metrics.rs 66.4 KB
Newer Older
1
// SPDX-FileCopyrightText: Copyright (c) 2024-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
2
3
// SPDX-License-Identifier: Apache-2.0

4
//! Metrics registry trait and implementation for Prometheus metrics
5
//!
6
7
//! This module provides a trait-based interface for creating and managing Prometheus metrics
//! with automatic label injection and hierarchical naming support.
8

9
pub mod frontend_perf;
10
pub mod prometheus_names;
11
pub mod request_plane;
12
pub mod tokio_perf;
13
14
pub mod transport_metrics;
pub mod work_handler_perf;
15

16
use parking_lot::Mutex;
17
18
19
20
21
use std::collections::HashSet;
use std::sync::Arc;

use crate::component::ComponentBuilder;
use anyhow;
22
23
use once_cell::sync::Lazy;
use regex::Regex;
24
25
26
use std::any::Any;
use std::collections::HashMap;

27
28
// Import commonly used items to avoid verbose prefixes
use prometheus_names::{
29
30
    build_component_metric_name, labels, name_prefix, sanitize_prometheus_label,
    sanitize_prometheus_name, work_handler,
31
32
33
34
};

// Pipeline imports for endpoint creation
use crate::pipeline::{
35
36
    AsyncEngine, AsyncEngineContextProvider, Error, ManyOut, ResponseStream, SingleIn, async_trait,
    network::Ingress,
37
38
39
40
41
};
use crate::protocols::annotated::Annotated;
use crate::stream;
use crate::stream::StreamExt;

42
43
44
// Prometheus imports
use prometheus::Encoder;

45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
/// Validate that a label slice has no duplicate keys.
/// Returns Ok(()) when all keys are unique; otherwise returns an error naming the duplicate key.
fn validate_no_duplicate_label_keys(labels: &[(&str, &str)]) -> anyhow::Result<()> {
    let mut seen_keys = std::collections::HashSet::new();
    for (key, _) in labels {
        if !seen_keys.insert(*key) {
            return Err(anyhow::anyhow!(
                "Duplicate label key '{}' found in labels",
                key
            ));
        }
    }
    Ok(())
}

60
61
62
/// ==============================
/// Prometheus section
/// ==============================
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
/// Trait that defines common behavior for Prometheus metric types
pub trait PrometheusMetric: prometheus::core::Collector + Clone + Send + Sync + 'static {
    /// Create a new metric with the given options
    fn with_opts(opts: prometheus::Opts) -> Result<Self, prometheus::Error>
    where
        Self: Sized;

    /// Create a new metric with histogram options and custom buckets
    /// This is a default implementation that will panic for non-histogram metrics
    fn with_histogram_opts_and_buckets(
        _opts: prometheus::HistogramOpts,
        _buckets: Option<Vec<f64>>,
    ) -> Result<Self, prometheus::Error>
    where
        Self: Sized,
    {
        panic!("with_histogram_opts_and_buckets is not implemented for this metric type");
    }

    /// Create a new metric with counter options and label names (for CounterVec)
    /// This is a default implementation that will panic for non-countervec metrics
    fn with_opts_and_label_names(
        _opts: prometheus::Opts,
        _label_names: &[&str],
    ) -> Result<Self, prometheus::Error>
    where
        Self: Sized,
    {
        panic!("with_opts_and_label_names is not implemented for this metric type");
    }
}

// Implement the trait for Counter, IntCounter, and Gauge
impl PrometheusMetric for prometheus::Counter {
    fn with_opts(opts: prometheus::Opts) -> Result<Self, prometheus::Error> {
        prometheus::Counter::with_opts(opts)
    }
}

impl PrometheusMetric for prometheus::IntCounter {
    fn with_opts(opts: prometheus::Opts) -> Result<Self, prometheus::Error> {
        prometheus::IntCounter::with_opts(opts)
    }
}

impl PrometheusMetric for prometheus::Gauge {
    fn with_opts(opts: prometheus::Opts) -> Result<Self, prometheus::Error> {
        prometheus::Gauge::with_opts(opts)
    }
}

impl PrometheusMetric for prometheus::IntGauge {
    fn with_opts(opts: prometheus::Opts) -> Result<Self, prometheus::Error> {
        prometheus::IntGauge::with_opts(opts)
    }
}

