: Arc<ECStore>) -> Result<DataUsageInfo, Error> {
let ttl = Duration::from_secs(DATA_USAGE_CACHE_TTL_SECS);
{
let cache = data_usage_snapshot_cache().read().await;
if let Some(cached) = cache.as_ref()
&& SystemTime::now().duration_since(cached.loaded_at).unwrap_or_default() < ttl
{
return Ok(cached.info.clone());
}
}
// Re-check under the write lock so concurrent expirations trigger a single
// backend read instead of a stampede.
let mut cache = data_usage_snapshot_cache().write().await;
if let Some(cached) = cache.as_ref()
&& SystemTime::now().duration_since(cached.loaded_at).unwrap_or_default() < ttl
{
return Ok(cached.info.clone());
}
let info = load_data_usage_from_backend(store).await?;
*cache = Some(CachedDataUsageSnapshot {
info: info.clone(),
loaded_at: SystemTime::now(),
});
Ok(info)
}
/// Aggregate usage information from local disk snapshots.
fn merge_snapshot(aggregated: &mut DataUsageInfo, mut snapshot: LocalUsageSnapshot, latest_update: &mut Option<SystemTime>) {
if let Some(update) = snapshot.last_update
@@ -548,6 +618,7 @@ impl BucketUsageAccumulator {
type UsageVersionPage = StorageListObjectVersionsInfo<ObjectInfo>;
pub async fn compute_bucket_usage(store: Arc<ECStore>, bucket_name: &str) -> Result<BucketUsageInfo, Error> {
LIVE_BUCKET_USAGE_COMPUTATIONS.fetch_add(1, Ordering::Relaxed);
let bucket = bucket_name.to_string();
compute_bucket_usage_with_pages(bucket_name, move |marker, version_marker| {
let store = Arc::clone(&store);
+56 -12
View File
@@ -406,10 +406,17 @@ pub struct AccountInfo {
#[derive(Debug, Serialize, Deserialize, Default)]
pub struct BucketAccessInfo {
pub name: String,
pub size: u64,
pub objects: u64,
pub object_sizes_histogram: HashMap<String, u64>,
pub object_versions_histogram: HashMap<String, u64>,
// Usage stats are absent (not zero) when no scanner snapshot covers the
// bucket yet, so clients can distinguish "unknown" from "empty"
// (rustfs/backlog#1306).
#[serde(default, skip_serializing_if = "Option::is_none")]
pub size: Option<u64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub objects: Option<u64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub object_sizes_histogram: Option<HashMap<String, u64>>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub object_versions_histogram: Option<HashMap<String, u64>>,
pub details: Option<BucketDetails>,
pub prefix_usage: HashMap<String, u64>,
#[serde(rename = "expiration", with = "time::serde::rfc3339::option")]
@@ -727,6 +734,43 @@ mod tests {
use time::OffsetDateTime;
use time::macros::datetime;
/// Wire pin (rustfs/backlog#1306): usage stats without a scanner snapshot
/// must be omitted from the JSON, not serialized as zeros.
#[test]
fn bucket_access_info_omits_absent_usage_stats() {
let info = BucketAccessInfo {
name: "no-snapshot".to_string(),
..Default::default()
};
let value = serde_json::to_value(&info).unwrap();
let obj = value.as_object().unwrap();
assert!(!obj.contains_key("size"));
assert!(!obj.contains_key("objects"));
assert!(!obj.contains_key("object_sizes_histogram"));
assert!(!obj.contains_key("object_versions_histogram"));
}
/// Wire pin (rustfs/backlog#1306): populated usage stats keep their
/// existing snake_case keys and numeric values.
#[test]
fn bucket_access_info_serializes_present_usage_stats() {
let info = BucketAccessInfo {
name: "snapshot".to_string(),
size: Some(1024),
objects: Some(7),
object_sizes_histogram: Some(HashMap::from([("1MiB-10MiB".to_string(), 7)])),
object_versions_histogram: Some(HashMap::from([("SINGLE_VERSION".to_string(), 7)])),
..Default::default()
};
let value = serde_json::to_value(&info).unwrap();
assert_eq!(value["size"], 1024);
assert_eq!(value["objects"], 7);
assert_eq!(value["object_sizes_histogram"]["1MiB-10MiB"], 7);
assert_eq!(value["object_versions_histogram"]["SINGLE_VERSION"], 7);
}
#[test]
fn test_account_status_try_from_invalid() {
let result = AccountStatus::try_from("invalid");
@@ -1029,10 +1073,10 @@ mod tests {
let bucket_info = BucketAccessInfo {
name: "test-bucket".to_string(),
size: 6000000,
objects: 150,
object_sizes_histogram: sizes_histogram,
object_versions_histogram: versions_histogram,
size: Some(6000000),
objects: Some(150),
object_sizes_histogram: Some(sizes_histogram),
object_versions_histogram: Some(versions_histogram),
details: Some(BucketDetails {
versioning: true,
versioning_suspended: false,
@@ -1049,10 +1093,10 @@ mod tests {
};
assert_eq!(bucket_info.name, "test-bucket");
assert_eq!(bucket_info.size, 6000000);
assert_eq!(bucket_info.objects, 150);
assert_eq!(bucket_info.object_sizes_histogram.len(), 2);
assert_eq!(bucket_info.object_versions_histogram.len(), 2);
assert_eq!(bucket_info.size, Some(6000000));
assert_eq!(bucket_info.objects, Some(150));
assert_eq!(bucket_info.object_sizes_histogram.as_ref().map(HashMap::len), Some(2));
assert_eq!(bucket_info.object_versions_histogram.as_ref().map(HashMap::len), Some(2));
assert!(bucket_info.details.is_some());
assert_eq!(bucket_info.prefix_usage.len(), 2);
assert!(bucket_info.created.is_some());