Compare commits

..

2 Commits

Author SHA1 Message Date
houseme 30706c60c9 Merge branch 'main' into overtrue/install-protocol-socket-oracle 2026-08-23 12:10:09 +08:00
overtrue 417fefc1bc ci: install protocol socket oracle 2026-08-23 04:03:05 +08:00
12 changed files with 123 additions and 1342 deletions
@@ -196,8 +196,11 @@ jobs:
cache-save-if: 'false'
install-build-packaging-tools: 'false'
- name: Verify protocol socket oracle
run: ss -tn state CLOSE-WAIT >/dev/null
- name: Install and verify protocol socket oracle
run: |
sudo apt-get update -qq
sudo apt-get install -y -qq iproute2
ss -tn state CLOSE-WAIT >/dev/null
# The suite owns fixed protocol ports and serializes its internal cases.
- name: Verify protocol e2e membership
+6 -3
View File
@@ -39,10 +39,11 @@ jobs:
env:
FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: "true"
steps:
- name: Checkout repository
- name: Checkout main branch
uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
with:
persist-credentials: false
ref: main
- name: Setup Rust environment
uses: ./.github/actions/setup
@@ -88,10 +89,11 @@ jobs:
# either casing.
NO_PROXY: 127.0.0.1,localhost
steps:
- name: Checkout repository
- name: Checkout main branch
uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
with:
persist-credentials: false
ref: main
- name: Setup Rust environment
uses: ./.github/actions/setup
@@ -176,10 +178,11 @@ jobs:
FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: "true"
NO_PROXY: 127.0.0.1,localhost
steps:
- name: Checkout repository
- name: Checkout main branch
uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
with:
persist-credentials: false
ref: main
- name: Setup Rust environment
uses: ./.github/actions/setup
+1 -89
View File
@@ -307,38 +307,6 @@ pub struct DiskUsageStatus {
pub snapshot_exists: bool,
}
/// A bounded reconciliation record for an object whose logical size could not
/// be trusted at the scanner boundary. The scanner persists these records in
/// its cache; keeping the model here avoids a second, incompatible accounting
/// representation in storage-facing crates.
#[derive(Debug, Default, Clone, Serialize, Deserialize, PartialEq, Eq)]
pub struct SizeReconciliationEntry {
/// Stable object/version identity key (not a metrics label).
pub key: String,
pub bucket: String,
pub object: String,
#[serde(default)]
pub version_id: Option<String>,
#[serde(default)]
pub generation: Option<String>,
/// Structured reason label; raw metadata values must never be stored here.
pub reason: String,
#[serde(default)]
pub physical_size: Option<u64>,
#[serde(default)]
pub first_seen: u64,
#[serde(default)]
pub attempts: u32,
}
/// Object scope refreshed by one scanner pass. Existing debts in this scope
/// are removed before the pass's unresolved records are inserted.
#[derive(Debug, Default, Clone, PartialEq, Eq)]
pub struct SizeReconciliationScope {
pub bucket: String,
pub object: String,
}
/// Size summary for a single object or group of objects
#[derive(Debug, Default, Clone)]
pub struct SizeSummary {
@@ -368,16 +336,6 @@ pub struct SizeSummary {
pub repl_target_stats: HashMap<String, ReplTargetSizeSummary>,
/// Per-tier accounting, keyed by storage class or remote tier name
pub tier_stats: HashMap<String, TierStats>,
/// Size-resolution debts observed while scanning this summary.
pub size_reconciliation: Vec<SizeReconciliationEntry>,
/// True when the per-object summary exceeded its bounded debt buffer.
/// Callers must retain prior ledger entries rather than treating the
/// partial list as a complete refresh.
pub size_reconciliation_truncated: bool,
/// Object scopes refreshed by this summary. They let the durable ledger
/// remove versions that resolved without allocating one key per healthy
/// version on the hot path.
pub reconciliation_scopes: Vec<SizeReconciliationScope>,
}
/// Replication target size summary
@@ -875,8 +833,7 @@ impl DataUsageEntry {
///
/// The canonical wire format is written by the hand-written map-encoded
/// `Serialize` on the scanner-side `DataUsageCacheInfo`
/// (`crates/scanner/src/data_usage_define.rs`), which carries the original 16
/// fields plus an optional reconciliation field.
/// (`crates/scanner/src/data_usage_define.rs`), which carries 16 fields.
/// This type decodes only the shared subset and is deliberately not
/// `Serialize`: a derived (array) encoding of this 6-field subset would
/// corrupt the cache for scanner readers, so no write path may exist here.
@@ -1820,51 +1777,6 @@ impl SizeSummary {
entry.pending_count = entry.pending_count.saturating_add(stats.pending_count);
entry.failed_count = entry.failed_count.saturating_add(stats.failed_count);
}
for entry in &other.size_reconciliation {
self.record_size_reconciliation(entry.clone());
}
self.size_reconciliation_truncated |= other.size_reconciliation_truncated;
for scope in &other.reconciliation_scopes {
self.record_reconciliation_scope(&scope.bucket, &scope.object);
}
}
/// Add one reconciliation debt, coalescing repeated observations in the
/// same object summary. The scanner cache applies its own larger bound.
pub fn record_size_reconciliation(&mut self, entry: SizeReconciliationEntry) {
const MAX_SUMMARY_RECONCILIATION_ENTRIES: usize = 1024;
if let Some(existing) = self.size_reconciliation.iter_mut().find(|value| value.key == entry.key) {
existing.reason = entry.reason;
existing.physical_size = entry.physical_size;
existing.generation = entry.generation;
existing.version_id = entry.version_id;
return;
}
if self.size_reconciliation.len() < MAX_SUMMARY_RECONCILIATION_ENTRIES {
self.size_reconciliation.push(entry);
} else {
self.size_reconciliation_truncated = true;
}
}
/// Mark one object scope as refreshed. Duplicate scopes are suppressed so
/// merging summaries remains bounded and deterministic.
pub fn record_reconciliation_scope(&mut self, bucket: &str, object: &str) {
if !self
.reconciliation_scopes
.iter()
.any(|scope| scope.bucket == bucket && scope.object == object)
{
if self.reconciliation_scopes.len() >= 1024 {
self.size_reconciliation_truncated = true;
return;
}
self.reconciliation_scopes.push(SizeReconciliationScope {
bucket: bucket.to_string(),
object: object.to_string(),
});
}
}
}
+6 -38
View File
@@ -784,24 +784,6 @@ pub(crate) fn create_deferred_bitrot_reader_with_stripe_handle(
///
/// # Returns
/// A Result containing the BitrotWriterWrapper or an error
/// Size hint handed to `DiskAPI::create_file` for a bitrot-wrapped shard.
///
/// A known length is grown by one checksum per shard so the on-disk file size
/// matches what the bitrot writer emits. A negative length is the
/// unknown-size sentinel (`HashReader::SIZE_PRESERVE_LAYER`, used by SSE and
/// compression) and must be preserved: `RemoteDisk::create_file` forwards it
/// in the `put_file_stream` query, and the receiver only treats `size > 0` as
/// a fixed body length when locating the authenticated trailer. Clamping it
/// to `0` would claim an empty body and misframe the stream. `0` stays `0`
/// because a genuinely empty object still means an empty body.
fn bitrot_create_file_size(length: i64, shard_size: usize, checksum_algo: &HashAlgorithm) -> i64 {
if length <= 0 {
return length;
}
let length = length as usize;
(length.div_ceil(shard_size) * checksum_algo.size() + length) as i64
}
pub async fn create_bitrot_writer(
is_inline_buffer: bool,
disk: Option<&DiskStore>,
@@ -814,7 +796,12 @@ pub async fn create_bitrot_writer(
let writer = if is_inline_buffer {
CustomWriter::new_inline_buffer()
} else if let Some(disk) = disk {
let length = bitrot_create_file_size(length, shard_size, &checksum_algo);
let length = if length > 0 {
let length = length as usize;
(length.div_ceil(shard_size) * checksum_algo.size() + length) as i64
} else {
0
};
let file = disk.create_file("", volume, path, length).await?;
#[cfg(feature = "hotpath")]
@@ -833,25 +820,6 @@ mod tests {
use rustfs_rio::ChunkReader;
use std::collections::VecDeque;
#[test]
fn bitrot_create_file_size_grows_known_length_by_checksums() {
// 10 bytes over 4-byte shards = 3 shards, each followed by a 32-byte hash.
assert_eq!(bitrot_create_file_size(10, 4, &HashAlgorithm::HighwayHash256), 10 + 3 * 32);
assert_eq!(bitrot_create_file_size(10, 4, &HashAlgorithm::None), 10);
}
#[test]
fn bitrot_create_file_size_keeps_empty_and_unknown_distinct() {
assert_eq!(bitrot_create_file_size(0, 4, &HashAlgorithm::HighwayHash256), 0);
// SSE/compression streams advertise SIZE_PRESERVE_LAYER (-1); the remote
// put_file_stream receiver relies on a non-positive size to parse the auth
// trailer from the stream tail, so the sentinel must survive untouched.
assert_eq!(
bitrot_create_file_size(rustfs_rio::HashReader::SIZE_PRESERVE_LAYER, 4, &HashAlgorithm::HighwayHash256),
rustfs_rio::HashReader::SIZE_PRESERVE_LAYER
);
}
struct TestChunkReader {
chunks: VecDeque<Bytes>,
}
+25 -86
View File
@@ -16,14 +16,14 @@
//!
