fix(scanner): report active first-cycle status (#5397)

* fix(scanner): report active first-cycle status

* fix(scanner): publish cycle activity consistently

* test(common): satisfy Rust 1.97 waker lint

* fix(scanner): satisfy Rust 1.97 clippy

---------

Co-authored-by: houseme <housemecn@gmail.com>
This commit is contained in:
Henry Guo
2026-07-29 22:33:25 +08:00
committed by GitHub
parent 3c0d315a9c
commit 3f20fbd77b
7 changed files with 846 additions and 102 deletions
+145 -33
View File
@@ -1114,6 +1114,8 @@ pub struct ScannerLastMinute {
pub struct ScannerMetricsReport {
pub collected_at: DateTime<Utc>,
pub current_cycle: u64,
#[serde(default)]
pub current_cycle_active: bool,
pub current_started: DateTime<Utc>,
pub cycles_completed_at: Vec<DateTime<Utc>>,
pub ongoing_buckets: usize,
@@ -2051,7 +2053,7 @@ impl Metrics {
pub fn record_scanner_transition_failed(&self, count: u64) {
self.scanner_transition_failed.fetch_add(count, Ordering::Relaxed);
self.record_scanner_source_failed(ScannerWorkSource::Lifecycle, count);
if !self.current_scan_cycle_work_active.load(Ordering::Relaxed) {
if !self.current_scan_cycle_work_active.load(Ordering::Acquire) {
self.record_last_cycle_scanner_source_work(ScannerWorkSource::Lifecycle, ScannerSourceWorkUpdate::failed(count));
}
}
@@ -2336,6 +2338,21 @@ impl Metrics {
*self.cycle_info.write().await = cycle;
}
/// Publish a scanner cycle and its work-accounting baseline as one state transition.
pub async fn start_scan_cycle_work_with_cycle(&self, cycle: CurrentCycle) -> ScanCycleWorkSnapshot {
let mut current_cycle = self.cycle_info.write().await;
let snapshot = self.start_scan_cycle_work();
*current_cycle = Some(cycle);
snapshot
}
/// Publish the completed work snapshot and idle cycle state as one state transition.
pub async fn finish_scan_cycle_work_with_cycle(&self, start: ScanCycleWorkSnapshot, cycle: CurrentCycle) {
let mut current_cycle = self.cycle_info.write().await;
self.finish_scan_cycle_work(start);
*current_cycle = Some(cycle);
}
/// Read the current cycle record.
pub async fn get_cycle(&self) -> Option<CurrentCycle> {
self.cycle_info.read().await.clone()
@@ -2464,7 +2481,7 @@ impl Metrics {
&self.current_scan_cycle_replication_repair_work_start,
&replication_repair_snapshot,
);
self.current_scan_cycle_work_active.store(true, Ordering::Relaxed);
self.current_scan_cycle_work_active.store(true, Ordering::Release);
snapshot
}
@@ -2476,11 +2493,11 @@ impl Metrics {
self.record_scan_cycle_work(work);
self.record_scan_cycle_source_work(&source_work);
self.record_scan_cycle_replication_repair_work(&replication_repair_work);
self.current_scan_cycle_work_active.store(false, Ordering::Relaxed);
self.current_scan_cycle_work_active.store(false, Ordering::Release);
}
pub fn current_scan_cycle_has_unresolved_heal_work(&self) -> bool {
if !self.current_scan_cycle_work_active.load(Ordering::Relaxed) {
if !self.current_scan_cycle_work_active.load(Ordering::Acquire) {
return false;
}
@@ -2746,13 +2763,41 @@ impl Metrics {
pub async fn report(&self) -> ScannerMetricsReport {
let mut m = ScannerMetricsReport::default();
let has_cycle = if let Some(cycle) = self.get_cycle().await {
m.current_cycle = cycle.current;
m.cycles_completed_at = cycle.cycle_completed;
m.current_started = cycle.started;
true
} else {
false
let has_cycle = {
let cycle = self.cycle_info.read().await;
let has_cycle = if let Some(cycle) = cycle.as_ref() {
m.current_cycle = cycle.current;
m.cycles_completed_at = cycle.cycle_completed.clone();
m.current_started = cycle.started;
true
} else {
false
};
m.current_cycle_active = self.current_scan_cycle_work_active.load(Ordering::Acquire);
if m.current_cycle_active {
let current_work = self.scan_cycle_work_since(self.current_scan_cycle_work_start());
let current_source_work = self.scanner_source_work_since(&self.current_scan_cycle_source_work_start_values());
let current_replication_repair_work =
self.scanner_replication_repair_work_since(&self.current_scan_cycle_replication_repair_work_start_values());
m.current_cycle_objects_scanned = current_work.objects_scanned;
m.current_cycle_directories_scanned = current_work.directories_scanned;
m.current_cycle_bucket_drive_scans = current_work.bucket_drive_scans;
m.current_cycle_bucket_drive_failures = current_work.bucket_drive_failures;
m.current_cycle_yield_events = current_work.yield_events;
m.current_cycle_yield_duration_seconds = current_work.yield_duration_millis as f64 / 1000.0;
m.current_cycle_throttle_sleep_events = current_work.throttle_sleep_events;
m.current_cycle_throttle_sleep_duration_seconds = current_work.throttle_sleep_duration_millis as f64 / 1000.0;
m.current_cycle_ilm_actions = current_work.ilm_actions;
m.current_cycle_lifecycle_expiry_actions = current_work.lifecycle_expiry_actions;
m.current_cycle_lifecycle_transition_actions = current_work.lifecycle_transition_actions;
m.current_cycle_heal_objects = current_work.heal_objects;
m.current_cycle_replication_checks = current_work.replication_checks;
m.current_cycle_usage_saves = current_work.usage_saves;
m.current_cycle_source_work = self.scanner_source_work_snapshots(&current_source_work);
m.current_cycle_replication_repair =
self.scanner_replication_repair_work_snapshots(&current_replication_repair_work);
}
has_cycle
};
if !has_cycle && let Some(init_time) = crate::get_global_init_time().await {
@@ -2793,28 +2838,6 @@ impl Metrics {
m.current_disk_scan_concurrency_limit = disk_scan_concurrency_limit;
m.current_disk_bucket_scans_queued = disk_bucket_scans_queued;
m.current_disk_bucket_scans_active = disk_bucket_scans_active;
if self.current_scan_cycle_work_active.load(Ordering::Relaxed) {
let current_work = self.scan_cycle_work_since(self.current_scan_cycle_work_start());
let current_source_work = self.scanner_source_work_since(&self.current_scan_cycle_source_work_start_values());
let current_replication_repair_work =
self.scanner_replication_repair_work_since(&self.current_scan_cycle_replication_repair_work_start_values());
m.current_cycle_objects_scanned = current_work.objects_scanned;
m.current_cycle_directories_scanned = current_work.directories_scanned;
m.current_cycle_bucket_drive_scans = current_work.bucket_drive_scans;
m.current_cycle_bucket_drive_failures = current_work.bucket_drive_failures;
m.current_cycle_yield_events = current_work.yield_events;
m.current_cycle_yield_duration_seconds = current_work.yield_duration_millis as f64 / 1000.0;
m.current_cycle_throttle_sleep_events = current_work.throttle_sleep_events;
