refactor(metrics): separate console stream from exports (#2583)

Cargo.lock

@@ -8082,6 +8082,7 @@ dependencies = [
 name = "rustfs-io-metrics"
 version = "0.0.5"
 dependencies = [
+ "criterion",
  "metrics",
  "num_cpus",
  "thiserror 2.0.18",

crates/io-metrics/Cargo.toml

@@ -24,6 +24,10 @@ description = "Metrics collection and reporting for RustFS (using metrics crate
 keywords = ["metrics", "zero-copy", "rustfs", "otel", "performance"]
 categories = ["development-tools", "filesystem"]
 
+[[bench]]
+name = "metrics_pipeline"
+harness = false
+
 [dependencies]
 metrics = { workspace = true }
 num_cpus = { workspace = true }
@@ -32,6 +36,7 @@ tokio = { workspace = true, features = ["sync","rt"] }
 tracing = { workspace = true }
 
 [dev-dependencies]
+criterion = { workspace = true }
 tokio = { workspace = true, features = ["test-util","rt","macros"] }
 
 [lints]

crates/io-metrics/README.md

@@ -31,6 +31,7 @@
 - **Bandwidth Monitoring**: Real-time bandwidth observation and analysis
 - **Performance Metrics**: I/O performance metrics collection
 - **Unified Configuration**: Centralized configuration management
+- **Exporter Boundary**: Emit via `metrics`, export via `rustfs-obs`, no Prometheus HTTP endpoint
 
 ## Features
 
@@ -199,7 +200,7 @@ cargo test --package rustfs-io-metrics
 cargo test --package rustfs-io-metrics --lib adaptive_ttl
 
 # Run benchmarks
-cargo bench --package rustfs-io-metrics
+cargo bench --package rustfs-io-metrics --bench metrics_pipeline
 ```
 
 ## Documentation

crates/io-metrics/README_zh.md

@@ -31,6 +31,7 @@
 - **带宽监控**：实时带宽观测和分析
 - **性能指标**：I/O 性能指标收集
 - **统一配置**：集中式配置管理
+- **导出边界**：通过 `metrics` 主动上报，由 `rustfs-obs` 负责 OTEL 导出，不提供 Prometheus HTTP 端点
 
 ## ✨ 核心功能
 
@@ -280,7 +281,7 @@ cargo test --package rustfs-io-metrics
 cargo test --package rustfs-io-metrics --lib adaptive_ttl
 
 # 运行基准测试
-cargo bench --package rustfs-io-metrics
+cargo bench --package rustfs-io-metrics --bench metrics_pipeline
 ```
 
 ## 📚 文档

crates/io-metrics/benches/metrics_pipeline.rs

@@ -0,0 +1,43 @@
+use criterion::{Criterion, criterion_group, criterion_main};
+use rustfs_io_metrics::{MetricsCollector, PerformanceMetrics, record_get_object_request_started};
+use std::hint::black_box;
+use std::sync::Arc;
+use std::time::Duration;
+
+fn bench_record_get_object_request_started(c: &mut Criterion) {
+    c.bench_function("record_get_object_request_started", |b| b.iter(record_get_object_request_started));
+}
+
+fn bench_update_concurrent_requests(c: &mut Criterion) {
+    let metrics = PerformanceMetrics::new();
+
+    c.bench_function("performance_metrics_update_concurrent_requests", |b| {
+        b.iter(|| metrics.update_concurrent_requests(black_box(64)))
+    });
+}
+
+fn bench_metrics_collector_record_io_operation(c: &mut Criterion) {
+    let runtime = tokio::runtime::Builder::new_current_thread()
+        .enable_all()
+        .build()
+        .expect("build tokio runtime for io-metrics benchmark");
+    let collector = MetricsCollector::new(Arc::new(PerformanceMetrics::new()), 256);
+
+    c.bench_function("metrics_collector_record_io_operation", |b| {
+        b.iter(|| {
+            runtime.block_on(collector.record_io_operation(
+                black_box(64 * 1024),
+                Duration::from_micros(black_box(250)),
+                black_box(true),
+            ))
+        })
+    });
+}
+
+criterion_group!(
+    benches,
+    bench_record_get_object_request_started,
+    bench_update_concurrent_requests,
+    bench_metrics_collector_record_io_operation
+);
+criterion_main!(benches);

crates/io-metrics/examples/metrics_example.rs

@@ -143,7 +143,7 @@ fn metrics_recording_example() {
 
     println!("  Recorded 10 cache operations (hits: 7, misses: 3)");
     println!("  Metrics reported via metrics crate");
-    println!("  View via Prometheus/Grafana");
+    println!("  Export via rustfs-obs OTEL pipeline");
 
     println!();
 }

crates/io-metrics/src/lib.rs

@@ -23,6 +23,8 @@
 //! - **PerformanceMetrics**: Shared atomic counter struct for advanced use cases
 //! - **MetricsCollector**: I/O operation tracking with percentile calculation
 //! - **AutoTuner**: Automatic performance optimization based on metrics
+//! - **No HTTP metrics endpoint**: consumers emit metrics through the `metrics` crate;
+//!   `rustfs-obs` owns OTEL initialization and export
 //!
 //! # Usage
 //!