//! `scripts/test/vault_ha_kms_live.sh` owns the official Vault containers and
//! kills the active node while this test continuously decrypts through a
//! surviving standby. KV2 and Transit must recover after the bounded circuit
//! interval, use a bounded number of attempts, and leave the circuit and
//! in-flight gauges at zero after a new leader is elected.
//! surviving standby. KV2 and Transit requests must remain successful, use a
//! bounded number of attempts, and leave the circuit and in-flight gauges at
//! zero after a new leader is elected.
use std::collections::HashMap;
use std::path::{Path, PathBuf};
use std::sync::Arc;
use std::sync::atomic::{AtomicBool, AtomicU64, Ordering};
use std::sync::{Arc, Mutex};
use std::time::Duration;
use metrics_util::MetricKind;
@@ -43,11 +43,6 @@ const OPERATION_ATTEMPTS: &str = "rustfs_kms_backend_operation_attempts";
const IN_FLIGHT: &str = "rustfs_kms_backend_in_flight";
const CIRCUIT_OPEN: &str = "rustfs_kms_backend_circuit_open";
const MAX_ATTEMPTS: u32 = 10;
const ATTEMPT_TIMEOUT: Duration = Duration::from_secs(2);
const HEALTHY_PROGRESS_TIMEOUT: Duration = Duration::from_secs(20);
// The circuit remains open for 30s after five failed attempts.
const POST_FAILOVER_PROGRESS_TIMEOUT: Duration = Duration::from_secs(35);
const FAILOVER_ERROR_POLL_INTERVAL: Duration = Duration::from_millis(100);
type MetricEntry = (
metrics_util::CompositeKey,
@@ -69,7 +64,7 @@ fn config(backend: KmsBackend, backend_config: BackendConfig) -> KmsConfig {
backend,
backend_config,
allow_insecure_dev_defaults: true,
timeout: ATTEMPT_TIMEOUT,
timeout: Duration::from_secs(2),
retry_attempts: MAX_ATTEMPTS,
enable_cache: false,
..KmsConfig::default()
@@ -169,31 +164,14 @@ fn retryable_failures(snapshot: &[MetricEntry], operation: &str) -> u64 {
.sum()
}
async fn wait_for_count(
counter: &AtomicU64,
failure: &Mutex<Option<String>>,
minimum: u64,
description: &str,
timeout: Duration,
) {
tokio::time::timeout(timeout, async {
async fn wait_for_count(counter: &AtomicU64, minimum: u64, description: &str) {
tokio::time::timeout(Duration::from_secs(20), async {
while counter.load(Ordering::SeqCst) < minimum {
if let Some(error) = failure.lock().expect("decrypt failure lock poisoned").as_ref() {
panic!(
"{description} worker failed after {} successful decrypts: {error}",
counter.load(Ordering::SeqCst)
);
}
tokio::time::sleep(Duration::from_millis(25)).await;
}
})
.await
.unwrap_or_else(|_| {
panic!(
"timed out after {timeout:?} waiting for {description}: completed {}, expected {minimum}",
counter.load(Ordering::SeqCst)
)
});
.unwrap_or_else(|_| panic!("timed out waiting for {description}"));
}
async fn wait_for_file(path: &Path, description: &str) {
@@ -211,8 +189,7 @@ async fn decrypt_loop<B: KmsBackendTrait + Send + Sync + 'static>(
request: DecryptRequest,
expected: Vec<u8>,
completed: Arc<AtomicU64>,
allow_failover_errors: Arc<AtomicBool>,
failure: Arc<Mutex<Option<String>>>,
failed: Arc<AtomicBool>,
stop: CancellationToken,
) {
while !stop.is_cancelled() {
@@ -220,18 +197,8 @@ async fn decrypt_loop<B: KmsBackendTrait + Send + Sync + 'static>(
Ok(response) if response.plaintext == expected => {
completed.fetch_add(1, Ordering::SeqCst);
}
Ok(_) => {
*failure.lock().expect("decrypt failure lock poisoned") =
Some("decrypt returned unexpected plaintext".to_string());
return;
}
Err(rustfs_kms::KmsError::BackendError { .. } | rustfs_kms::KmsError::OperationTimedOut { .. })
if allow_failover_errors.load(Ordering::SeqCst) =>
{
tokio::time::sleep(FAILOVER_ERROR_POLL_INTERVAL).await;
}
Err(error) => {
*failure.lock().expect("decrypt failure lock poisoned") = Some(error.to_string());
Ok(_) | Err(_) => {
failed.store(true, Ordering::SeqCst);
return;
}
}
@@ -329,9 +296,7 @@ async fn exercise_failover(snapshotter: &Snapshotter) {
);
let stop = CancellationToken::new();
let allow_failover_errors = Arc::new(AtomicBool::new(false));
let kv2_failure = Arc::new(Mutex::new(None));
let transit_failure = Arc::new(Mutex::new(None));
let failed = Arc::new(AtomicBool::new(false));
let kv2_completed = Arc::new(AtomicU64::new(0));
let transit_completed = Arc::new(AtomicU64::new(0));
let kv2_worker = tokio::spawn(decrypt_loop(
@@ -339,8 +304,7 @@ async fn exercise_failover(snapshotter: &Snapshotter) {
kv2_request,
kv2_data_key.plaintext_key,
Arc::clone(&kv2_completed),
Arc::clone(&allow_failover_errors),
Arc::clone(&kv2_failure),
Arc::clone(&failed),
stop.clone(),
));
let transit_worker = tokio::spawn(decrypt_loop(
@@ -348,21 +312,12 @@ async fn exercise_failover(snapshotter: &Snapshotter) {
transit_request,
transit_data_key.plaintext_key,
Arc::clone(&transit_completed),
Arc::clone(&allow_failover_errors),
Arc::clone(&transit_failure),
Arc::clone(&failed),
stop.clone(),
));
wait_for_count(&kv2_completed, &kv2_failure, 2, "two healthy KV2 decrypts", HEALTHY_PROGRESS_TIMEOUT).await;
wait_for_count(
&transit_completed,
&transit_failure,
2,
"two healthy Transit decrypts",
HEALTHY_PROGRESS_TIMEOUT,
)
.await;
allow_failover_errors.store(true, Ordering::SeqCst);
wait_for_count(&kv2_completed, 2, "two healthy KV2 decrypts").await;
wait_for_count(&transit_completed, 2, "two healthy Transit decrypts").await;
std::fs::write(&marker, b"ready").expect("publish failover readiness marker");
wait_for_file(&elected, "the replacement Vault leader").await;
@@ -371,39 +326,18 @@ async fn exercise_failover(snapshotter: &Snapshotter) {
let kv2_after_election = kv2_completed.load(Ordering::SeqCst) + 2;
let transit_after_election = transit_completed.load(Ordering::SeqCst) + 2;
wait_for_count(
&kv2_completed,
&kv2_failure,
kv2_after_election,
"post-failover KV2 decrypts",
POST_FAILOVER_PROGRESS_TIMEOUT,
)
.await;
wait_for_count(
&transit_completed,
&transit_failure,
transit_after_election,
"post-failover Transit decrypts",
POST_FAILOVER_PROGRESS_TIMEOUT,
)
.await;
wait_for_count(&kv2_completed, kv2_after_election, "post-failover KV2 decrypts").await;
wait_for_count(&transit_completed, transit_after_election, "post-failover Transit decrypts").await;
stop.cancel();
kv2_worker.await.expect("KV2 decrypt worker must join");
transit_worker.await.expect("Transit decrypt worker must join");
assert!(
kv2_failure.lock().expect("KV2 failure lock poisoned").is_none(),
"no KV2 decrypt may fail or return different plaintext"
);
assert!(
transit_failure.lock().expect("Transit failure lock poisoned").is_none(),
"no Transit decrypt may fail or return different plaintext"
);
assert!(!failed.load(Ordering::SeqCst), "no decrypt may fail or return different plaintext");
}
#[test]
#[ignore = "requires a real three-node Vault Raft cluster; run scripts/test/vault_ha_kms_live.sh"]
fn vault_raft_leader_failure_recovers_kv2_and_transit_decrypts() {
fn vault_raft_leader_failure_preserves_kv2_and_transit_decrypts() {
let recorder = DebuggingRecorder::new();
let snapshotter = recorder.snapshotter();
metrics::with_local_recorder(&recorder, || {
@@ -415,6 +349,11 @@ fn vault_raft_leader_failure_recovers_kv2_and_transit_decrypts() {
});
let snapshot = snapshotter.snapshot().into_vec();
assert_eq!(
counter_value(&snapshot, OPERATIONS_TOTAL, &[("outcome", "circuit_open")]),
0,
"a bounded leader election must not open the circuit"
);
assert_eq!(
counter_value(&snapshot, OPERATIONS_TOTAL, &[("outcome", "budget_exhausted")]),
0,
+4 -33
View File
@@ -54,7 +54,6 @@ const ERR_LIFECYCLE_INVALID_EXPIRED_OBJECT_ALL_VERSIONS: &str =
"Days must be a positive integer and Date must not be specified inside Expiration with ExpiredObjectAllVersions";
const ERR_LIFECYCLE_INVALID_DEL_MARKER_EXPIRATION_DAYS: &str = "Days must be a positive integer with DelMarkerExpiration";
const ERR_LIFECYCLE_INVALID_RULE_ID_TOO_LONG: &str = "Rule ID must be at most 255 characters";
const ERR_LIFECYCLE_INVALID_RULE_ID_EMPTY: &str = "Rule ID must not be empty";
const ERR_LIFECYCLE_INVALID_RULE_STATUS: &str = "Rule status must be either Enabled or Disabled";