m.current_cycle_throttle_sleep_duration_seconds = current_work.throttle_sleep_duration_millis as f64 / 1000.0;
m.current_cycle_ilm_actions = current_work.ilm_actions;
m.current_cycle_lifecycle_expiry_actions = current_work.lifecycle_expiry_actions;
m.current_cycle_lifecycle_transition_actions = current_work.lifecycle_transition_actions;
m.current_cycle_heal_objects = current_work.heal_objects;
m.current_cycle_replication_checks = current_work.replication_checks;
m.current_cycle_usage_saves = current_work.usage_saves;
m.current_cycle_source_work = self.scanner_source_work_snapshots(&current_source_work);
m.current_cycle_replication_repair = self.scanner_replication_repair_work_snapshots(&current_replication_repair_work);
}
let last_cycle_result = self.last_scan_cycle_result.load(Ordering::Relaxed);
m.last_cycle_result = scan_cycle_result_label(last_cycle_result).to_string();
m.last_cycle_result_code = last_cycle_result as u64;
@@ -4142,6 +4165,8 @@ mod tests {
let report = metrics.report().await;
assert!(report.current_cycle_active);
assert_eq!(report.current_cycle, 0);
assert_eq!(report.current_cycle_objects_scanned, 7);
assert_eq!(report.current_cycle_directories_scanned, 3);
assert_eq!(report.current_cycle_bucket_drive_scans, 2);
@@ -4158,6 +4183,8 @@ mod tests {
metrics.finish_scan_cycle_work(start);
let report = metrics.report().await;
assert!(!report.current_cycle_active);
assert_eq!(report.current_cycle, 0);
assert_eq!(report.current_cycle_objects_scanned, 0);
assert_eq!(report.current_cycle_directories_scanned, 0);
assert_eq!(report.current_cycle_bucket_drive_scans, 0);
@@ -4184,6 +4211,91 @@ mod tests {
assert_eq!(report.last_cycle_usage_saves, 2);
}
#[tokio::test]
async fn scan_cycle_activity_and_cycle_state_publish_together() {
let metrics = Metrics::new();
let cycle_started = Utc::now() - chrono::Duration::seconds(5);
let active_cycle = CurrentCycle {
current: 12,
next: 13,
started: cycle_started,
..Default::default()
};
let cycle_state = metrics.cycle_info.read().await;
let mut start_transition = Box::pin(metrics.start_scan_cycle_work_with_cycle(active_cycle));
let waker = std::task::Waker::noop();
let mut context = std::task::Context::from_waker(waker);
assert!(start_transition.as_mut().poll(&mut context).is_pending());
assert!(!metrics.current_scan_cycle_work_active.load(Ordering::Acquire));
drop(cycle_state);
let start = start_transition.await;
let active = metrics.report().await;
assert!(active.current_cycle_active);
assert_eq!(active.current_cycle, 12);
assert_eq!(active.current_started, cycle_started);
let idle_cycle = CurrentCycle {
current: 0,
next: 13,
started: cycle_started,
..Default::default()
};
let cycle_state = metrics.cycle_info.read().await;
let mut finish_transition = Box::pin(metrics.finish_scan_cycle_work_with_cycle(start, idle_cycle));
assert!(finish_transition.as_mut().poll(&mut context).is_pending());
assert!(metrics.current_scan_cycle_work_active.load(Ordering::Acquire));
drop(cycle_state);
finish_transition.await;
let idle = metrics.report().await;
assert!(!idle.current_cycle_active);
assert_eq!(idle.current_cycle, 0);
}
#[tokio::test]
async fn report_keeps_cycle_identity_and_work_in_one_snapshot() {
let metrics = Metrics::new();
let cycle_ten = CurrentCycle {
current: 10,
next: 11,
started: Utc::now() - chrono::Duration::seconds(10),
..Default::default()
};
let cycle_ten_start = metrics.start_scan_cycle_work_with_cycle(cycle_ten.clone()).await;
metrics.operations[Metric::ScanObject as usize].store(1, Ordering::Relaxed);
let paths = metrics.current_paths.write().await;
let mut report = Box::pin(metrics.report());
let waker = std::task::Waker::noop();
let mut context = std::task::Context::from_waker(waker);
assert!(report.as_mut().poll(&mut context).is_pending());
metrics
.finish_scan_cycle_work_with_cycle(cycle_ten_start, CurrentCycle { current: 0, ..cycle_ten })
.await;
let cycle_eleven_start = metrics
.start_scan_cycle_work_with_cycle(CurrentCycle {
current: 11,
next: 12,
started: Utc::now(),
..Default::default()
})
.await;
metrics.operations[Metric::ScanObject as usize].store(101, Ordering::Relaxed);
drop(paths);
let snapshot = report.await;
assert_eq!(snapshot.current_cycle, 10);
assert_eq!(snapshot.current_cycle_objects_scanned, 1);
metrics
.finish_scan_cycle_work_with_cycle(cycle_eleven_start, CurrentCycle::default())
.await;
}
#[tokio::test]
async fn scanner_cycle_ilm_actions_ignore_global_ilm_work() {
let metrics = Metrics::new();
-4
View File
@@ -207,10 +207,6 @@ pub(crate) async fn ensure_boot_time() {
GLOBAL_BOOT_TIME.get_or_init(|| async { SystemTime::now() }).await;
}
pub(crate) async fn scanner_init_time() -> Option<chrono::DateTime<chrono::Utc>> {
rustfs_common::get_global_init_time().await
}
pub(crate) async fn root_disk_threshold_for_erasure_disk() -> Option<u64> {
if is_erasure_sd().await {
None
@@ -71,6 +71,7 @@ fn to_madmin_scanner_metrics(metrics: rustfs_common::metrics::ScannerMetricsRepo
MadminScannerMetrics {
collected_at: metrics.collected_at,
current_cycle: metrics.current_cycle,
current_cycle_active: Some(metrics.current_cycle_active),
current_started: metrics.current_started,
cycles_completed_at: metrics.cycles_completed_at,
ongoing_buckets: metrics.ongoing_buckets,
@@ -398,10 +399,7 @@ pub async fn collect_local_metrics(types: MetricType, opts: &CollectMetricsOpts)
if types.contains(&MetricType::SCANNER) {
debug!("start get scanner metrics");
let mut metrics = global_metrics().report().await;
if let Some(init_time) = runtime_sources::scanner_init_time().await {
metrics.current_started = init_time;
}
let metrics = global_metrics().report().await;
real_time_metrics.aggregated.scanner = Some(to_madmin_scanner_metrics(metrics));
}
@@ -540,7 +538,9 @@ async fn collect_local_disks_metrics(disks: &HashSet<String>) -> HashMap<String,
#[cfg(test)]
mod test {
use super::*;
use rustfs_common::metrics::CurrentCycle;
use rustfs_io_metrics::internode_metrics::global_internode_metrics;
use serial_test::serial;
use std::time::Duration;
#[test]
@@ -588,7 +588,10 @@ mod test {
#[test]
fn scanner_metrics_mapping_preserves_partial_source_status() {
let current_started = Utc::now() - chrono::Duration::seconds(5);
let scanner = to_madmin_scanner_metrics(rustfs_common::metrics::ScannerMetricsReport {
current_cycle_active: true,
current_started,
last_cycle_partial_source: "usage".to_string(),
last_cycle_partial_source_code: 1,