rustfs/src/admin/handlers/metrics.rs

@@ -12,22 +12,29 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+//! Console realtime metrics API.
+//!
+//! This preserves the console's fixed `/admin/v3/metrics` contract while
+//! keeping the response format explicitly NDJSON. It is not a Prometheus text
+//! exposition endpoint.
+
 use crate::admin::router::Operation;
 use bytes::Bytes;
 use futures::{Stream, StreamExt};
-use http::Uri;
+use http::{HeaderMap, HeaderValue, Uri};
 use hyper::StatusCode;
 use matchit::Params;
 use rustfs_ecstore::metrics_realtime::{CollectMetricsOpts, MetricType, collect_local_metrics};
 use rustfs_madmin::metrics::RealtimeMetrics;
 use rustfs_madmin::utils::parse_duration;
+use s3s::header::CONTENT_TYPE;
 use s3s::stream::{ByteStream, DynByteStream};
 use s3s::{Body, S3Request, S3Response, S3Result, StdError, s3_error};
 use serde::{Deserialize, Serialize};
 use std::collections::{HashMap, HashSet};
 use std::pin::Pin;
 use std::task::{Context, Poll};
-use std::time::Duration as std_Duration;
+use std::time::Duration as StdDuration;
 use tokio::sync::mpsc;
 use tokio::time::interval;
 use tokio::{select, spawn};
@@ -36,6 +43,7 @@ use tracing::{debug, error, warn};
 
 const DEFAULT_METRICS_SAMPLES: u64 = 1;
 const MAX_METRICS_SAMPLES: u64 = 120;
+const CONSOLE_METRICS_CONTENT_TYPE: &str = "application/x-ndjson";
 
 #[derive(Debug, Serialize, Deserialize)]
 struct MetricsParams {
@@ -173,14 +181,15 @@ pub struct MetricsHandler {}
 impl Operation for MetricsHandler {
     async fn call(&self, req: S3Request<Body>, params: Params<'_, '_>) -> S3Result<S3Response<(StatusCode, Body)>> {
         debug!("handle MetricsHandler, uri: {:?}, params: {:?}", req.uri, params);
-        let Some(_cred) = req.credentials else { return Err(s3_error!(InvalidRequest, "get cred failed")) };
-        debug!("validated metrics request credentials");
+        let Some(_cred) = req.credentials else {
+            return Err(s3_error!(InvalidRequest, "get cred failed"));
+        };
+        debug!("validated console metrics request credentials");
 
         let mp = extract_metrics_init_params(&req.uri);
         debug!("mp: {:?}", mp);
 
-        let tick = parse_duration(&mp.tick).unwrap_or_else(|_| std_Duration::from_secs(3));
-
+        let tick = parse_duration(&mp.tick).unwrap_or_else(|_| StdDuration::from_secs(3));
         let mut n = resolve_sample_count(&mp);
 
         let types = if mp.types != 0 {
@@ -193,53 +202,43 @@ impl Operation for MetricsHandler {
             s.split(',').filter(|part| !part.is_empty()).map(String::from).collect()
         }
 
-        let disks = parse_comma_separated(&mp.disks);
         let by_disk = mp.by_disk == "true";
-        let disk_map = disks;
-
-        let job_id = mp.by_job_id;
-        let hosts = parse_comma_separated(&mp.hosts);
         let by_host = mp.by_host == "true";
-        let host_map = hosts;
-
-        let d_id = mp.by_dep_id;
         let mut interval = interval(tick);
-
         let opts = CollectMetricsOpts {
-            hosts: host_map,
-            disks: disk_map,
-            job_id,
-            dep_id: d_id,
+            hosts: parse_comma_separated(&mp.hosts),
+            disks: parse_comma_separated(&mp.disks),
+            job_id: mp.by_job_id,
+            dep_id: mp.by_dep_id,
         };
         let (tx, rx) = mpsc::channel(10);
         let in_stream: DynByteStream = Box::pin(MetricsStream {
             inner: ReceiverStream::new(rx),
         });
         let body = Body::from(in_stream);
+
         spawn(async move {
             while n > 0 {
-                let mut m = RealtimeMetrics::default();
-                let m_local = collect_local_metrics(types, &opts).await;
-                m.merge(m_local);
+                let mut metrics = RealtimeMetrics::default();
+                let local_metrics = collect_local_metrics(types, &opts).await;
+                metrics.merge(local_metrics);
 
                 if !by_host {
-                    m.by_host = HashMap::new();
+                    metrics.by_host = HashMap::new();
                 }
                 if !by_disk {
-                    m.by_disk = HashMap::new();
+                    metrics.by_disk = HashMap::new();
                 }
 