const ERR_LIFECYCLE_DEL_MARKER_WITH_TAGS: &str = "Rule with DelMarkerExpiration cannot have tags based filtering";
const ERR_LIFECYCLE_EXPIRED_OBJECT_DELETE_MARKER_WITH_TAGS: &str =
@@ -403,13 +402,10 @@ impl Lifecycle for BucketLifecycleConfiguration {
NoncurrentVersionTransitionOps::validate(transition)?;
}
}
if let Some(id) = &r.id {
if id.is_empty() {
return Err(std::io::Error::other(ERR_LIFECYCLE_INVALID_RULE_ID_EMPTY));
}
if id.len() > 255 {
return Err(std::io::Error::other(ERR_LIFECYCLE_INVALID_RULE_ID_TOO_LONG));
}
if let Some(id) = &r.id
&& id.len() > 255
{
return Err(std::io::Error::other(ERR_LIFECYCLE_INVALID_RULE_ID_TOO_LONG));
}
r.validate()?;
if let Some(object_lock_enabled) = lr.object_lock_enabled.as_ref()
@@ -3734,31 +3730,6 @@ mod tests {
.expect("empty prefix with filter should be valid");
}
#[tokio::test]
async fn validate_rejects_empty_rule_id() {
let lc = BucketLifecycleConfiguration {
expiry_updated_at: None,
rules: vec![LifecycleRule {
status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
expiration: Some(LifecycleExpiration {
days: Some(30),
..Default::default()
}),
abort_incomplete_multipart_upload: None,
del_marker_expiration: None,
filter: None,
id: Some(String::new()),
noncurrent_version_expiration: None,
noncurrent_version_transitions: None,
prefix: None,
transitions: None,
}],
};
let error = lc.validate(&ObjectLockConfiguration::default()).await.unwrap_err();
assert_eq!(error.to_string(), ERR_LIFECYCLE_INVALID_RULE_ID_EMPTY);
}
// --- TASK-004 tests: ExpiredObjectAllVersions ---
#[tokio::test]
+5 -50
View File
@@ -29,8 +29,7 @@ use rustfs_config::ENV_SCANNER_CACHE_SAVE_TIMEOUT_SECS;
pub use rustfs_data_usage::{
AllTierStats, BucketTargetUsageInfo, BucketUsageInfo, DATA_USAGE_OBJECT_NAME, DATA_USAGE_OBSERVED_OBJECT_NAME,
DataUsageEntry, DataUsageHash, DataUsageHashMap, DataUsageInfo, LEGACY_DATA_USAGE_OBJECT_NAME, PrefixUsageEntry,
PrefixUsageQuery, PrefixUsageSummary, ReplTargetSizeSummary, SizeReconciliationEntry, SizeReconciliationScope, SizeSummary,
TierStats, hash_path, prefix_usage_in_cache,
PrefixUsageQuery, PrefixUsageSummary, ReplTargetSizeSummary, SizeSummary, TierStats, hash_path, prefix_usage_in_cache,
};
use rustfs_utils::path::{SLASH_SEPARATOR, path_join_buf};
use tokio::time::{Duration, Instant, sleep, timeout};
@@ -193,10 +192,6 @@ const MAX_DATA_USAGE_CACHE_DEPTH: usize = 1024;
pub trait ScannerSizeSummaryExt {
/// Fold one object's contribution into the summary, including its tier.
fn actions_accounting(&mut self, oi: &ObjectInfo, size: i64, actual_size: i64);
/// Fold counters and physical tier usage for an object whose metadata is
/// valid but whose logical size is currently unavailable. Logical totals
/// stay unchanged.
fn actions_accounting_unknown(&mut self, oi: &ObjectInfo);
}
impl ScannerSizeSummaryExt for SizeSummary {
@@ -230,34 +225,6 @@ impl ScannerSizeSummaryExt for SizeSummary {
});
}
}
fn actions_accounting_unknown(&mut self, oi: &ObjectInfo) {
if oi.delete_marker {
self.delete_markers = self.delete_markers.saturating_add(1);
return;
}
if oi.version_id.is_some_and(|v| !v.is_nil()) {
self.versions = self.versions.saturating_add(1);
}
if oi.transitioned_object.free_version {
return;
}
let tier = if oi.transitioned_object.status == TRANSITION_COMPLETE {
oi.transitioned_object.tier.clone()
} else {
oi.storage_class.clone().unwrap_or_else(|| storageclass::STANDARD.to_string())
};
if let Some(tier_stats) = self.tier_stats.get_mut(&tier) {
*tier_stats = tier_stats.add(&TierStats {
total_size: u64::try_from(oi.size).unwrap_or(0),
num_versions: 1,
num_objects: u64::from(oi.is_latest),
});
}
}
}
// ===== Cache-related data structures =====
@@ -377,10 +344,6 @@ pub struct DataUsageCacheInfo {
pub scan_plan_digest: Option<DataUsageScanPlanDigest>,
#[serde(default)]
pub cache_key_format: u16,
/// Bounded durable debts for versions whose logical size was not trusted.
/// The map key is an identity key, never a user-controlled metric label.
#[serde(default)]
pub size_reconciliation: HashMap<String, SizeReconciliationEntry>,
}
impl Serialize for DataUsageCacheInfo {
@@ -390,8 +353,7 @@ impl Serialize for DataUsageCacheInfo {
{
// Keep this metadata map-encoded so older readers can ignore fields
// appended by newer scanner versions during rolling upgrades.
let field_count = 16 + usize::from(!self.size_reconciliation.is_empty());
let mut state = serializer.serialize_map(Some(field_count))?;
let mut state = serializer.serialize_map(Some(16))?;
state.serialize_entry("name", &self.name)?;
state.serialize_entry("next_cycle", &self.next_cycle)?;
state.serialize_entry("leader_epoch", &self.leader_epoch)?;
@@ -408,9 +370,6 @@ impl Serialize for DataUsageCacheInfo {
state.serialize_entry("snapshot_complete", &self.snapshot_complete)?;
state.serialize_entry("scan_plan_digest", &self.scan_plan_digest)?;
state.serialize_entry("cache_key_format", &self.cache_key_format)?;
if !self.size_reconciliation.is_empty() {
state.serialize_entry("size_reconciliation", &self.size_reconciliation)?;
}
state.end()
}
}
@@ -469,18 +428,14 @@ impl DataUsageCache {
self.checked_flatten(name).is_some()
});
if !reusable {
let (pending_heals, size_reconciliation) = if self.info.name == name {
(
std::mem::take(&mut self.info.pending_heals),
std::mem::take(&mut self.info.size_reconciliation),
)
let pending_heals = if self.info.name == name {
std::mem::take(&mut self.info.pending_heals)
} else {
(Vec::new(), HashMap::new())
Vec::new()
};
*self = Self::default();
self.info.name = name.to_string();
self.info.pending_heals = pending_heals;
self.info.size_reconciliation = size_reconciliation;
}
self.info.next_cycle = next_cycle;
@@ -673,34 +673,6 @@ fn size_summary_actions_accounting_accumulates_tier_stats() {
);
}
#[test]
fn size_summary_unknown_accounting_keeps_physical_tier_and_version_only() {
let mut summary = SizeSummary::default();
summary
.tier_stats
.insert(storageclass::STANDARD.to_string(), TierStats::default());
let object = ObjectInfo {
size: 12,
storage_class: Some(storageclass::STANDARD.to_string()),
version_id: Some(uuid::Uuid::new_v4()),
is_latest: true,
..Default::default()
};
summary.actions_accounting_unknown(&object);
assert_eq!(summary.total_size, 0, "unknown logical size must not become zero or physical bytes");
assert_eq!(summary.versions, 1);
assert_eq!(
summary.tier_stats.get(storageclass::STANDARD),
Some(&TierStats {
total_size: 12,
num_versions: 1,
num_objects: 1,
})
);
}
#[test]
fn test_data_usage_entry_merge_sums_failed_objects() {
let mut left = DataUsageEntry {
@@ -1107,16 +1079,6 @@ fn data_usage_cache_prepare_for_scan_preserves_pending_heal_only_progress() {
scan_plan_digest: Some(TEST_PLAN_DIGEST),
cache_key_format: DATA_USAGE_CACHE_KEY_FORMAT,
pending_heals: vec![pending_heal.clone()],
size_reconciliation: HashMap::from([(
"size-key".to_string(),
SizeReconciliationEntry {
key: "size-key".to_string(),
bucket: "bucket".to_string(),
object: "prefix/object".to_string(),
reason: "invalid_declared_size".to_string(),
..Default::default()
},
)]),
..Default::default()
},
..Default::default()
@@ -1126,7 +1088,6 @@ fn data_usage_cache_prepare_for_scan_preserves_pending_heal_only_progress() {
assert_eq!(outcome, DataUsageCachePrepareOutcome::Reused);
assert_eq!(cache.info.pending_heals, vec![pending_heal]);
assert!(cache.info.size_reconciliation.contains_key("size-key"));
assert!(cache.cache.is_empty());
assert!(!cache.info.snapshot_complete);
}
+5 -129
View File
@@ -20,9 +20,8 @@ use std::time::{Duration, Instant, SystemTime};
use crate::ReplTargetSizeSummary;
use crate::data_usage_define::{
DATA_USAGE_SCAN_CHECKPOINT_VERSION, DataUsageCache, DataUsageCacheInfo, DataUsageEntry, DataUsageHash, DataUsageHashMap,
DataUsageScanCheckpoint, DataUsageScanCheckpointReason, PendingScannerHeal, PendingScannerHealKind, ScannerSizeSummaryExt,