partial_cycles_by_source: vec![rustfs_common::metrics::ScannerSourceCycleSnapshot {
@@ -598,6 +601,8 @@ mod test {
..Default::default()
});
assert_eq!(scanner.current_cycle_active, Some(true));
assert_eq!(scanner.current_started, current_started);
assert_eq!(scanner.last_cycle_partial_source, "usage");
assert_eq!(scanner.last_cycle_partial_source_code, 1);
let usage = scanner
@@ -608,6 +613,39 @@ mod test {
assert_eq!(usage.cycles, 2);
}
#[tokio::test]
#[serial]
async fn collect_local_metrics_preserves_scanner_cycle_started_time() {
let previous_init_time = *rustfs_common::globals::GLOBAL_INIT_TIME.read().await;
let previous_cycle = global_metrics().get_cycle().await;
let init_time = Utc::now() - chrono::Duration::hours(1);
let cycle_started = Utc::now() - chrono::Duration::seconds(5);
*rustfs_common::globals::GLOBAL_INIT_TIME.write().await = Some(init_time);
let cycle = CurrentCycle {
current: 0,
next: 1,
started: cycle_started,
..Default::default()
};
let cycle_start = global_metrics().start_scan_cycle_work_with_cycle(cycle).await;
let realtime = collect_local_metrics(MetricType::SCANNER, &CollectMetricsOpts::default()).await;
global_metrics()
.finish_scan_cycle_work_with_cycle(cycle_start, previous_cycle.clone().unwrap_or_default())
.await;
global_metrics().set_cycle(previous_cycle).await;
*rustfs_common::globals::GLOBAL_INIT_TIME.write().await = previous_init_time;
let encoded = rmp_serde::to_vec_named(&realtime).expect("realtime metrics should encode");
let decoded: RealtimeMetrics = rmp_serde::from_slice(&encoded).expect("realtime metrics should decode");
let mut aggregated = RealtimeMetrics::default();
aggregated.merge(decoded);
let scanner = aggregated.aggregated.scanner.expect("scanner metrics");
assert_eq!(scanner.current_cycle_active, Some(true));
assert_eq!(scanner.current_started, cycle_started);
}
#[test]
fn scanner_metrics_mapping_preserves_pacing_pressure() {
let scanner = to_madmin_scanner_metrics(rustfs_common::metrics::ScannerMetricsReport {
+309 -6
View File
@@ -545,6 +545,8 @@ pub struct ScannerMetrics {
pub collected_at: DateTime<Utc>,
#[serde(rename = "current_cycle")]
pub current_cycle: u64,
#[serde(rename = "current_cycle_active", default, skip_serializing_if = "Option::is_none")]
pub current_cycle_active: Option<bool>,
#[serde(rename = "current_started")]
pub current_started: DateTime<Utc>,
#[serde(rename = "cycle_complete_times")]
@@ -718,7 +720,31 @@ pub struct ScannerMetrics {
}
impl ScannerMetrics {
pub fn is_current_cycle_active(&self) -> bool {
self.current_cycle_active.unwrap_or(self.current_cycle > 0)
}
pub fn merge(&mut self, other: &Self) {
// Legacy nodes omit the activity field and use a non-zero cycle as
// their active signal. New nodes publish explicit first-cycle and idle
// states, including cycle zero.
let self_cycle_active = self.is_current_cycle_active();
let other_cycle_active = other.is_current_cycle_active();
let self_cycle_authority = (
self_cycle_active,
self.current_cycle,
self.cycles_completed_at.len(),
self.cycles_completed_at.as_slice(),
self.current_started,
);
let other_cycle_authority = (
other_cycle_active,
other.current_cycle,
other.cycles_completed_at.len(),
other.cycles_completed_at.as_slice(),
other.current_started,
);
let other_cycle_is_authoritative = self_cycle_authority < other_cycle_authority;
let other_is_newer = self.collected_at < other.collected_at;
if other_is_newer {
self.collected_at = other.collected_at;
@@ -854,15 +880,12 @@ impl ScannerMetrics {
self.ongoing_buckets = other.ongoing_buckets;
}
if self.current_cycle < other.current_cycle {
if other_cycle_is_authoritative {
self.current_cycle = other.current_cycle;
self.cycles_completed_at = other.cycles_completed_at.clone();
self.current_started = other.current_started;
}
if other.cycles_completed_at.len() > self.cycles_completed_at.len() {
self.cycles_completed_at = other.cycles_completed_at.clone();
}
self.current_cycle_active = Some(self_cycle_active || other_cycle_active);
if !other.life_time_ops.is_empty() && self.life_time_ops.is_empty() {
self.life_time_ops = other.life_time_ops.clone();
@@ -931,7 +954,13 @@ impl Metrics {
if let Some(scanner) = other.scanner.as_ref() {
match self.scanner {
Some(ref mut s_scanner) => s_scanner.merge(scanner),
None => self.scanner = Some(scanner.clone()),
None => {
let mut scanner = scanner.clone();
if scanner.current_cycle_active.is_none() {
scanner.current_cycle_active = Some(scanner.is_current_cycle_active());
}
self.scanner = Some(scanner);
}
}
}
@@ -1389,6 +1418,280 @@ pub struct Operations {
mod tests {
use super::*;
#[test]
fn scanner_metrics_serializes_cycle_active_presence() {
let missing_value = serde_json::to_value(ScannerMetrics::default()).expect("scanner metrics should serialize");
assert!(missing_value.get("current_cycle_active").is_none());
let missing: ScannerMetrics =
serde_json::from_value(missing_value).expect("older scanner metrics without cycle-active should decode");
assert_eq!(missing.current_cycle_active, None);
let explicit_false_value = serde_json::to_value(ScannerMetrics {
current_cycle_active: Some(false),
..Default::default()
})
.expect("scanner metrics with explicit cycle-active should serialize");
assert_eq!(explicit_false_value["current_cycle_active"], serde_json::Value::Bool(false));
let explicit_false: ScannerMetrics =
serde_json::from_value(explicit_false_value).expect("explicit cycle-active should decode");
assert_eq!(explicit_false.current_cycle_active, Some(false));
}
#[test]
fn scanner_metrics_merge_prefers_an_active_first_cycle() {
let collected_at = Utc::now();
let idle_started = collected_at - chrono::Duration::hours(1);
let active_started = collected_at - chrono::Duration::seconds(5);
let mut scanner = ScannerMetrics {
collected_at,
current_cycle: 0,
current_cycle_active: Some(false),
current_started: idle_started,
..Default::default()
};
scanner.merge(&ScannerMetrics {
collected_at: collected_at + chrono::Duration::seconds(1),
current_cycle: 0,
current_cycle_active: Some(true),
current_started: active_started,
..Default::default()
});
assert_eq!(scanner.current_cycle_active, Some(true));
assert_eq!(scanner.current_cycle, 0);
assert_eq!(scanner.current_started, active_started);
}
#[test]
fn metrics_merge_preserves_explicit_active_first_cycle() {
let mut aggregated = Metrics::default();
aggregated.merge(&Metrics {
scanner: Some(ScannerMetrics {
current_cycle_active: Some(true),
..Default::default()
}),
..Default::default()
});
let scanner = aggregated.scanner.expect("aggregated scanner metrics");
