feat(scanner): add runtime scanner controls and status (#3203)

* feat(scanner): add runtime scanner controls and status

* fix(scanner): validate persisted scanner config

* docs(scanner): clarify start delay behavior

---------

Co-authored-by: Henry Guo <marshawcoco@users.noreply.github.com>
Co-authored-by: houseme <housemecn@gmail.com>
This commit is contained in:
Henry Guo
2026-06-04 13:16:26 +08:00
committed by GitHub
parent f708c22b0e
commit 6850457707
16 changed files with 1394 additions and 126 deletions
+44 -78
View File
@@ -12,16 +12,17 @@
// See the License for the specific language governing permissions and
// limitations under the License.
use std::sync::{
Arc,
atomic::{AtomicU64, Ordering},
};
use std::sync::Arc;
use crate::data_usage_define::{BACKGROUND_HEAL_INFO_PATH, DATA_USAGE_BLOOM_NAME_PATH, DATA_USAGE_OBJ_NAME_PATH};
use crate::runtime_config::{
current_scanner_runtime_config, lookup_scanner_runtime_config, refresh_scanner_runtime_config_from_global,
scanner_bitrot_cycle, scanner_cycle_interval, scanner_start_delay, set_scanner_default_cycle_secs,
};
use crate::scanner_budget::{ScannerCycleBudget, ScannerCycleBudgetConfig, ScannerCycleBudgetReason};
use crate::scanner_folder::{data_usage_update_dir_cycles, heal_object_select_prob};
use crate::scanner_io::ScannerIO;
use crate::sleeper::{SCANNER_SLEEPER, scanner_speed_from_env_or_default, set_scanner_default_speed};
use crate::sleeper::{SCANNER_SLEEPER, set_scanner_default_speed};
use crate::{DataUsageInfo, ScannerActivityGuard, ScannerError};
use chrono::{DateTime, Utc};
use rustfs_common::heal_channel::HealScanMode;
@@ -29,11 +30,12 @@ use rustfs_common::metrics::{
CurrentCycle, Metric, Metrics, ScanCyclePartialReason, emit_scan_cycle_complete, emit_scan_cycle_partial, global_metrics,
};
use rustfs_config::ScannerSpeed;
#[cfg(test)]
use rustfs_config::{
DEFAULT_SCANNER_BITROT_CYCLE_SECS, DEFAULT_SCANNER_CYCLE_MAX_DIRECTORIES, DEFAULT_SCANNER_CYCLE_MAX_DURATION_SECS,
DEFAULT_SCANNER_CYCLE_MAX_OBJECTS, ENV_SCANNER_BITROT_CYCLE_SECS, ENV_SCANNER_CYCLE, ENV_SCANNER_CYCLE_MAX_DIRECTORIES,
ENV_SCANNER_CYCLE_MAX_DURATION_SECS, ENV_SCANNER_CYCLE_MAX_OBJECTS, ENV_SCANNER_SPEED, ENV_SCANNER_START_DELAY_SECS,
ENV_SCANNER_BITROT_CYCLE_SECS, ENV_SCANNER_CYCLE_MAX_DIRECTORIES, ENV_SCANNER_CYCLE_MAX_DURATION_SECS,
ENV_SCANNER_CYCLE_MAX_OBJECTS,
};
use rustfs_config::{ENV_SCANNER_CYCLE, ENV_SCANNER_SPEED, ENV_SCANNER_START_DELAY_SECS};
use rustfs_ecstore::StorageAPI as _;
use rustfs_ecstore::bucket::lifecycle::lifecycle::Lifecycle as _;
use rustfs_ecstore::bucket::metadata_sys::{get_lifecycle_config, get_replication_config};
@@ -50,11 +52,9 @@ use tokio::time::{Duration, Instant};
use tokio_util::sync::CancellationToken;
use tracing::{debug, error, info, instrument, warn};
const ENV_SCANNER_START_DELAY_SECS_DEPRECATED: &str = "RUSTFS_DATA_SCANNER_START_DELAY_SECS";
const SINGLE_DISK_SCANNER_CYCLE_SECS: u64 = 24 * 60 * 60;
const NO_DEFAULT_CYCLE_OVERRIDE: u64 = 0;