SizeReconciliationEntry, SizeSummary, hash_path,
DATA_USAGE_SCAN_CHECKPOINT_VERSION, DataUsageCache, DataUsageEntry, DataUsageHash, DataUsageHashMap, DataUsageScanCheckpoint,
DataUsageScanCheckpointReason, PendingScannerHeal, PendingScannerHealKind, ScannerSizeSummaryExt, SizeSummary, hash_path,
};
use crate::error::ScannerError;
use crate::runtime_config::{
@@ -98,9 +97,6 @@ const METRIC_SCANNER_EXCESS_FOLDERS_TOTAL: &str = "rustfs_scanner_excess_folders
const METRIC_SCANNER_PENDING_HEAL_PRUNE_TOTAL: &str = "rustfs_scanner_pending_heal_prune_total";
const METRIC_SCANNER_PENDING_HEAL_MALFORMED_TOTAL: &str = "rustfs_scanner_pending_heal_malformed_total";
const MAX_PENDING_SCANNER_HEAL_RETRIES_PER_BUCKET: usize = 128;
const MAX_SIZE_RECONCILIATION_ENTRIES_PER_BUCKET: usize = 10_000;
const MAX_SIZE_RECONCILIATION_BYTES_PER_BUCKET: usize = 8 * 1024 * 1024;
const MAX_SIZE_RECONCILIATION_AGE_SECS: u64 = 7 * 24 * 60 * 60;
// --- scanner excess alerts as S3 notification events (rustfs/backlog#1868) --
//
@@ -368,7 +364,7 @@ impl PendingScannerAccounting<'_> {
fn apply(self, size_summary: &mut SizeSummary, cumulative_size: &mut i64, queued: bool) {
let size = if queued { self.expired_size } else { self.retained_size };
size_summary.actions_accounting(self.object, size, self.retained_size);
*cumulative_size = cumulative_size.saturating_add(size);
*cumulative_size += size;
}
}
@@ -675,65 +671,10 @@ pub struct FolderScanner {
skip_heal: Arc<std::sync::atomic::AtomicBool>,
local_disk: Arc<Disk>,
pending_heals_changed: bool,
pending_size_reconciliation_keys: HashSet<String>,
pending_size_reconciliation_scopes: HashSet<String>,
pending_size_reconciliation_truncated: bool,
#[cfg(test)]
list_path_raw_options_observer: Option<mpsc::UnboundedSender<ListPathRawTimeoutSnapshot>>,
}
fn size_reconciliation_entry_bytes(entry: &SizeReconciliationEntry) -> usize {
entry.key.len()
+ entry.bucket.len()
+ entry.object.len()
+ entry.version_id.as_deref().map_or(0, str::len)
+ entry.generation.as_deref().map_or(0, str::len)
+ entry.reason.len()
+ std::mem::size_of::<u64>()
+ std::mem::size_of::<u32>()
}
fn size_reconciliation_scope_key(bucket: &str, object: &str) -> String {
format!("{}:{}|{}:{}", bucket.len(), bucket, object.len(), object)
}
fn prune_size_reconciliation(info: &mut DataUsageCacheInfo, now: u64) {
info.size_reconciliation.retain(|key, entry| {
if entry.first_seen == 0 || entry.first_seen > now {
entry.first_seen = now;
}
key == &entry.key
&& entry.key.len() <= 4096
&& entry.bucket.len() <= 512
&& entry.object.len() <= 512
&& entry.version_id.as_deref().is_none_or(|value| value.len() <= 64)
&& entry.generation.as_deref().is_none_or(|value| value.len() <= 64)
&& entry.reason.len() <= 64
&& now.saturating_sub(entry.first_seen) <= MAX_SIZE_RECONCILIATION_AGE_SECS
});
while info.size_reconciliation.len() > MAX_SIZE_RECONCILIATION_ENTRIES_PER_BUCKET
|| info
.size_reconciliation
.values()
.map(size_reconciliation_entry_bytes)
.sum::<usize>()
> MAX_SIZE_RECONCILIATION_BYTES_PER_BUCKET
{
let oldest = info
.size_reconciliation
.iter()
.min_by(|(left_key, left), (right_key, right)| {
left.first_seen.cmp(&right.first_seen).then_with(|| left_key.cmp(right_key))
})
.map(|(key, _)| key.clone());
let Some(oldest) = oldest else {
break;
};
info.size_reconciliation.remove(&oldest);
}
}
impl FolderScanner {
fn now_secs() -> u64 {
SystemTime::now()
@@ -807,60 +748,6 @@ impl FolderScanner {
}
}
/// Apply the per-object size-resolution ledger updates in one place. The
/// scanner cache is the durable boundary; both working copies are updated
/// so an incremental publication cannot lose a debt or its resolution.
fn apply_size_reconciliation(&mut self, summary: &SizeSummary) {
let now = Self::now_secs();
self.pending_size_reconciliation_keys
.extend(summary.size_reconciliation.iter().map(|entry| entry.key.clone()));
self.pending_size_reconciliation_scopes.extend(
summary
.reconciliation_scopes
.iter()
.map(|scope| size_reconciliation_scope_key(&scope.bucket, &scope.object)),
);
self.pending_size_reconciliation_truncated |= summary.size_reconciliation_truncated;
for info in [&mut self.new_cache.info, &mut self.update_cache.info] {
for incoming in &summary.size_reconciliation {
if let Some(existing) = info.size_reconciliation.get_mut(&incoming.key) {
existing.reason = incoming.reason.clone();
existing.physical_size = incoming.physical_size;
existing.generation = incoming.generation.clone();
existing.version_id = incoming.version_id.clone();
existing.attempts = existing.attempts.saturating_add(1);
continue;
}
if size_reconciliation_entry_bytes(incoming) > MAX_SIZE_RECONCILIATION_BYTES_PER_BUCKET {
continue;
}
let mut entry = incoming.clone();
entry.first_seen = now;
entry.attempts = 1;
info.size_reconciliation.insert(entry.key.clone(), entry);
}
}
}
fn finish_size_reconciliation_batch(&mut self) {
let now = Self::now_secs();
let current_keys = std::mem::take(&mut self.pending_size_reconciliation_keys);
let scopes = std::mem::take(&mut self.pending_size_reconciliation_scopes);
let truncated = std::mem::replace(&mut self.pending_size_reconciliation_truncated, false);
for info in [&mut self.new_cache.info, &mut self.update_cache.info] {
if !truncated {
info.size_reconciliation.retain(|key, entry| {
!scopes.contains(&size_reconciliation_scope_key(&entry.bucket, &entry.object)) || current_keys.contains(key)
});
}
prune_size_reconciliation(info, now);
}
}
fn record_scan_resume_hint(&mut self, folder: &str) {
self.new_cache.info.scan_resume_after = Some(folder.to_string());
self.update_cache.info.scan_resume_after = Some(folder.to_string());
@@ -1539,7 +1426,6 @@ impl FolderScanner {
abandoned_children.remove(&path_join_buf(&[&item.bucket, &item.object_path()]));
apply_scanner_size_summary(into, &sz);
self.apply_size_reconciliation(&sz);
into.objects += 1;
object_count += 1;
self.budget.record_object_scanned();
@@ -2219,7 +2105,6 @@ impl FolderScanner {
}
}
self.finish_size_reconciliation_batch();
done_folder();
let scanned_objects = u64::try_from(into.objects).unwrap_or(u64::MAX);
emit_scanner_folder_trace(&self.root, &folder.name, scanned_objects, trace_started_at, "completed");
@@ -2305,17 +2190,10 @@ pub async fn scan_data_folder(
skip_heal,
local_disk,
pending_heals_changed: false,
pending_size_reconciliation_keys: HashSet::new(),
pending_size_reconciliation_scopes: HashSet::new(),
pending_size_reconciliation_truncated: false,
#[cfg(test)]
list_path_raw_options_observer: None,
};
let now = FolderScanner::now_secs();
prune_size_reconciliation(&mut scanner.new_cache.info, now);
prune_size_reconciliation(&mut scanner.update_cache.info, now);
// Check if context is cancelled
if ctx.is_cancelled() {
return Err(ScannerError::Other("Operation cancelled".to_string()));
@@ -2339,9 +2217,7 @@ pub async fn scan_data_folder(
new_cache.force_compact(DATA_SCANNER_COMPACT_AT_CHILDREN);
new_cache.info.last_update = Some(SystemTime::now());
new_cache.info.next_cycle = cache.info.next_cycle;
let unresolved_objects = root.failed_objects > 0
|| !new_cache.info.failed_objects.is_empty()
|| !new_cache.info.size_reconciliation.is_empty();
let unresolved_objects = root.failed_objects > 0 || !new_cache.info.failed_objects.is_empty();
new_cache.info.snapshot_complete = !unresolved_objects;
let had_scan_checkpoint = cache.info.scan_checkpoint.is_some() || new_cache.info.scan_checkpoint.is_some();
new_cache.info.scan_resume_after = None;
@@ -2369,7 +2245,7 @@ pub async fn scan_data_folder(
if root_has_progress {
new_cache.replace_hashed(&root_hash, &None, &root);
}
if partial_cache_is_useful(&root, pending_heals_changed) || !new_cache.info.size_reconciliation.is_empty() {
if partial_cache_is_useful(&root, pending_heals_changed) {
if new_cache.root().is_some() {
new_cache.force_compact(DATA_SCANNER_COMPACT_AT_CHILDREN);
}
+65 -809
View File
@@ -13,7 +13,6 @@
// limitations under the License.