assert_eq!(scanner.current_cycle_active, Some(true));
assert_eq!(scanner.current_cycle, 0);
}
#[test]
fn scanner_metrics_merge_preserves_legacy_nonzero_active_signal() {
let collected_at = Utc::now();
let mut scanner = ScannerMetrics {
collected_at,
..Default::default()
};
scanner.merge(&ScannerMetrics {
collected_at: collected_at + chrono::Duration::seconds(1),
current_cycle: 7,
..Default::default()
});
assert_eq!(scanner.current_cycle_active, Some(true));
assert_eq!(scanner.current_cycle, 7);
}
#[test]
fn metrics_merge_normalizes_first_legacy_scanner_snapshot() {
let legacy = Metrics {
scanner: Some(ScannerMetrics {
current_cycle: 7,
..Default::default()
}),
..Default::default()
};
let mut aggregated = Metrics::default();
aggregated.merge(&legacy);
let scanner = aggregated.scanner.expect("aggregated scanner metrics");
assert_eq!(scanner.current_cycle_active, Some(true));
assert_eq!(scanner.current_cycle, 7);
}
#[test]
fn scanner_metrics_merge_preserves_explicit_inactive_nonzero_cycle() {
let collected_at = Utc::now();
let mut scanner = ScannerMetrics::default();
scanner.merge(&ScannerMetrics {
collected_at,
current_cycle: 7,
current_cycle_active: Some(false),
..Default::default()
});
assert_eq!(scanner.current_cycle_active, Some(false));
assert_eq!(scanner.current_cycle, 7);
}
#[test]
fn scanner_metrics_merge_cycle_active_is_order_independent() {
let collected_at = Utc::now();
let legacy_active = ScannerMetrics {
collected_at,
current_cycle: 7,
current_started: collected_at - chrono::Duration::seconds(10),
..Default::default()
};
let explicit_idle = ScannerMetrics {
collected_at,
current_cycle: 0,
current_cycle_active: Some(false),
current_started: collected_at - chrono::Duration::hours(1),
..Default::default()
};
let mut legacy_first = legacy_active.clone();
legacy_first.merge(&explicit_idle);
let mut legacy_second = explicit_idle.clone();
legacy_second.merge(&legacy_active);
assert_eq!(legacy_first.current_cycle_active, Some(true));
assert_eq!(legacy_second.current_cycle_active, Some(true));
assert_eq!(legacy_first.current_cycle, 7);
assert_eq!(legacy_second.current_cycle, 7);
let explicit_inactive_nonzero = ScannerMetrics {
collected_at,
current_cycle: 7,
current_cycle_active: Some(false),
..Default::default()
};
let mut inactive_first = explicit_inactive_nonzero.clone();
inactive_first.merge(&explicit_idle);
let mut inactive_second = explicit_idle;
inactive_second.merge(&explicit_inactive_nonzero);
assert_eq!(inactive_first.current_cycle_active, Some(false));
assert_eq!(inactive_second.current_cycle_active, Some(false));
assert_eq!(inactive_first.current_cycle, 7);
assert_eq!(inactive_second.current_cycle, 7);
}
#[test]
fn scanner_metrics_merge_cycle_authority_is_order_independent() {
let collected_at = Utc::now();
let completion = collected_at - chrono::Duration::minutes(1);
let earlier_active = ScannerMetrics {
collected_at,
current_cycle: 7,
current_cycle_active: Some(true),
current_started: collected_at - chrono::Duration::seconds(10),
cycles_completed_at: vec![completion],
..Default::default()
};
let later_active = ScannerMetrics {
collected_at: collected_at + chrono::Duration::seconds(1),
current_cycle: 7,
current_cycle_active: Some(true),
current_started: collected_at - chrono::Duration::seconds(5),
cycles_completed_at: vec![completion],
..Default::default()
};
let mut earlier_first = earlier_active.clone();
earlier_first.merge(&later_active);
let mut later_first = later_active.clone();
later_first.merge(&earlier_active);
assert_eq!(earlier_first.current_started, later_active.current_started);
assert_eq!(later_first.current_started, later_active.current_started);
assert_eq!(earlier_first.cycles_completed_at, later_active.cycles_completed_at);
assert_eq!(later_first.cycles_completed_at, later_active.cycles_completed_at);
let stale_idle = ScannerMetrics {
collected_at,
current_cycle_active: Some(false),
current_started: collected_at - chrono::Duration::hours(1),
..Default::default()
};
let completed_idle = ScannerMetrics {
collected_at: collected_at + chrono::Duration::seconds(1),
current_cycle_active: Some(false),
current_started: collected_at - chrono::Duration::seconds(5),
cycles_completed_at: vec![completion],
..Default::default()
};
let mut stale_first = stale_idle.clone();
stale_first.merge(&completed_idle);
let mut completed_first = completed_idle.clone();
completed_first.merge(&stale_idle);
assert_eq!(stale_first.current_started, completed_idle.current_started);
assert_eq!(completed_first.current_started, completed_idle.current_started);
assert_eq!(stale_first.cycles_completed_at, completed_idle.cycles_completed_at);
assert_eq!(completed_first.cycles_completed_at, completed_idle.cycles_completed_at);
}
#[test]
fn scanner_metrics_merge_cycle_authority_is_associative() {
let collected_at = Utc::now();
let older_completion = collected_at - chrono::Duration::minutes(3);
let last_completion = collected_at - chrono::Duration::minutes(1);
let cycle_seven = ScannerMetrics {
collected_at,
current_cycle: 7,
current_cycle_active: Some(true),
current_started: collected_at - chrono::Duration::seconds(10),
cycles_completed_at: vec![older_completion, last_completion],
..Default::default()
};
let cycle_eight = ScannerMetrics {
collected_at: collected_at + chrono::Duration::seconds(1),
current_cycle: 8,
current_cycle_active: Some(true),
current_started: collected_at - chrono::Duration::seconds(5),
cycles_completed_at: vec![last_completion],
..Default::default()
};
let newer_idle = ScannerMetrics {
collected_at: collected_at + chrono::Duration::hours(1),
current_cycle_active: Some(false),
current_started: collected_at - chrono::Duration::hours(1),
..Default::default()
};
let mut left_associative = cycle_seven.clone();
left_associative.merge(&cycle_eight);
left_associative.merge(&newer_idle);
let mut right_group = cycle_eight.clone();
right_group.merge(&newer_idle);
let mut right_associative = cycle_seven.clone();
right_associative.merge(&right_group);
assert_eq!(left_associative.current_cycle, 8);
assert_eq!(right_associative.current_cycle, 8);
assert_eq!(left_associative.current_started, cycle_eight.current_started);
assert_eq!(right_associative.current_started, cycle_eight.current_started);
assert_eq!(left_associative.cycles_completed_at, cycle_eight.cycles_completed_at);
assert_eq!(right_associative.cycles_completed_at, cycle_eight.cycles_completed_at);
for order in [
[&cycle_seven, &cycle_eight, &newer_idle],