static SCANNER_DEFAULT_CYCLE_SECS: AtomicU64 = AtomicU64::new(NO_DEFAULT_CYCLE_OVERRIDE);
#[cfg(test)]
const ENV_SCANNER_START_DELAY_SECS_DEPRECATED: &str = "RUSTFS_DATA_SCANNER_START_DELAY_SECS";
/// Returns the base cycle interval.
/// Priority order:
@@ -62,54 +62,28 @@ static SCANNER_DEFAULT_CYCLE_SECS: AtomicU64 = AtomicU64::new(NO_DEFAULT_CYCLE_O
/// 2. RUSTFS_SCANNER_START_DELAY_SECS (for backward compatibility)
/// 3. Deployment-specific default cycle override
/// 4. RUSTFS_SCANNER_SPEED preset
#[cfg(test)]
fn cycle_interval() -> Duration {
if let Some(secs) = rustfs_utils::get_env_opt_u64(ENV_SCANNER_CYCLE) {
return Duration::from_secs(secs);
}
if let Some(secs) = scanner_start_delay_secs() {
return Duration::from_secs(secs);
}
if let Some(secs) = scanner_default_cycle_secs() {
return Duration::from_secs(secs);
}
scanner_speed_from_env_or_default().cycle_interval()
}
fn scanner_default_cycle_secs() -> Option<u64> {
match SCANNER_DEFAULT_CYCLE_SECS.load(Ordering::Relaxed) {
NO_DEFAULT_CYCLE_OVERRIDE => None,
secs => Some(secs),
}
}
fn set_scanner_default_cycle_secs(secs: Option<u64>) {
SCANNER_DEFAULT_CYCLE_SECS.store(secs.unwrap_or(NO_DEFAULT_CYCLE_OVERRIDE), Ordering::Relaxed);
}
fn scanner_start_delay_secs() -> Option<u64> {
let deprecated = [ENV_SCANNER_START_DELAY_SECS_DEPRECATED];
rustfs_utils::get_env_opt_u64_with_aliases(ENV_SCANNER_START_DELAY_SECS, &deprecated)
}
fn scanner_cycle_max_duration() -> Option<Duration> {
match rustfs_utils::get_env_u64(ENV_SCANNER_CYCLE_MAX_DURATION_SECS, DEFAULT_SCANNER_CYCLE_MAX_DURATION_SECS) {
0 => None,
secs => Some(Duration::from_secs(secs)),
}
}
fn scanner_cycle_count_budget(env: &str, default: u64) -> Option<u64> {
match rustfs_utils::get_env_u64(env, default) {
0 => None,
count => Some(count),
}
resolve_scanner_runtime_config().cycle_interval
}
fn scanner_cycle_budget_config() -> ScannerCycleBudgetConfig {
ScannerCycleBudgetConfig {
max_duration: scanner_cycle_max_duration(),
max_objects: scanner_cycle_count_budget(ENV_SCANNER_CYCLE_MAX_OBJECTS, DEFAULT_SCANNER_CYCLE_MAX_OBJECTS),
max_directories: scanner_cycle_count_budget(ENV_SCANNER_CYCLE_MAX_DIRECTORIES, DEFAULT_SCANNER_CYCLE_MAX_DIRECTORIES),
resolve_scanner_runtime_config().cycle_budget
}
#[cfg(test)]
fn scanner_cycle_max_duration() -> Option<Duration> {
resolve_scanner_runtime_config().cycle_budget.max_duration
}
fn resolve_scanner_runtime_config() -> crate::runtime_config::ScannerRuntimeConfig {
let config = rustfs_ecstore::config::get_global_server_config();
match lookup_scanner_runtime_config(config.as_ref()) {
Ok(config) => config,
Err(err) => {
warn!(error = %err, "Failed to resolve scanner runtime config, using last applied config");
current_scanner_runtime_config()
}
}
}
@@ -125,7 +99,7 @@ fn scan_cycle_partial_reason(reason: Option<ScannerCycleBudgetReason>) -> ScanCy
/// Compute a randomized inter-cycle sleep.
// Delay is scan interval +- 10%, with a floor of 1 second.
fn randomized_cycle_delay() -> Duration {