/// Per-object scan actions: ScannerItem, the get-size failure policy, and the heal/ILM admission helpers.
use super::*;
use sha2::{Digest as _, Sha256};
/// Cached folder information for scanning
#[derive(Clone, Debug)]
@@ -33,263 +32,6 @@ pub(super) enum GetSizeFailureAction {
HealMetadata { object: String },
}
#[derive(Clone, Copy, Debug, PartialEq, Eq)]
pub(super) enum SizeResolutionReason {
CompressedSizeUnknown,
InvalidPhysicalSize,
UnsupportedCompression,
InvalidObjectSize,
InvalidPartSize,
InvalidDeclaredSize,
SizeOverflowOrMismatch,
}
impl SizeResolutionReason {
fn as_str(self) -> &'static str {
match self {
Self::CompressedSizeUnknown => "compressed_size_unknown",
Self::InvalidPhysicalSize => "invalid_physical_size",
Self::UnsupportedCompression => "unsupported_compression",
Self::InvalidObjectSize => "invalid_object_size",
Self::InvalidPartSize => "invalid_part_size",
Self::InvalidDeclaredSize => "invalid_declared_size",
Self::SizeOverflowOrMismatch => "size_overflow_or_mismatch",
}
}
}
#[derive(Clone, Debug, PartialEq, Eq)]
pub(super) enum SizeResolution {
Known { logical: i64, physical: i64 },
Unknown { physical: i64, reason: SizeResolutionReason },
Corrupt { physical: i64, reason: SizeResolutionReason },
}
impl SizeResolution {
fn known_size(&self) -> Option<i64> {
match self {
Self::Known { logical, .. } => Some(*logical),
Self::Unknown { .. } | Self::Corrupt { .. } => None,
}
}
}
fn size_reconciliation_key(oi: &ObjectInfo, reason: SizeResolutionReason) -> String {
let version = oi
.version_id
.filter(|version| !version.is_nil())
.map(|version| version.to_string())
.unwrap_or_default();
let generation = oi
.data_dir
.filter(|generation| !generation.is_nil())
.map(|generation| generation.to_string())
.unwrap_or_default();
// Length-prefix each component so an object key containing the separator
// cannot alias another identity. S3 keys are bounded in normal operation;
// oversized persisted values use a digest so a corrupt metadata record
// cannot grow the ledger without bound.
fn component(value: &str) -> String {
const MAX_COMPONENT_LEN: usize = 512;
if value.len() <= MAX_COMPONENT_LEN {
return format!("{}:{}", value.len(), value);
}
let digest = Sha256::digest(value.as_bytes());
let digest = hex_simd::encode_to_string(digest, hex_simd::AsciiCase::Lower);
format!("hash:{}:{}", value.len(), digest)
}
format!(
"{}|{}|{}|{}|{}",
component(&oi.bucket),
component(&oi.name),
component(&version),
component(&generation),
component(reason.as_str())
)
}
pub(super) fn bounded_reconciliation_field(value: &str) -> String {
const MAX_FIELD_LEN: usize = 512;
if value.len() <= MAX_FIELD_LEN {
return value.to_string();
}
let digest = hex_simd::encode_to_string(Sha256::digest(value.as_bytes()), hex_simd::AsciiCase::Lower);
let prefix_len = MAX_FIELD_LEN - 65;
let prefix = value
.char_indices()
.take_while(|(offset, ch)| offset.saturating_add(ch.len_utf8()) <= prefix_len)
.map(|(_, ch)| ch)
.collect::<String>();
format!("{}~{}", prefix, digest)
}
fn record_size_resolution(summary: &mut SizeSummary, oi: &ObjectInfo, resolution: &SizeResolution) {
match resolution {
SizeResolution::Known { .. } => {}
SizeResolution::Unknown { physical, reason } | SizeResolution::Corrupt { physical, reason } => {
summary.record_size_reconciliation(SizeReconciliationEntry {
key: size_reconciliation_key(oi, *reason),
bucket: bounded_reconciliation_field(&oi.bucket),
object: bounded_reconciliation_field(&oi.name),
version_id: oi
.version_id
.filter(|version| !version.is_nil())
.map(|version| version.to_string()),
generation: oi
.data_dir
.filter(|generation| !generation.is_nil())
.map(|generation| generation.to_string()),
reason: reason.as_str().to_string(),
physical_size: u64::try_from(*physical).ok(),
first_seen: 0,
attempts: 0,
});
}
}
}
/// Resolve the size metadata once at the scanner trust boundary. A compressed
/// -1 sentinel is valid legacy metadata, but it cannot participate in normal
/// logical-size accounting or size-filtered lifecycle rules.
pub(super) fn resolve_size(oi: &ObjectInfo) -> SizeResolution {
let physical = oi.size;
if physical < 0 {
return SizeResolution::Corrupt {
physical,
reason: SizeResolutionReason::InvalidPhysicalSize,
};
}
let compressed = match oi.compression_read_plan() {
Ok((_, _, compressed)) => compressed,
Err(_) => {
return SizeResolution::Corrupt {
physical,
reason: SizeResolutionReason::UnsupportedCompression,
};
}
};
if oi.actual_size < -1 || (oi.actual_size == -1 && !compressed) {
return SizeResolution::Corrupt {
physical,
reason: SizeResolutionReason::InvalidObjectSize,
};
}
// Match ObjectInfo::get_actual_size: a positive in-memory value is the
// authoritative decoded size. Stale declared/part metadata must not turn
// an otherwise valid object into a false corruption report.
if oi.actual_size > 0 {
return SizeResolution::Known {
logical: oi.actual_size,
physical,
};
}
if oi
.parts
.iter()
.any(|part| part.actual_size < -1 || (part.actual_size < 0 && !compressed))
{
return SizeResolution::Corrupt {
physical,
reason: SizeResolutionReason::InvalidPartSize,
};
}
let declared = rustfs_utils::http::get_str(&oi.user_defined, rustfs_utils::http::SUFFIX_ACTUAL_SIZE);
let declared = match declared {
Some(value) if value.is_empty() => {
return SizeResolution::Corrupt {
physical,
reason: SizeResolutionReason::InvalidDeclaredSize,
};
}
Some(value) => match value.parse::<i64>() {
Ok(value) if value >= 0 => Some(value),
_ => {
return SizeResolution::Corrupt {
physical,
reason: SizeResolutionReason::InvalidDeclaredSize,
};
}
},
None => None,
};
let logical = match oi.get_actual_size() {
Ok(size) if size == -1 && compressed && declared.is_none() => {
return SizeResolution::Unknown {
physical,
reason: SizeResolutionReason::CompressedSizeUnknown,
};
}
Ok(size) if size >= 0 => size,
Ok(_) | Err(_) => {
return SizeResolution::Corrupt {
physical,
reason: SizeResolutionReason::SizeOverflowOrMismatch,
};
}
};
if compressed && logical == 0 && physical != 0 && oi.parts.is_empty() && declared.is_none() {
return SizeResolution::Corrupt {
physical,
reason: SizeResolutionReason::SizeOverflowOrMismatch,
};
}
SizeResolution::Known { logical, physical }
}
fn resolve_sizes(object_infos: &[ObjectInfo]) -> Vec<SizeResolution> {
object_infos.iter().map(resolve_size).collect()
}
fn lifecycle_rule_has_size_filter(lifecycle: &BucketLifecycleConfiguration, rule_id: &str) -> bool {
let filter_has_size = |filter: &s3s::dto::LifecycleRuleFilter| {
filter.object_size_greater_than.is_some()
|| filter.object_size_less_than.is_some()
|| filter
.and
.as_ref()
.is_some_and(|and| and.object_size_greater_than.is_some() || and.object_size_less_than.is_some())
};
lifecycle
.rules
.iter()
.find(|rule| {
if rule_id.is_empty() {
rule.id.as_deref().is_none_or(str::is_empty)
} else {
rule.id.as_deref() == Some(rule_id)
}
})
.and_then(|rule| rule.filter.as_ref())
.is_some_and(filter_has_size)
}
fn lifecycle_event_allowed(resolution: &SizeResolution, event: &Event, lifecycle: &BucketLifecycleConfiguration) -> bool {
match resolution {
// Missing or invalid logical size only defers actions whose selected
// rule actually depends on that size. Time/version-only actions retain
// their existing semantics, including intrinsic events without a rule ID.