[&cycle_seven, &newer_idle, &cycle_eight],
[&cycle_eight, &cycle_seven, &newer_idle],
[&cycle_eight, &newer_idle, &cycle_seven],
[&newer_idle, &cycle_seven, &cycle_eight],
[&newer_idle, &cycle_eight, &cycle_seven],
] {
let mut merged = ScannerMetrics::default();
for scanner in order {
merged.merge(scanner);
}
assert_eq!(merged.current_cycle_active, Some(true));
assert_eq!(merged.current_cycle, 8);
assert_eq!(merged.current_started, cycle_eight.current_started);
assert_eq!(merged.cycles_completed_at, cycle_eight.cycles_completed_at);
}
}
#[test]
fn scanner_metrics_merge_aggregates_partial_cycles_by_source() {
let collected_at = Utc::now();
+7 -7
View File
@@ -262,11 +262,11 @@ async fn obs_site_replication_stats() -> ReplicationStats {
}
fn current_scanner_cycle_age_seconds(
current_cycle: u64,
current_cycle_active: bool,
current_started: chrono::DateTime<Utc>,
now: chrono::DateTime<Utc>,
) -> u64 {
if current_cycle == 0 {
if !current_cycle_active {
0
} else {
now.signed_duration_since(current_started).num_seconds().max(0) as u64
@@ -1090,7 +1090,7 @@ pub async fn collect_scanner_metric_stats() -> Option<ScannerStats> {
let reference_time = metrics.cycles_completed_at.last().copied().unwrap_or(metrics.current_started);
let last_activity_seconds = now.signed_duration_since(reference_time).num_seconds().max(0) as u64;
let active_paths = metrics.active_scan_paths as u64;
let current_cycle_age_seconds = current_scanner_cycle_age_seconds(metrics.current_cycle, metrics.current_started, now);
let current_cycle_age_seconds = current_scanner_cycle_age_seconds(metrics.current_cycle_active, metrics.current_started, now);
let current_scan_mode = scanner_scan_mode_code(&metrics.current_scan_mode);
let current_cycle_age = current_cycle_age_seconds as f64;
let last_cycle_duration = metrics.last_cycle_duration_seconds;
@@ -1464,21 +1464,21 @@ mod tests {
fn current_scanner_cycle_age_seconds_returns_zero_when_idle() {
let now = Utc::now();
assert_eq!(current_scanner_cycle_age_seconds(0, now - chrono::Duration::seconds(30), now), 0);
assert_eq!(current_scanner_cycle_age_seconds(false, now - chrono::Duration::seconds(30), now), 0);
}
#[test]
fn current_scanner_cycle_age_seconds_clamps_future_start() {
let now = Utc::now();
assert_eq!(current_scanner_cycle_age_seconds(4, now + chrono::Duration::seconds(30), now), 0);
assert_eq!(current_scanner_cycle_age_seconds(true, now + chrono::Duration::seconds(30), now), 0);
}
#[test]
fn current_scanner_cycle_age_seconds_reports_active_elapsed_time() {
fn current_scanner_cycle_age_seconds_reports_active_first_cycle_elapsed_time() {
let now = Utc::now();
assert_eq!(current_scanner_cycle_age_seconds(4, now - chrono::Duration::seconds(45), now), 45);
assert_eq!(current_scanner_cycle_age_seconds(true, now - chrono::Duration::seconds(45), now), 45);
}
#[test]
+1 -1
View File
@@ -47,7 +47,7 @@ rmp-serde = { workspace = true }
hmac = { workspace = true }
sha2 = { workspace = true }
rustfs-filemeta = { workspace = true }
tokio-util = { workspace = true, features = ["io", "compat"] }
tokio-util = { workspace = true, features = ["io", "compat", "rt"] }
rustfs-ecstore = { workspace = true }
rustfs-storage-api = { workspace = true }
http = { workspace = true }
+342 -47
View File
@@ -14,6 +14,8 @@
use std::collections::BTreeMap;
use std::future::Future;
#[cfg(test)]
use std::sync::Mutex as StdMutex;
use std::sync::{Arc, LazyLock, RwLock};
use crate::ScannerObjectIO;
@@ -38,8 +40,8 @@ use bytes::Bytes;
use chrono::{DateTime, Utc};
use rustfs_common::heal_channel::HealScanMode;
use rustfs_common::metrics::{
CurrentCycle, Metric, Metrics, ScanCyclePartialReason, ScannerUsageSaveResult, ScannerWorkSource, emit_scan_cycle_complete,
emit_scan_cycle_partial_with_source, emit_scan_cycle_superseded, global_metrics,
CurrentCycle, Metric, Metrics, ScanCyclePartialReason, ScanCycleWorkSnapshot, ScannerUsageSaveResult, ScannerWorkSource,
emit_scan_cycle_complete, emit_scan_cycle_partial_with_source, emit_scan_cycle_superseded, global_metrics,
};
use rustfs_config::ScannerSpeed;
#[cfg(test)]
@@ -50,9 +52,12 @@ use rustfs_config::{
use rustfs_config::{ENV_SCANNER_CYCLE, ENV_SCANNER_SPEED, ENV_SCANNER_START_DELAY_SECS};
use serde::{Deserialize, Serialize};
use sha2::{Digest as _, Sha256};
#[cfg(test)]
use tokio::sync::Notify;
use tokio::sync::mpsc;
use tokio::time::{Duration, Instant};
use tokio_util::sync::CancellationToken;
use tokio_util::task::AbortOnDropHandle;
use tracing::{debug, error, info, instrument, warn};
use crate::storage_api::scan::{
@@ -93,6 +98,44 @@ const SCANNER_CYCLE_STATE_MAGIC: &[u8; 8] = b"RSCYC001";
const SCANNER_CYCLE_STATE_HEADER_LEN: usize = 24;
#[cfg(test)]
const ENV_SCANNER_START_DELAY_SECS_DEPRECATED: &str = "RUSTFS_DATA_SCANNER_START_DELAY_SECS";
#[cfg(test)]
type ScannerCycleStatePersistTestHook = (u64, Arc<Notify>);
#[cfg(test)]
static SCANNER_CYCLE_STATE_PERSIST_TEST_HOOK: LazyLock<StdMutex<Option<ScannerCycleStatePersistTestHook>>> =
LazyLock::new(|| StdMutex::new(None));
#[cfg(test)]
struct ScannerCycleStatePersistTestHookGuard;
#[cfg(test)]
impl Drop for ScannerCycleStatePersistTestHookGuard {
fn drop(&mut self) {
*SCANNER_CYCLE_STATE_PERSIST_TEST_HOOK
.lock()
.unwrap_or_else(|poisoned| poisoned.into_inner()) = None;
}
}
#[cfg(test)]
fn set_scanner_cycle_state_persist_test_hook(leader_epoch: u64, reached: Arc<Notify>) -> ScannerCycleStatePersistTestHookGuard {
*SCANNER_CYCLE_STATE_PERSIST_TEST_HOOK
.lock()
.unwrap_or_else(|poisoned| poisoned.into_inner()) = Some((leader_epoch, reached));
ScannerCycleStatePersistTestHookGuard
}
#[cfg(test)]
fn notify_scanner_cycle_state_persist_test_hook(leader_epoch: u64) {
let reached = SCANNER_CYCLE_STATE_PERSIST_TEST_HOOK
.lock()
.unwrap_or_else(|poisoned| poisoned.into_inner())
.as_ref()
.filter(|(expected_epoch, _)| *expected_epoch == leader_epoch)
.map(|(_, reached)| reached.clone());
if let Some(reached) = reached {
reached.notify_one();
}
}
#[derive(Debug, thiserror::Error)]
enum ScannerCycleStateError {
@@ -1691,10 +1734,10 @@ fn data_usage_persist_timeout() -> Duration {
DataUsageCache::persistence_timeout()
}
async fn mark_scan_cycle_idle(cycle_info: &mut CurrentCycle) {
async fn mark_scan_cycle_idle(cycle_info: &mut CurrentCycle, cycle_metrics_guard: &mut ScannerCycleMetricsGuard) {
cycle_info.current = 0;
global_metrics().clear_current_scan_mode();
global_metrics().set_cycle(Some(cycle_info.clone())).await;
cycle_metrics_guard.finish(cycle_info.clone()).await;
}