randomized_cycle_delay_for(cycle_interval())
randomized_cycle_delay_for(scanner_cycle_interval())
}
fn randomized_cycle_delay_for(interval: Duration) -> Duration {
@@ -137,7 +111,7 @@ fn randomized_cycle_delay_for(interval: Duration) -> Duration {
}
fn initial_scanner_delay() -> Duration {
initial_scanner_delay_for(scanner_start_delay_secs())
initial_scanner_delay_for(scanner_start_delay().map(|duration| duration.as_secs()))
}
fn initial_scanner_delay_for(start_delay_secs: Option<u64>) -> Duration {
@@ -150,6 +124,9 @@ pub async fn init_data_scanner(ctx: CancellationToken, storeapi: Arc<ECStore>) {
configure_scanner_defaults(&storeapi).await;
// Force init global sleeper so config is read once at startup.
let _ = &*SCANNER_SLEEPER;
if let Err(err) = refresh_scanner_runtime_config_from_global() {
warn!(error = %err, "Failed to apply scanner runtime config at startup");
}
let ctx_clone = ctx;
let storeapi_clone = storeapi;
@@ -281,24 +258,9 @@ async fn configure_scanner_defaults(storeapi: &Arc<ECStore>) {
}
}
#[cfg(test)]
fn bitrot_scan_cycle() -> Option<Duration> {
let Ok(value) = std::env::var(ENV_SCANNER_BITROT_CYCLE_SECS) else {
return Some(Duration::from_secs(DEFAULT_SCANNER_BITROT_CYCLE_SECS));
};
match value.trim().to_ascii_lowercase().as_str() {
"0" | "true" | "on" | "yes" => Some(Duration::ZERO),
"false" | "off" | "no" | "disabled" => None,
value => value.parse::<u64>().ok().map(Duration::from_secs).or_else(|| {
warn!(
env = ENV_SCANNER_BITROT_CYCLE_SECS,
value,
default_secs = DEFAULT_SCANNER_BITROT_CYCLE_SECS,
"Invalid scanner bitrot cycle, using default"
);
Some(Duration::from_secs(DEFAULT_SCANNER_BITROT_CYCLE_SECS))
}),
}
resolve_scanner_runtime_config().bitrot_cycle
}
fn get_cycle_scan_mode(
@@ -476,9 +438,11 @@ async fn mark_scan_cycle_idle(cycle_info: &mut CurrentCycle) {
#[instrument(skip_all)]
async fn run_data_scanner_cycle(ctx: &CancellationToken, storeapi: &Arc<ECStore>, cycle_info: &mut CurrentCycle) {
let _activity_guard = ScannerActivityGuard::new();
SCANNER_SLEEPER.refresh_from_env();
let configured_cycle_interval = cycle_interval();
let configured_bitrot_cycle = bitrot_scan_cycle();
if let Err(err) = refresh_scanner_runtime_config_from_global() {
warn!(error = %err, "Failed to refresh scanner runtime config, using last applied config");
}
let configured_cycle_interval = scanner_cycle_interval();
let configured_bitrot_cycle = scanner_bitrot_cycle();
let cycle_budget_config = scanner_cycle_budget_config();
global_metrics().record_scanner_cycle_config(
configured_cycle_interval,
@@ -790,10 +754,12 @@ mod tests {
#[serial]
fn test_initial_scanner_delay_uses_cycle_without_explicit_start_delay() {
with_var(ENV_SCANNER_CYCLE, Some("120"), || {
crate::runtime_config::refresh_scanner_runtime_config_for_tests();
let delay = initial_scanner_delay_for(None);
assert!(delay >= Duration::from_secs(108));
assert!(delay <= Duration::from_secs(132));
});
crate::runtime_config::refresh_scanner_runtime_config_for_tests();
}
#[test]