SizeResolution::Unknown { .. } | SizeResolution::Corrupt { .. } => {
!lifecycle_rule_has_size_filter(lifecycle, &event.rule_id)
}
SizeResolution::Known { .. } => true,
}
}
/// A successful newer-noncurrent batch consumes both known and unresolved
/// versions from the retained-version alert count. The two accounting paths
/// are separate because only known sizes can contribute byte totals.
fn remaining_versions_after_queued_noncurrent(remaining_versions: usize, known_count: usize, unknown_count: usize) -> usize {
remaining_versions.saturating_sub(known_count.saturating_add(unknown_count))
}
/// How the corrupt-metadata branch records the repair after attempting an
/// MRF intent (backlog#1894 axis A).
#[derive(Debug, PartialEq, Eq)]
@@ -577,48 +319,34 @@ impl ScannerItem {
"Scanner lifecycle evaluation started"
);
let resolved_sizes = resolve_sizes(&object_infos);
if let Some(first) = object_infos.first() {
size_summary.record_reconciliation_scope(
&bounded_reconciliation_field(&first.bucket),
&bounded_reconciliation_field(&first.name),
);
}
for (oi, resolution) in object_infos.iter().zip(resolved_sizes.iter()) {
record_size_resolution(size_summary, oi, resolution);
}
let has_corrupt_size = resolved_sizes
.iter()
.any(|resolution| matches!(resolution, SizeResolution::Corrupt { .. }));
// `versioning_config` is resolved once per object by the caller
// (`get_size`) and handed in; only `prefix_enabled` is consulted here.
let Some(lifecycle) = self.lifecycle.clone() else {
let mut cumulative_size: i64 = 0;
for (oi, resolved_size) in object_infos.iter().zip(resolved_sizes.iter()) {
let accounting_size = match resolved_size {
SizeResolution::Known { logical, .. } => *logical,
// A valid compressed legacy sentinel has no logical size,
// but heal and replication still need to run. The
// physical size is only an input to those operations; it
// is not folded into the logical total below.
SizeResolution::Unknown { physical, .. } => {
self.heal_actions(oi, *physical, size_summary).await;
size_summary.actions_accounting_unknown(oi);
continue;
}
SizeResolution::Corrupt { .. } => {
size_summary.actions_accounting_unknown(oi);
let Some(lifecycle) = self.lifecycle.as_ref() else {
let mut cumulative_size = 0;
for oi in object_infos.iter() {
let actual_size = match oi.get_actual_size() {
Ok(size) => size,
Err(_) => {
warn!(
target: "rustfs::scanner::folder",
event = EVENT_SCANNER_LIFECYCLE_ACTION,
component = LOG_COMPONENT_SCANNER,
subsystem = LOG_SUBSYSTEM_LIFECYCLE,
bucket = %self.bucket,
object = %oi.name,
state = "size_lookup_failed",
"Scanner lifecycle action used fallback size"
);
continue;
}
};
let size = self.heal_actions(oi, accounting_size, size_summary).await;
let size = self.heal_actions(oi, actual_size, size_summary).await;
size_summary.actions_accounting(oi, size, accounting_size);
size_summary.actions_accounting(oi, size, actual_size);
cumulative_size = cumulative_size.saturating_add(size);
cumulative_size += size;
}
self.alert_excessive_versions(object_infos.len(), cumulative_size);
@@ -672,108 +400,25 @@ impl ScannerItem {
let mut to_delete_objs: Vec<ObjectToDelete> = Vec::new();
let mut noncurrent_events: Vec<Event> = Vec::new();
let mut noncurrent_accounting: Vec<PendingScannerAccounting<'_>> = Vec::new();
let mut noncurrent_unknown: Vec<&ObjectInfo> = Vec::new();
let mut cumulative_size = 0;
let mut remaining_versions = object_infos.len();
'eventLoop: {
for (i, event) in events.iter().enumerate() {
let oi = &object_infos[i];
let known_size = resolved_sizes[i].known_size();
if has_corrupt_size
&& matches!(
event.action,
IlmAction::DeleteAllVersionsAction | IlmAction::DelMarkerDeleteAllVersionsAction
)
{
// An all-version delete would also remove a corrupt
// sibling that could not be reconciled safely.
continue;
}
if !lifecycle_event_allowed(&resolved_sizes[i], event, &lifecycle) {
// An unknown logical size must not make an otherwise
// non-destructive scan disappear from heal/physical-tier
// accounting. Size-filtered or deferred events remain
// pending, so retain the version-only physical counters.
if let SizeResolution::Unknown { physical, .. } = &resolved_sizes[i] {
self.heal_actions(oi, *physical, size_summary).await;
size_summary.actions_accounting_unknown(oi);
}
continue;
}
let actual_size = match known_size {
Some(size) => size,
None => {
match event.action {
IlmAction::DeleteAction
| IlmAction::DeleteRestoredAction
| IlmAction::DeleteRestoredVersionAction
| IlmAction::DeleteAllVersionsAction
| IlmAction::DelMarkerDeleteAllVersionsAction => {
let done_ilm = Metrics::time_ilm(event.action);
let trace_started_at = trace_start_instant();
let queued = apply_expiry_rule(event, &LcEventSrc::Scanner, oi).await;
emit_scanner_ilm_action_trace(&self.bucket, &oi.name, event.action, 1, queued, trace_started_at);
if record_scanner_ilm_action_if_queued(global_metrics(), event.action, 1, queued) {
done_ilm(1)();
if event.action == IlmAction::DeleteAllVersionsAction
|| event.action == IlmAction::DelMarkerDeleteAllVersionsAction
{
remaining_versions = 0;
}
} else if matches!(
event.action,
IlmAction::DeleteAction
| IlmAction::DeleteRestoredAction
| IlmAction::DeleteRestoredVersionAction
) {
size_summary.actions_accounting_unknown(oi);
} else {
size_summary.actions_accounting_unknown(oi);
for (j, retained) in object_infos.iter().enumerate().skip(i + 1) {
match &resolved_sizes[j] {
SizeResolution::Known { logical, .. } => PendingScannerAccounting {
object: retained,
retained_size: *logical,
expired_size: 0,
}
.apply(size_summary, &mut cumulative_size, false),
SizeResolution::Unknown { .. } => {
size_summary.actions_accounting_unknown(retained);
}
SizeResolution::Corrupt { .. } => {}
}
}
}
}
IlmAction::DeleteVersionAction => {
if let Some(opt) = object_opts.get(i) {
to_delete_objs.push(ObjectToDelete {
object_name: opt.name.clone(),
version_id: opt.version_id,
..Default::default()
});
noncurrent_events.push(event.clone());
noncurrent_unknown.push(oi);
}
}
IlmAction::TransitionAction | IlmAction::TransitionVersionAction => {
let trace_started_at = trace_start_instant();
let queued = apply_transition_rule(event, &LcEventSrc::Scanner, oi).await;
emit_scanner_ilm_action_trace(&self.bucket, &oi.name, event.action, 1, queued, trace_started_at);
if record_scanner_ilm_action_if_queued(global_metrics(), event.action, 1, queued) {
let done_ilm = Metrics::time_ilm(event.action);
done_ilm(1)();
}
size_summary.actions_accounting_unknown(oi);
}
IlmAction::NoneAction | IlmAction::ActionCount => {
if let SizeResolution::Unknown { physical, .. } = &resolved_sizes[i] {
self.heal_actions(oi, *physical, size_summary).await;
}
size_summary.actions_accounting_unknown(oi);
}
}
continue;
let actual_size = match oi.get_actual_size() {
Ok(size) => size,
Err(_) => {
warn!(
target: "rustfs::scanner::folder",
event = EVENT_SCANNER_LIFECYCLE_ACTION,
component = LOG_COMPONENT_SCANNER,
subsystem = LOG_SUBSYSTEM_LIFECYCLE,
bucket = %self.bucket,
object = %oi.name,
state = "size_lookup_failed",
"Scanner lifecycle action used fallback size"
);
0
}
};
@@ -801,24 +446,36 @@ impl ScannerItem {
done_ilm(1)();
remaining_versions = 0;
} else {
if let Some(actual_size) = known_size {
PendingScannerAccounting {
object: oi,
retained_size: actual_size,
expired_size: 0,
}
.apply(size_summary, &mut cumulative_size, false);
for retained in object_infos.iter().skip(i + 1) {
let retained_size = match retained.get_actual_size() {
Ok(size) => size,
Err(_) => {
warn!(
target: "rustfs::scanner::folder",
event = EVENT_SCANNER_LIFECYCLE_ACTION,
component = LOG_COMPONENT_SCANNER,
subsystem = LOG_SUBSYSTEM_LIFECYCLE,
bucket = %self.bucket,
object = %retained.name,
state = "size_lookup_failed",
"Scanner lifecycle action used fallback size"
);
0
}
};
PendingScannerAccounting {
object: oi,
retained_size: actual_size,
object: retained,
retained_size,
expired_size: 0,
}
.apply(size_summary, &mut cumulative_size, false);