fn encode_scanner_cycle_state(cycle_info: &CurrentCycle, leader_epoch: u64) -> Result<Vec<u8>, ScannerCycleStateError> {
@@ -2218,6 +2261,8 @@ async fn persist_scanner_cycle_state(
return false;
}
#[cfg(test)]
notify_scanner_cycle_state_persist_test_hook(leader_epoch);
match save_config_with_preconditions(storeapi.clone(), &DATA_USAGE_BLOOM_NAME_PATH, buf.clone(), revision.preconditions())
.await
{
@@ -2340,7 +2385,6 @@ async fn persist_scanner_cycle_state(
if persisted_cycle.next >= cycle_info.next {
*cycle_info = persisted_cycle;
global_metrics().set_cycle(Some(cycle_info.clone())).await;
debug!(
target: "rustfs::scanner",
event = EVENT_SCANNER_PERSIST_STATE,
@@ -2406,6 +2450,7 @@ async fn finalize_partial_scan_cycle(
cycle_info: &mut CurrentCycle,
revision: &mut DataUsageCacheRevision,
leader_epoch: u64,
cycle_metrics_guard: &mut ScannerCycleMetricsGuard,
) -> bool {
// A budget-limited cycle is deliberate pacing, not a failure. The cycle counter
// must still advance (and persist) because per-bucket next_cycle is stamped from
@@ -2422,11 +2467,14 @@ async fn finalize_partial_scan_cycle(
error = %err,
"Scanner partial cycle could not advance"
);
mark_scan_cycle_idle(cycle_info).await;
mark_scan_cycle_idle(cycle_info, cycle_metrics_guard).await;
return false;
}
mark_scan_cycle_idle(cycle_info).await;
persist_scanner_cycle_state(ctx, storeapi, cycle_info, revision, leader_epoch).await
cycle_info.current = 0;
global_metrics().clear_current_scan_mode();
let persisted = persist_scanner_cycle_state(ctx, storeapi, cycle_info, revision, leader_epoch).await;
cycle_metrics_guard.finish(cycle_info.clone()).await;
persisted
}
async fn persist_required_scanner_cycle_floor(
@@ -2436,6 +2484,7 @@ async fn persist_required_scanner_cycle_floor(
revision: &mut DataUsageCacheRevision,
leader_epoch: u64,
required_cycle: u64,
cycle_metrics_guard: &mut ScannerCycleMetricsGuard,
) -> bool {
if required_cycle <= cycle_info.current || required_cycle == u64::MAX {
error!(
@@ -2448,13 +2497,16 @@ async fn persist_required_scanner_cycle_floor(
state = "invalid_cache_cycle_floor",
"Scanner cache cycle floor is invalid"
);
mark_scan_cycle_idle(cycle_info).await;
mark_scan_cycle_idle(cycle_info, cycle_metrics_guard).await;
return false;
}
cycle_info.next = cycle_info.next.max(required_cycle);
mark_scan_cycle_idle(cycle_info).await;
persist_scanner_cycle_state(ctx, storeapi, cycle_info, revision, leader_epoch).await
cycle_info.current = 0;
global_metrics().clear_current_scan_mode();
let persisted = persist_scanner_cycle_state(ctx, storeapi, cycle_info, revision, leader_epoch).await;
cycle_metrics_guard.finish(cycle_info.clone()).await;
persisted
}
async fn await_scanner_cycle_with_lock_fence<Cycle, LockLost>(
@@ -2513,7 +2565,7 @@ async fn run_data_scanner_cycle(
let now = Instant::now();
cycle_info.started = Utc::now();
global_metrics().set_cycle(Some(cycle_info.clone())).await;
let mut cycle_metrics_guard = ScannerCycleMetricsGuard::new(cycle_info.clone()).await;
let mut background_heal_info = read_background_heal_info(storeapi.clone()).await;
@@ -2564,14 +2616,14 @@ async fn run_data_scanner_cycle(
"Scanner cycle could not capture the data usage persistence baseline"
);
emit_scan_cycle_complete(false, cycle_start.elapsed());
mark_scan_cycle_idle(cycle_info).await;
mark_scan_cycle_idle(cycle_info, &mut cycle_metrics_guard).await;
return ScannerCycleOutcome::Failed;
}
};
let (sender, receiver) = mpsc::channel::<DataUsageInfo>(1);
let storeapi_clone = storeapi.clone();
let ctx_clone = ctx.clone();
let mut usage_persist_task = tokio::spawn(async move {
let mut usage_persist_task = AbortOnDropHandle::new(tokio::spawn(async move {
store_data_usage_in_backend_with_outcome_for_epoch_and_baseline(
ctx_clone,
storeapi_clone,
@@ -2580,10 +2632,9 @@ async fn run_data_scanner_cycle(
Some(usage_persist_baseline),
)
.await
});
}));
let done_cycle = Metrics::time(Metric::ScanCycle);
let cycle_work_start = global_metrics().start_scan_cycle_work();
let cycle_budget = ScannerCycleBudget::new(ctx, cycle_budget_config);
let scan_result = storeapi
.clone()
@@ -2640,7 +2691,6 @@ async fn run_data_scanner_cycle(
}
};
let unresolved_heal_work = global_metrics().current_scan_cycle_has_unresolved_heal_work();
global_metrics().finish_scan_cycle_work(cycle_work_start);
let scan_cycle_result = match scan_result {
Ok(result) => result,
@@ -2663,7 +2713,7 @@ async fn run_data_scanner_cycle(
{
save_background_heal_info(storeapi.clone(), new_heal_info).await;
}
mark_scan_cycle_idle(cycle_info).await;
mark_scan_cycle_idle(cycle_info, &mut cycle_metrics_guard).await;
return ScannerCycleOutcome::Failed;
}
};
@@ -2678,7 +2728,7 @@ async fn run_data_scanner_cycle(
"Scanner cycle stopped before committing cycle state"
);
emit_scan_cycle_complete(false, cycle_start.elapsed());
mark_scan_cycle_idle(cycle_info).await;
mark_scan_cycle_idle(cycle_info, &mut cycle_metrics_guard).await;
return ScannerCycleOutcome::Failed;
}
if let Some(required_cycle) = scan_cycle_result.required_cycle_floor() {
@@ -2700,6 +2750,7 @@ async fn run_data_scanner_cycle(
cycle_revision,
leader_epoch,
required_cycle,
&mut cycle_metrics_guard,
)
.await
{
@@ -2719,7 +2770,7 @@ async fn run_data_scanner_cycle(
"Scanner cycle completed without a durable data usage snapshot"
);
emit_scan_cycle_complete(false, cycle_start.elapsed());
mark_scan_cycle_idle(cycle_info).await;
mark_scan_cycle_idle(cycle_info, &mut cycle_metrics_guard).await;
return ScannerCycleOutcome::Failed;
}
if budget_elapsed {
@@ -2743,7 +2794,16 @@ async fn run_data_scanner_cycle(
scan_cycle_partial_reason(budget_reason),
scan_cycle_partial_source(budget_reason),
);
return if finalize_partial_scan_cycle(ctx, storeapi.clone(), cycle_info, cycle_revision, leader_epoch).await {
return if finalize_partial_scan_cycle(
ctx,
storeapi.clone(),
cycle_info,
cycle_revision,
leader_epoch,
&mut cycle_metrics_guard,
)
.await
{
ScannerCycleOutcome::Partial
} else {
ScannerCycleOutcome::Failed
@@ -2792,7 +2852,7 @@ async fn run_data_scanner_cycle(
"Scanner cycle completed without a durable data usage snapshot"
);
emit_scan_cycle_complete(false, cycle_start.elapsed());
mark_scan_cycle_idle(cycle_info).await;
mark_scan_cycle_idle(cycle_info, &mut cycle_metrics_guard).await;
return ScannerCycleOutcome::Failed;
}
ScannerCycleOutcome::Partial => {
@@ -2818,7 +2878,16 @@ async fn run_data_scanner_cycle(
);
}
emit_scan_cycle_partial_with_source(cycle_start.elapsed(), ScanCyclePartialReason::Unknown, None);
