fix(capacity): harden scope registry, scan symlink guard, and test temp dir cleanup (#2432)

Co-authored-by: heihutu <heihutu@gmail.com>
Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
Co-authored-by: houseme <4829346+houseme@users.noreply.github.com>

crates/io-metrics/src/capacity_metrics.rs

@@ -29,6 +29,12 @@ pub fn record_capacity_cache_miss() {
     counter!("rustfs.capacity.cache.misses").increment(1);
 }
 
+/// Record how capacity cache was served to the caller.
+#[inline(always)]
+pub fn record_capacity_cache_served(state: &'static str) {
+    counter!("rustfs.capacity.cache.served.total", "state" => state).increment(1);
+}
+
 /// Record current capacity gauge.
 #[inline(always)]
 pub fn record_capacity_current_bytes(used_bytes: u64) {
@@ -55,6 +61,44 @@ pub fn record_capacity_update_failed(source: &'static str) {
     counter!("rustfs.capacity.update.failures", "source" => source).increment(1);
 }
 
+/// Record a capacity refresh request.
+#[inline(always)]
+pub fn record_capacity_refresh_request(mode: &'static str, source: &'static str) {
+    counter!("rustfs.capacity.refresh.requests.total", "mode" => mode, "source" => source).increment(1);
+}
+
+/// Record a refresh joiner waiting for an inflight refresh.
+#[inline(always)]
+pub fn record_capacity_refresh_joiner(source: &'static str) {
+    counter!("rustfs.capacity.refresh.joiners.total", "source" => source).increment(1);
+}
+
+/// Record the number of inflight capacity refreshes.
+#[inline(always)]
+pub fn record_capacity_refresh_inflight(count: usize) {
+    gauge!("rustfs.capacity.refresh.inflight").set(count as f64);
+}
+
+/// Record the final result of a capacity refresh.
+#[inline(always)]
+pub fn record_capacity_refresh_result(source: &'static str, result: &'static str, duration: Duration) {
+    counter!("rustfs.capacity.refresh.result.total", "source" => source, "result" => result).increment(1);
+    histogram!("rustfs.capacity.refresh.duration.seconds", "source" => source, "result" => result).record(duration.as_secs_f64());
+}
+
+/// Record the refresh scope selected for a capacity refresh.
+#[inline(always)]
+pub fn record_capacity_refresh_scope(scope: &'static str, disk_count: usize) {
+    counter!("rustfs.capacity.refresh.scope.total", "scope" => scope).increment(1);
+    histogram!("rustfs.capacity.refresh.scope.disks", "scope" => scope).record(disk_count as f64);
+}
+
+/// Record the current number of dirty disks tracked by capacity management.
+#[inline(always)]
+pub fn record_capacity_dirty_disk_count(count: usize) {
+    gauge!("rustfs.capacity.dirty.disks").set(count as f64);
+}
+
 /// Record capacity write activity.
 #[inline(always)]
 pub fn record_capacity_write_operation(write_frequency: usize) {
@@ -98,3 +142,33 @@ pub fn record_capacity_scan_sampling(sampled_count: usize, estimated: bool) {
     )
     .increment(1);
 }
+
+/// Record the scan mode used for a capacity result.
+#[inline(always)]
+pub fn record_capacity_scan_mode(mode: &'static str) {
+    counter!("rustfs.capacity.scan.mode.total", "mode" => mode).increment(1);
+}
+
+/// Record per-disk capacity scan statistics.
+#[inline(always)]
+pub fn record_capacity_scan_disk(
+    disk: &str,
+    duration: Duration,
+    file_count: usize,
+    sampled_count: usize,
+    estimated: bool,
+    partial_errors: bool,
+) {
+    histogram!("rustfs.capacity.scan.disk.duration.seconds", "disk" => disk.to_owned()).record(duration.as_secs_f64());
+    histogram!("rustfs.capacity.scan.disk.files", "disk" => disk.to_owned()).record(file_count as f64);
+    histogram!("rustfs.capacity.scan.disk.sampled", "disk" => disk.to_owned()).record(sampled_count as f64);
+    counter!(
+        "rustfs.capacity.scan.disk.estimated.total",
+        "disk" => disk.to_owned(),
+        "estimated" => if estimated { "true" } else { "false" }
+    )
+    .increment(1);
+    if partial_errors {
+        counter!("rustfs.capacity.scan.disk.partial_errors.total", "disk" => disk.to_owned()).increment(1);
+    }
+}

crates/io-metrics/src/lib.rs

@@ -74,9 +74,12 @@ pub use adaptive_ttl::{
 
 // Capacity metrics exports
 pub use capacity_metrics::{
-    record_capacity_cache_hit, record_capacity_cache_miss, record_capacity_current_bytes, record_capacity_dynamic_timeout,
-    record_capacity_scan_sampling, record_capacity_stall_detected, record_capacity_symlink, record_capacity_timeout_fallback,
-    record_capacity_update_completed, record_capacity_update_failed, record_capacity_write_operation,
+    record_capacity_cache_hit, record_capacity_cache_miss, record_capacity_cache_served, record_capacity_current_bytes,
+    record_capacity_dirty_disk_count, record_capacity_dynamic_timeout, record_capacity_refresh_inflight,
+    record_capacity_refresh_joiner, record_capacity_refresh_request, record_capacity_refresh_result,
+    record_capacity_refresh_scope, record_capacity_scan_disk, record_capacity_scan_mode, record_capacity_scan_sampling,
+    record_capacity_stall_detected, record_capacity_symlink, record_capacity_timeout_fallback, record_capacity_update_completed,
+    record_capacity_update_failed, record_capacity_write_operation,
 };
 
 // I/O metrics exports