}
for (j, retained) in object_infos.iter().enumerate().skip(i + 1) {
if let Some(retained_size) = resolved_sizes[j].known_size() {
PendingScannerAccounting {
object: retained,
retained_size,
expired_size: 0,
}
.apply(size_summary, &mut cumulative_size, false);
}
}
}
break 'eventLoop;
}
@@ -854,13 +511,11 @@ impl ScannerItem {
version_id: opt.version_id,
..Default::default()
});
if let Some(actual_size) = known_size {
noncurrent_accounting.push(PendingScannerAccounting {
object: oi,
retained_size: actual_size,
expired_size: 0,
});
}
noncurrent_accounting.push(PendingScannerAccounting {
object: oi,
retained_size: actual_size,
expired_size: 0,
});
account_now = false;
}
noncurrent_events.push(event.clone());
@@ -893,7 +548,7 @@ impl ScannerItem {
if account_now {
size_summary.actions_accounting(oi, size, actual_size);
cumulative_size = cumulative_size.saturating_add(size);
cumulative_size += size;
}
}
}
@@ -921,20 +576,11 @@ impl ScannerItem {
}
if record_scanner_ilm_action_if_queued(global_metrics(), action, count, queued) {
done_ilm(count)();
remaining_versions = remaining_versions_after_queued_noncurrent(
remaining_versions,
noncurrent_accounting.len(),
noncurrent_unknown.len(),
);
remaining_versions = remaining_versions.saturating_sub(noncurrent_accounting.len());
}
for pending in noncurrent_accounting {
pending.apply(size_summary, &mut cumulative_size, queued);
}
if !queued {
for object in noncurrent_unknown {
size_summary.actions_accounting_unknown(object);
}
}
}
self.alert_excessive_versions(remaining_versions, cumulative_size);
}
@@ -1283,394 +929,4 @@ mod tests {
assert_eq!(item.object_name, "object");
assert_eq!(item.object_path(), "object");
}
#[test]
fn size_resolution_rejects_negative_overflow_and_unknown_compression() {
let compressed = |actual_size: i64, declared: Option<&str>| {
let mut user_defined = HashMap::new();
rustfs_utils::http::insert_str(&mut user_defined, rustfs_utils::http::SUFFIX_COMPRESSION, "zstd".to_string());
if let Some(declared) = declared {
rustfs_utils::http::insert_str(&mut user_defined, rustfs_utils::http::SUFFIX_ACTUAL_SIZE, declared.to_string());
}
ObjectInfo {
size: 12,
actual_size,
user_defined: Arc::new(user_defined),
..Default::default()
}
};
let normal = ObjectInfo {
size: 12,
actual_size: 10,
..Default::default()
};
assert_eq!(
resolve_size(&normal),
SizeResolution::Known {
logical: 10,
physical: 12
}
);
let stale_declared_metadata = ObjectInfo {
size: 12,
actual_size: 10,
user_defined: Arc::new(HashMap::from([("x-rustfs-internal-actual-size".to_string(), "not-a-size".to_string())])),
parts: Arc::new(vec![rustfs_filemeta::ObjectPartInfo {
actual_size: -2,
..Default::default()
}]),
..Default::default()
};
assert_eq!(
resolve_size(&stale_declared_metadata),
SizeResolution::Known {
logical: 10,
physical: 12
}
);
assert_eq!(
resolve_size(&compressed(0, Some("9"))),
SizeResolution::Known {
logical: 9,
physical: 12
}
);
assert_eq!(
resolve_size(&compressed(-1, None)),
SizeResolution::Unknown {
physical: 12,
reason: SizeResolutionReason::CompressedSizeUnknown,
}
);
assert!(matches!(
resolve_size(&compressed(0, Some("not-a-size"))),
SizeResolution::Corrupt {
reason: SizeResolutionReason::InvalidDeclaredSize,
..
}
));
assert!(matches!(
resolve_size(&ObjectInfo {
size: 12,
actual_size: -2,
..Default::default()
}),
SizeResolution::Corrupt { .. }
));
assert!(matches!(resolve_size(&compressed(0, Some("-1"))), SizeResolution::Corrupt { .. }));
assert!(matches!(resolve_size(&compressed(0, Some(""))), SizeResolution::Corrupt { .. }));
let unsupported = {
let mut object = compressed(0, None);
let mut metadata = (*object.user_defined).clone();
rustfs_utils::http::insert_str(&mut metadata, rustfs_utils::http::SUFFIX_COMPRESSION, "unsupported".to_string());
object.user_defined = Arc::new(metadata);
object
};
assert!(matches!(resolve_size(&unsupported), SizeResolution::Corrupt { .. }));
let invalid_part = {
let mut object = compressed(0, None);
object.parts = Arc::new(vec![rustfs_filemeta::ObjectPartInfo {
size: 12,
actual_size: -2,
..Default::default()
}]);
object
};
assert!(matches!(resolve_size(&invalid_part), SizeResolution::Corrupt { .. }));
let overflow = {
let mut object = compressed(0, None);
object.parts = Arc::new(vec![
rustfs_filemeta::ObjectPartInfo {
size: 1,
actual_size: i64::MAX,
..Default::default()
},
rustfs_filemeta::ObjectPartInfo {
size: 1,
actual_size: 1,
..Default::default()
},
]);
object
};
assert!(matches!(resolve_size(&overflow), SizeResolution::Corrupt { .. }));
let mismatch = compressed(0, None);
assert!(matches!(resolve_size(&mismatch), SizeResolution::Corrupt { .. }));
assert_eq!(
resolve_size(&ObjectInfo {
size: 0,
actual_size: 0,
..Default::default()
}),
SizeResolution::Known { logical: 0, physical: 0 }
);
}
#[test]
fn size_resolution_records_and_replays_one_identity() {
let version_id = uuid::Uuid::new_v4();
let generation = uuid::Uuid::new_v4();
let mut metadata = HashMap::new();
rustfs_utils::http::insert_str(&mut metadata, rustfs_utils::http::SUFFIX_COMPRESSION, "zstd".to_string());
rustfs_utils::http::insert_str(&mut metadata, rustfs_utils::http::SUFFIX_ACTUAL_SIZE, "not-a-number".to_string());
let corrupt = ObjectInfo {
bucket: "bucket".to_string(),
name: "object".to_string(),
size: 12,
version_id: Some(version_id),
data_dir: Some(generation),
user_defined: Arc::new(metadata),
..Default::default()
};
let mut summary = SizeSummary::default();
let resolution = resolve_size(&corrupt);
record_size_resolution(&mut summary, &corrupt, &resolution);
record_size_resolution(&mut summary, &corrupt, &resolution);
assert_eq!(summary.size_reconciliation.len(), 1);
assert_eq!(summary.size_reconciliation[0].reason, "invalid_declared_size");
assert_eq!(summary.size_reconciliation[0].physical_size, Some(12));
let known = ObjectInfo {
actual_size: 12,
user_defined: Arc::new(HashMap::new()),
..corrupt.clone()
};
record_size_resolution(&mut summary, &known, &resolve_size(&known));
summary.record_reconciliation_scope(&known.bucket, &known.name);
assert_eq!(summary.reconciliation_scopes.len(), 1);
assert_eq!(summary.reconciliation_scopes[0].bucket, "bucket");
}
#[test]
fn malformed_size_has_same_ilm_accounting() {
let mut metadata = HashMap::new();
rustfs_utils::http::insert_str(&mut metadata, rustfs_utils::http::SUFFIX_COMPRESSION, "zstd".to_string());
rustfs_utils::http::insert_str(&mut metadata, rustfs_utils::http::SUFFIX_ACTUAL_SIZE, "invalid".to_string());
let object = ObjectInfo {
bucket: "bucket".to_string(),
name: "object".to_string(),
size: 12,
user_defined: Arc::new(metadata),
..Default::default()
};
let resolution = resolve_size(&object);
let mut without_ilm = SizeSummary::default();
let mut with_ilm = SizeSummary::default();
record_size_resolution(&mut without_ilm, &object, &resolution);
record_size_resolution(&mut with_ilm, &object, &resolution);
assert_eq!(without_ilm.size_reconciliation, with_ilm.size_reconciliation);
assert_eq!(without_ilm.total_size, 0);
assert_eq!(with_ilm.total_size, 0);
assert!(without_ilm.tier_stats.is_empty());
assert!(with_ilm.tier_stats.is_empty());
}
#[test]
fn size_resolution_parses_once_per_version() {
let objects = vec![
ObjectInfo {
bucket: "bucket".to_string(),
name: "one".to_string(),
size: 1,
actual_size: 1,
..Default::default()
},
ObjectInfo {
bucket: "bucket".to_string(),
name: "two".to_string(),
size: 2,
actual_size: -2,
..Default::default()
},
];
let resolutions = resolve_sizes(&objects);
assert_eq!(resolutions.len(), objects.len());
assert!(matches!(resolutions[0], SizeResolution::Known { logical: 1, .. }));
assert!(matches!(resolutions[1], SizeResolution::Corrupt { .. }));
}
#[test]
fn queued_unknown_noncurrent_versions_are_removed_from_alert_count() {
assert_eq!(remaining_versions_after_queued_noncurrent(3, 1, 2), 0);
assert_eq!(remaining_versions_after_queued_noncurrent(7, 2, 1), 4);