return if finalize_partial_scan_cycle(ctx, storeapi.clone(), cycle_info, cycle_revision, leader_epoch).await {
return if finalize_partial_scan_cycle(
ctx,
storeapi.clone(),
cycle_info,
cycle_revision,
leader_epoch,
&mut cycle_metrics_guard,
)
.await
{
ScannerCycleOutcome::Partial
} else {
ScannerCycleOutcome::Failed
@@ -2834,7 +2903,16 @@ async fn run_data_scanner_cycle(
state = "superseded",
"Scanner cycle usage snapshot was superseded by concurrent namespace activity"
);
if finalize_partial_scan_cycle(ctx, storeapi.clone(), cycle_info, cycle_revision, leader_epoch).await {
if finalize_partial_scan_cycle(
ctx,
storeapi.clone(),
cycle_info,
cycle_revision,
leader_epoch,
&mut cycle_metrics_guard,
)
.await
{
emit_scan_cycle_superseded(cycle_start.elapsed());
return ScannerCycleOutcome::Superseded;
}
@@ -2853,7 +2931,7 @@ async fn run_data_scanner_cycle(
error = %err,
"Scanner completed cycle could not advance"
);
mark_scan_cycle_idle(cycle_info).await;
mark_scan_cycle_idle(cycle_info, &mut cycle_metrics_guard).await;
emit_scan_cycle_complete(false, cycle_start.elapsed());
return ScannerCycleOutcome::Failed;
}
@@ -2862,9 +2940,8 @@ async fn run_data_scanner_cycle(
global_metrics().clear_current_scan_mode();
retain_recent_cycle_completions(&mut cycle_info.cycle_completed);
global_metrics().set_cycle(Some(cycle_info.clone())).await;
if !persist_scanner_cycle_state(ctx, storeapi.clone(), cycle_info, cycle_revision, leader_epoch).await {
mark_scan_cycle_idle(cycle_info).await;
cycle_metrics_guard.finish(cycle_info.clone()).await;
emit_scan_cycle_complete(false, cycle_start.elapsed());
return ScannerCycleOutcome::Failed;
}
@@ -2888,9 +2965,37 @@ async fn run_data_scanner_cycle(
"Scanner cycle completed"
);
cycle_metrics_guard.finish(cycle_info.clone()).await;
scanner_cycle_outcome_with_pending_maintenance(ScannerCycleOutcome::Completed, pending_maintenance_work)
}
struct ScannerCycleMetricsGuard {
start: Option<ScanCycleWorkSnapshot>,
}
impl ScannerCycleMetricsGuard {
async fn new(cycle: CurrentCycle) -> Self {
Self {
start: Some(global_metrics().start_scan_cycle_work_with_cycle(cycle).await),
}
}
async fn finish(&mut self, cycle: CurrentCycle) {
if let Some(start) = self.start {
global_metrics().finish_scan_cycle_work_with_cycle(start, cycle).await;
self.start = None;
}
}
}
impl Drop for ScannerCycleMetricsGuard {
fn drop(&mut self) {
if let Some(start) = self.start.take() {
global_metrics().finish_scan_cycle_work(start);
}
}
}
async fn record_scanner_leader_lock_lost(message: &'static str) {
reset_scanner_cycle_schedule();
record_scanner_leader_lock_state("lost");
@@ -3443,7 +3548,7 @@ enum DataUsagePersistTaskResult {
async fn wait_for_data_usage_persist_task(
ctx: &CancellationToken,
task: &mut tokio::task::JoinHandle<DataUsagePersistOutcome>,
task: &mut AbortOnDropHandle<DataUsagePersistOutcome>,
timeout: Duration,
) -> DataUsagePersistTaskResult {
tokio::select! {
@@ -3840,8 +3945,9 @@ mod tests {
use super::*;
use crate::EcstoreResult;
use crate::{
ScannerGetObjectReader as GetObjectReader, ScannerObjectInfo as ObjectInfo, ScannerObjectOptions as ObjectOptions,
ScannerPutObjReader as PutObjReader,
Endpoint, EndpointServerPools, Endpoints, InstanceContext, PoolEndpoints, ScannerGetObjectReader as GetObjectReader,
ScannerObjectInfo as ObjectInfo, ScannerObjectOptions as ObjectOptions, ScannerPutObjReader as PutObjReader,
init_bucket_metadata_sys_for_scanner_tests, init_ecstore_config_for_scanner_tests, init_local_disks_with_instance_ctx,
};
use serial_test::serial;
use std::collections::HashMap;
@@ -3853,6 +3959,47 @@ mod tests {
const TEST_DEFAULT_SCANNER_CYCLE_SECS: u64 = 24 * 60 * 60;
async fn setup_scanner_cycle_store() -> (tempfile::TempDir, Arc<ECStore>) {
init_ecstore_config_for_scanner_tests();
let temp_dir = tempfile::tempdir().expect("scanner cycle test directory should be created");
let mut endpoints = Vec::new();
for disk_index in 0..4 {
let disk_path = temp_dir.path().join(format!("disk{disk_index}"));
tokio::fs::create_dir_all(&disk_path)
.await
.expect("scanner cycle test disk should be created");
let mut endpoint =
Endpoint::try_from(disk_path.to_str().expect("disk path should be utf8")).expect("endpoint should parse");
endpoint.set_pool_index(0);
endpoint.set_set_index(0);
endpoint.set_disk_index(disk_index);
endpoints.push(endpoint);
}
let endpoint_pools = EndpointServerPools::from(vec![PoolEndpoints {
legacy: false,
set_count: 1,
drives_per_set: 4,
endpoints: Endpoints::from(endpoints),
cmd_line: "scanner-cycle-metrics".to_string(),
platform: format!("OS: {} | Arch: {}", std::env::consts::OS, std::env::consts::ARCH),
}]);
let instance_ctx = Arc::new(InstanceContext::new());
init_local_disks_with_instance_ctx(&instance_ctx, endpoint_pools.clone())
.await
.expect("scanner cycle test disks should initialize");
let store = ECStore::new_with_instance_ctx(
"127.0.0.1:0".parse().expect("test address should parse"),
endpoint_pools,
CancellationToken::new(),
instance_ctx,
)
.await
.expect("scanner cycle test ECStore should initialize");
init_bucket_metadata_sys_for_scanner_tests(store.clone()).await;
(temp_dir, store)
}
fn assert_run_data_scanner_signature<F, Fut>(_run: F)
where
F: Fn(CancellationToken, Arc<ECStore>) -> Fut,
@@ -4312,9 +4459,9 @@ mod tests {
};
global_metrics().set_current_scan_mode(HealScanMode::Deep);
global_metrics().set_cycle(Some(cycle_info.clone())).await;
let mut cycle_metrics_guard = ScannerCycleMetricsGuard::new(cycle_info.clone()).await;
mark_scan_cycle_idle(&mut cycle_info).await;
mark_scan_cycle_idle(&mut cycle_info, &mut cycle_metrics_guard).await;
let published = global_metrics()
.get_cycle()
@@ -4330,6 +4477,123 @@ mod tests {
global_metrics().set_cycle(None).await;
}
#[tokio::test]
#[serial]
async fn scanner_cycle_metrics_guard_covers_published_first_cycle_lifetime() {
let cycle_started = Utc::now() - chrono::Duration::seconds(5);
let mut cycle_info = CurrentCycle {
current: 0,
next: 1,
started: cycle_started,
..Default::default()
};
let mut guard = ScannerCycleMetricsGuard::new(cycle_info.clone()).await;
let setup_report = global_metrics().report().await;
assert!(setup_report.current_cycle_active);
assert_eq!(setup_report.current_cycle, 0);
assert_eq!(setup_report.current_started, cycle_started);
mark_scan_cycle_idle(&mut cycle_info, &mut guard).await;
let idle_report = global_metrics().report().await;
assert!(!idle_report.current_cycle_active);
global_metrics().set_cycle(None).await;
}
#[tokio::test]