-                m.finally = n <= 1;
+                metrics.finally = n <= 1;
 
-                // todo write resp
-                match serde_json::to_vec(&m) {
-                    Ok(mut re) => {
-                        // NDJSON framing allows stream clients to parse incremental records.
-                        re.push(b'\n');
-                        let _ = tx.send(Ok(Bytes::from(re))).await;
+                match serde_json::to_vec(&metrics) {
+                    Ok(mut encoded) => {
+                        encoded.push(b'\n');
+                        let _ = tx.send(Ok(Bytes::from(encoded))).await;
                     }
-                    Err(e) => {
-                        error!("MetricsHandler: json encode failed, err: {:?}", e);
+                    Err(err) => {
+                        error!("MetricsHandler: json encode failed, err: {:?}", err);
                         return;
                     }
                 }
@@ -256,13 +255,19 @@ impl Operation for MetricsHandler {
             }
         });
 
-        Ok(S3Response::new((StatusCode::OK, body)))
+        let mut header = HeaderMap::new();
+        header.insert(CONTENT_TYPE, HeaderValue::from_static(CONSOLE_METRICS_CONTENT_TYPE));
+
+        Ok(S3Response::with_headers((StatusCode::OK, body), header))
     }
 }
 
 #[cfg(test)]
 mod tests {
-    use super::{DEFAULT_METRICS_SAMPLES, MAX_METRICS_SAMPLES, extract_metrics_init_params, resolve_sample_count};
+    use super::{
+        CONSOLE_METRICS_CONTENT_TYPE, DEFAULT_METRICS_SAMPLES, MAX_METRICS_SAMPLES, extract_metrics_init_params,
+        resolve_sample_count,
+    };
     use http::Uri;
 
     #[test]
@@ -288,4 +293,9 @@ mod tests {
 
         assert_eq!(resolve_sample_count(&mp), MAX_METRICS_SAMPLES);
     }
+
+    #[test]
+    fn metrics_handler_uses_ndjson_content_type() {
+        assert_eq!(CONSOLE_METRICS_CONTENT_TYPE, "application/x-ndjson");
+    }
 }

rustfs/src/storage/backpressure.rs

@@ -24,7 +24,7 @@
 //!
 //! - Configurable buffer size with high/low watermarks
 //! - Backpressure state monitoring and events
-//! - Prometheus metrics for backpressure events
+//! - Backpressure metrics emitted through the shared metrics pipeline
 //! - Graceful handling of slow consumers
 //!
 //! # Architecture

rustfs/src/storage/concurrency/io_schedule.rs

@@ -1683,7 +1683,7 @@ impl<T> IoPriorityQueue<T> {
 
 /// Global metrics for I/O priority queue monitoring.
 ///
-/// These metrics are exposed for Prometheus scraping and provide
+/// These metrics are emitted through the shared metrics pipeline and provide
 /// visibility into the priority queue behavior.
 #[allow(dead_code)]
 pub struct IoPriorityMetrics {

rustfs/src/storage/deadlock_detector.rs

@@ -34,7 +34,7 @@
 //! - Request resource tracking (locks, memory, file handles)
 //! - Lock wait graph analysis for cycle detection
 //! - Configurable detection interval and hang threshold
-//! - Prometheus metrics for deadlock events
+//! - Deadlock metrics emitted through the shared metrics pipeline
 //! - Detailed diagnostic logging
 //!
 //! # Usage

rustfs/src/storage/lock_optimizer.rs

@@ -37,7 +37,7 @@
 //! - Early lock release after metadata read
 //! - Lock hold time monitoring
 //! - Configurable optimization (can be disabled for debugging)
-//! - Prometheus metrics for lock contention analysis
+//! - Lock contention metrics emitted through the shared metrics pipeline
 //!
 //! # Architecture
 //!

rustfs/src/storage/timeout_wrapper.rs

@@ -34,7 +34,7 @@
 //! - Configurable request-level timeout (default 30 seconds)
 //! - Automatic cancellation of sub-tasks on timeout
 //! - Resource cleanup on timeout (locks, memory, file handles)
-//! - Prometheus metrics for timeout monitoring
+//! - Timeout metrics emitted through `rustfs-io-metrics`
 
 // Allow dead_code for public API that may be used by external modules or future features
 #![allow(dead_code)]
@@ -42,7 +42,7 @@
 //! - Configurable request-level timeout (default 30 seconds)
 //! - Automatic cancellation of sub-tasks on timeout
 //! - Resource cleanup on timeout (locks, memory, file handles)
-//! - Prometheus metrics for timeout monitoring
+//! - Timeout metrics emitted through `rustfs-io-metrics`
 //!
 //! # Usage
 //!