assert_eq!(remaining_versions_after_queued_noncurrent(usize::MAX, usize::MAX, usize::MAX), 0);
}
#[test]
fn malformed_size_blocks_size_dependent_transition_but_allows_time_only_expiry() {
let size_filtered = BucketLifecycleConfiguration {
rules: vec![s3s::dto::LifecycleRule {
status: s3s::dto::ExpirationStatus::from_static(s3s::dto::ExpirationStatus::ENABLED),
expiration: None,
abort_incomplete_multipart_upload: None,
del_marker_expiration: None,
id: Some("size".to_string()),
filter: Some(s3s::dto::LifecycleRuleFilter {
object_size_greater_than: Some(1),
..Default::default()
}),
noncurrent_version_expiration: None,
noncurrent_version_transitions: None,
prefix: None,
transitions: None,
}],
..Default::default()
};
let unknown = SizeResolution::Unknown {
physical: 12,
reason: SizeResolutionReason::CompressedSizeUnknown,
};
let size_event = Event {
action: IlmAction::DeleteAction,
rule_id: "size".to_string(),
..Default::default()
};
assert!(!lifecycle_event_allowed(&unknown, &size_event, &size_filtered));
assert!(!lifecycle_event_allowed(
&unknown,
&Event {
action: IlmAction::TransitionAction,
rule_id: "size".to_string(),
..Default::default()
},
&size_filtered
));
let mixed_filters = BucketLifecycleConfiguration {
rules: vec![
size_filtered.rules[0].clone(),
s3s::dto::LifecycleRule {
status: s3s::dto::ExpirationStatus::from_static(s3s::dto::ExpirationStatus::ENABLED),
expiration: None,
abort_incomplete_multipart_upload: None,
del_marker_expiration: None,
id: Some("time".to_string()),
filter: None,
noncurrent_version_expiration: None,
noncurrent_version_transitions: None,
prefix: None,
transitions: None,
},
],
..Default::default()
};
assert!(lifecycle_event_allowed(
&unknown,
&Event {
action: IlmAction::DeleteAction,
rule_id: "time".to_string(),
..Default::default()
},
&mixed_filters
));
assert!(lifecycle_event_allowed(
&unknown,
&Event {
action: IlmAction::TransitionAction,
..Default::default()
},
&BucketLifecycleConfiguration::default()
));
assert!(lifecycle_event_allowed(
&SizeResolution::Corrupt {
physical: 12,
reason: SizeResolutionReason::InvalidDeclaredSize,
},
&Event {
action: IlmAction::DeleteAction,
..Default::default()
},
&BucketLifecycleConfiguration::default()
));
assert!(!lifecycle_event_allowed(
&SizeResolution::Corrupt {
physical: 12,
reason: SizeResolutionReason::InvalidDeclaredSize,
},
&Event {
action: IlmAction::DeleteAction,
rule_id: "size".to_string(),
..Default::default()
},
&size_filtered
));
assert!(lifecycle_rule_has_size_filter(
&BucketLifecycleConfiguration {
rules: vec![s3s::dto::LifecycleRule {
status: s3s::dto::ExpirationStatus::from_static(s3s::dto::ExpirationStatus::ENABLED),
expiration: None,
abort_incomplete_multipart_upload: None,
del_marker_expiration: None,
id: None,
filter: Some(s3s::dto::LifecycleRuleFilter {
object_size_greater_than: Some(1),
..Default::default()
}),
noncurrent_version_expiration: None,
noncurrent_version_transitions: None,
prefix: None,
transitions: None,
}],
..Default::default()
},
""
));
assert!(lifecycle_event_allowed(
&SizeResolution::Known {
logical: 10,
physical: 12,
},
&Event {
action: IlmAction::DeleteAllVersionsAction,
..Default::default()
},
&BucketLifecycleConfiguration::default()
));
assert!(lifecycle_event_allowed(
&unknown,
&Event {
action: IlmAction::DeleteAction,
rule_id: "time-only".to_string(),
..Default::default()
},
&BucketLifecycleConfiguration::default()
));
}
#[tokio::test]
async fn long_object_size_reconciliation_scope_uses_bounded_identity() {
let object_name = "o".repeat(600);
let mut item = scanner_item_with_prefix("");
item.object_name = object_name.clone();
let mut metadata = HashMap::new();
rustfs_utils::http::insert_str(&mut metadata, rustfs_utils::http::SUFFIX_COMPRESSION, "zstd".to_string());
let object = ObjectInfo {
bucket: item.bucket.clone(),
name: object_name.clone(),
size: 12,
actual_size: -1,
version_id: Some(uuid::Uuid::new_v4()),
user_defined: Arc::new(metadata),
..Default::default()
};
let mut summary = SizeSummary::default();
item.apply_actions(vec![object], None, VersioningConfiguration::default(), &mut summary)
.await;
let bounded_bucket = bounded_reconciliation_field(&item.bucket);
let bounded_object = bounded_reconciliation_field(&object_name);
assert_eq!(summary.reconciliation_scopes[0].bucket, bounded_bucket);
assert_eq!(summary.reconciliation_scopes[0].object, bounded_object);
assert_eq!(summary.size_reconciliation[0].object, bounded_object);
assert_eq!(summary.versions, 1);
assert_eq!(summary.total_size, 0);
}
}
@@ -326,9 +326,6 @@ async fn build_test_scanner() -> (FolderScanner, std::path::PathBuf) {
skip_heal: Arc::new(AtomicBool::new(false)),
local_disk: disk,
pending_heals_changed: false,
pending_size_reconciliation_keys: HashSet::new(),
pending_size_reconciliation_scopes: HashSet::new(),
pending_size_reconciliation_truncated: false,
list_path_raw_options_observer: None,
};
@@ -391,66 +388,6 @@ async fn test_record_failed_ttl_zero_noop() {
assert!(!scanner.should_skip_failed("path2"));
}
#[tokio::test]
async fn malformed_size_reconciliation_replays_after_restart() {
let (mut scanner, temp_dir) = build_test_scanner().await;
let _guard = TestGuard::new(60, 100, &mut scanner, temp_dir);
let entry = SizeReconciliationEntry {
key: "1:b|6:object|0:|0:".to_string(),
bucket: "b".to_string(),
object: "object".to_string(),
reason: "invalid_declared_size".to_string(),
physical_size: Some(12),
..Default::default()
};
let mut summary = SizeSummary::default();
summary.record_size_reconciliation(entry.clone());
summary.record_reconciliation_scope("b", "object");
scanner.apply_size_reconciliation(&summary);
scanner.apply_size_reconciliation(&summary);
assert_eq!(scanner.new_cache.info.size_reconciliation.len(), 1);
assert_eq!(scanner.update_cache.info.size_reconciliation.len(), 1);
assert_eq!(scanner.new_cache.info.size_reconciliation[&entry.key].attempts, 2);
let encoded = rmp_serde::to_vec_named(&scanner.new_cache.info).expect("size ledger should encode");
let decoded: crate::data_usage_define::DataUsageCacheInfo =
rmp_serde::from_slice(&encoded).expect("size ledger should decode");
assert_eq!(decoded.size_reconciliation.len(), 1);
assert_eq!(decoded.size_reconciliation[&entry.key].reason, "invalid_declared_size");
let mut resolved = SizeSummary::default();
resolved.record_reconciliation_scope("b", "object");
scanner.apply_size_reconciliation(&resolved);
assert!(scanner.new_cache.info.size_reconciliation.is_empty());
assert!(scanner.update_cache.info.size_reconciliation.is_empty());
}
#[tokio::test]
async fn malformed_size_reconciliation_clears_bounded_long_object_scope() {
let (mut scanner, temp_dir) = build_test_scanner().await;
let _guard = TestGuard::new(60, 100, &mut scanner, temp_dir);
let long_object = "o".repeat(600);
let bounded_object = item_actions::bounded_reconciliation_field(&long_object);
let entry = SizeReconciliationEntry {
key: "long-object-key".to_string(),
bucket: "b".to_string(),
object: bounded_object,
reason: "invalid_declared_size".to_string(),
..Default::default()
};
let mut summary = SizeSummary::default();
summary.record_size_reconciliation(entry);
scanner.apply_size_reconciliation(&summary);
assert_eq!(scanner.new_cache.info.size_reconciliation.len(), 1);
let mut resolved = SizeSummary::default();
resolved.record_reconciliation_scope("b", &long_object);
scanner.apply_size_reconciliation(&resolved);
assert!(scanner.new_cache.info.size_reconciliation.is_empty());
}
#[test]
fn test_classify_get_size_failure_marks_metadata_heal_object_path() {
let temp_dir = std::env::temp_dir();
+1 -1
View File
@@ -241,7 +241,7 @@ env \
RUSTFS_TEST_VAULT_FAILOVER_MARKER="$MARKER" \
RUSTFS_TEST_VAULT_OLD_LEADER="$OLD_LEADER" \
cargo test -p rustfs-kms --test vault_ha_failover_live \
vault_raft_leader_failure_recovers_kv2_and_transit_decrypts -- \
vault_raft_leader_failure_preserves_kv2_and_transit_decrypts -- \
--ignored --nocapture --test-threads=1 &
TEST_PID=$!