#[serial]
async fn scanner_cycle_metrics_guard_keeps_active_cycle_published_during_finalization() {
let mut cycle_info = CurrentCycle {
current: 12,
next: 13,
started: Utc::now(),
..Default::default()
};
let mut guard = ScannerCycleMetricsGuard::new(cycle_info.clone()).await;
cycle_info.current = 0;
tokio::task::yield_now().await;
let finalizing_report = global_metrics().report().await;
assert!(finalizing_report.current_cycle_active);
assert_eq!(finalizing_report.current_cycle, 12);
guard.finish(cycle_info).await;
let idle_report = global_metrics().report().await;
assert!(!idle_report.current_cycle_active);
assert_eq!(idle_report.current_cycle, 0);
global_metrics().set_cycle(None).await;
}
#[tokio::test]
#[serial]
async fn scanner_cycle_metrics_guard_drop_clears_activity() {
let guard = ScannerCycleMetricsGuard::new(CurrentCycle {
current: 12,
next: 13,
started: Utc::now(),
..Default::default()
})
.await;
assert!(global_metrics().report().await.current_cycle_active);
drop(guard);
assert!(!global_metrics().report().await.current_cycle_active);
global_metrics().set_cycle(None).await;
}
#[tokio::test]
#[serial]
async fn run_data_scanner_cycle_publishes_activity_for_owner_lifetime() {
let (_temp_dir, store) = setup_scanner_cycle_store().await;
let ctx = CancellationToken::new();
let mut cycle_info = CurrentCycle::default();
let mut revision = DataUsageCacheRevision::Missing;
let leader_epoch = u64::MAX - 1;
let state_persist_reached = Arc::new(Notify::new());
let _state_persist_hook = set_scanner_cycle_state_persist_test_hook(leader_epoch, state_persist_reached.clone());
let state_lock = store
.new_ns_lock(RUSTFS_META_BUCKET, DATA_USAGE_BLOOM_NAME_PATH.as_str())
.await
.expect("scanner cycle state lock should be created");
let state_guard = state_lock
.get_write_lock(Duration::from_secs(1))
.await
.expect("scanner cycle state lock should be acquired");
let mut cycle = Box::pin(run_data_scanner_cycle(&ctx, &store, &mut cycle_info, &mut revision, leader_epoch));
let waker = std::task::Waker::noop();
let mut context = std::task::Context::from_waker(waker);
assert!(cycle.as_mut().poll(&mut context).is_pending());
let active = global_metrics().report().await;
assert!(active.current_cycle_active);
assert_eq!(active.current_cycle, 0);
tokio::time::timeout(Duration::from_secs(30), async {
tokio::select! {
outcome = &mut cycle => panic!("scanner cycle finished before state persistence was released: {outcome:?}"),
_ = state_persist_reached.notified() => {}
}
})
.await
.expect("scanner cycle should reach state persistence");
let finalizing = global_metrics().report().await;
assert!(finalizing.current_cycle_active);
drop(state_guard);
let outcome = tokio::time::timeout(Duration::from_secs(30), cycle)
.await
.expect("scanner cycle should finish");
assert!(matches!(
outcome,
ScannerCycleOutcome::Completed | ScannerCycleOutcome::CompletedWithPendingMaintenance
));
assert!(!global_metrics().report().await.current_cycle_active);
global_metrics().set_cycle(None).await;
}
#[tokio::test]
#[serial]
async fn test_finalize_partial_scan_cycle_advances_and_persists_counter() {
@@ -4342,8 +4606,11 @@ mod tests {
cycle_completed: vec![],
started: Utc::now(),
};
let mut cycle_metrics_guard = ScannerCycleMetricsGuard::new(cycle_info.clone()).await;
assert!(finalize_partial_scan_cycle(&ctx, store.clone(), &mut cycle_info, &mut revision, 1).await);
assert!(
finalize_partial_scan_cycle(&ctx, store.clone(), &mut cycle_info, &mut revision, 1, &mut cycle_metrics_guard,).await
);
assert_eq!(cycle_info.next, 13);
assert_eq!(cycle_info.current, 0);
@@ -4377,8 +4644,20 @@ mod tests {
cycle_completed: vec![],
started: Utc::now(),
};
let mut cycle_metrics_guard = ScannerCycleMetricsGuard::new(cycle_info.clone()).await;
assert!(persist_required_scanner_cycle_floor(&ctx, store.clone(), &mut cycle_info, &mut revision, 7, 19).await);
assert!(
persist_required_scanner_cycle_floor(
&ctx,
store.clone(),
&mut cycle_info,
&mut revision,
7,
19,
&mut cycle_metrics_guard,
)
.await
);
assert_eq!(cycle_info.current, 0);
assert_eq!(cycle_info.next, 19);
@@ -4405,22 +4684,37 @@ mod tests {
cycle_completed: vec![],
started: Utc::now(),
};
let mut cycle_metrics_guard = ScannerCycleMetricsGuard::new(cycle_info.clone()).await;
assert!(!persist_required_scanner_cycle_floor(&ctx, store.clone(), &mut cycle_info, &mut revision, 7, 12).await);
assert_eq!(cycle_info.next, 12);
assert_eq!(revision, DataUsageCacheRevision::Missing);
assert!(
!persist_required_scanner_cycle_floor(
&ctx,
store.clone(),
&mut CurrentCycle {
current: 12,
next: 12,
..Default::default()
},
&mut cycle_info,
&mut revision,
7,
12,
&mut cycle_metrics_guard,
)
.await
);
assert_eq!(cycle_info.next, 12);
assert_eq!(revision, DataUsageCacheRevision::Missing);
let mut max_cycle_info = CurrentCycle {
current: 12,
next: 12,
..Default::default()
};
let mut max_cycle_metrics_guard = ScannerCycleMetricsGuard::new(max_cycle_info.clone()).await;
assert!(
!persist_required_scanner_cycle_floor(
&ctx,
store.clone(),
&mut max_cycle_info,
&mut revision,
7,
u64::MAX,
&mut max_cycle_metrics_guard,
)
.await
);
@@ -4685,8 +4979,9 @@ mod tests {
cycle_completed: vec![],
started: Utc::now(),
};
let mut cycle_metrics_guard = ScannerCycleMetricsGuard::new(cycle_info.clone()).await;
assert!(!finalize_partial_scan_cycle(&ctx, store, &mut cycle_info, &mut revision, 1).await);
assert!(!finalize_partial_scan_cycle(&ctx, store, &mut cycle_info, &mut revision, 1, &mut cycle_metrics_guard,).await);
assert_eq!(cycle_info.next, 13);
assert_eq!(cycle_info.current, 0);
assert_eq!(revision, DataUsageCacheRevision::Missing);
@@ -5943,10 +6238,10 @@ mod tests {
#[tokio::test]
async fn data_usage_persist_wait_aborts_when_scanner_is_cancelled() {
let ctx = CancellationToken::new();
let mut task = tokio::spawn(async {
let mut task = AbortOnDropHandle::new(tokio::spawn(async {
std::future::pending::<()>().await;
DataUsagePersistOutcome::Saved
});
}));
ctx.cancel();
let result = wait_for_data_usage_persist_task(&ctx, &mut task, Duration::from_secs(60)).await;
@@ -5958,10 +6253,10 @@ mod tests {
#[tokio::test(start_paused = true)]
async fn data_usage_persist_wait_aborts_after_timeout() {
let ctx = CancellationToken::new();
let mut task = tokio::spawn(async {
let mut task = AbortOnDropHandle::new(tokio::spawn(async {
std::future::pending::<()>().await;
DataUsagePersistOutcome::Saved
});
}));
let result = wait_for_data_usage_persist_task(&ctx, &mut task, Duration::from_secs(30)).await;