120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
impl PrometheusMetric for prometheus::GaugeVec {
    fn with_opts(_opts: prometheus::Opts) -> Result<Self, prometheus::Error> {
        Err(prometheus::Error::Msg(
            "GaugeVec requires label names, use with_opts_and_label_names instead".to_string(),
        ))
    }

    fn with_opts_and_label_names(
        opts: prometheus::Opts,
        label_names: &[&str],
    ) -> Result<Self, prometheus::Error> {
        prometheus::GaugeVec::new(opts, label_names)
    }
}

135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
impl PrometheusMetric for prometheus::IntGaugeVec {
    fn with_opts(_opts: prometheus::Opts) -> Result<Self, prometheus::Error> {
        Err(prometheus::Error::Msg(
            "IntGaugeVec requires label names, use with_opts_and_label_names instead".to_string(),
        ))
    }

    fn with_opts_and_label_names(
        opts: prometheus::Opts,
        label_names: &[&str],
    ) -> Result<Self, prometheus::Error> {
        prometheus::IntGaugeVec::new(opts, label_names)
    }
}

150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
impl PrometheusMetric for prometheus::IntCounterVec {
    fn with_opts(_opts: prometheus::Opts) -> Result<Self, prometheus::Error> {
        Err(prometheus::Error::Msg(
            "IntCounterVec requires label names, use with_opts_and_label_names instead".to_string(),
        ))
    }

    fn with_opts_and_label_names(
        opts: prometheus::Opts,
        label_names: &[&str],
    ) -> Result<Self, prometheus::Error> {
        prometheus::IntCounterVec::new(opts, label_names)
    }
}

165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
// Implement the trait for Histogram
impl PrometheusMetric for prometheus::Histogram {
    fn with_opts(opts: prometheus::Opts) -> Result<Self, prometheus::Error> {
        // Convert Opts to HistogramOpts
        let histogram_opts = prometheus::HistogramOpts::new(opts.name, opts.help);
        prometheus::Histogram::with_opts(histogram_opts)
    }

    fn with_histogram_opts_and_buckets(
        mut opts: prometheus::HistogramOpts,
        buckets: Option<Vec<f64>>,
    ) -> Result<Self, prometheus::Error> {
        if let Some(custom_buckets) = buckets {
            opts = opts.buckets(custom_buckets);
        }
        prometheus::Histogram::with_opts(opts)
    }
}

// Implement the trait for CounterVec
impl PrometheusMetric for prometheus::CounterVec {
    fn with_opts(_opts: prometheus::Opts) -> Result<Self, prometheus::Error> {
        // This will panic - CounterVec needs label names
        panic!("CounterVec requires label names, use with_opts_and_label_names instead");
    }

    fn with_opts_and_label_names(
        opts: prometheus::Opts,
        label_names: &[&str],
    ) -> Result<Self, prometheus::Error> {
        prometheus::CounterVec::new(opts, label_names)
    }
}

199
200
201
202
203
204
/// ==============================
/// Metrics section
/// ==============================
/// Public helper function to create metrics - accessible for Python bindings
pub fn create_metric<T: PrometheusMetric, H: MetricsHierarchy + ?Sized>(
    hierarchy: &H,
205
206
207
208
209
    metric_name: &str,
    metric_desc: &str,
    labels: &[(&str, &str)],
    buckets: Option<Vec<f64>>,
    const_labels: Option<&[&str]>,
210
) -> anyhow::Result<T> {
211
    // Validate that user-provided labels don't have duplicate keys
212
    validate_no_duplicate_label_keys(labels)?;
213
    // Note: stored labels functionality has been removed
214

215
216
    let basename = hierarchy.basename();
    let parent_hierarchies = hierarchy.parent_hierarchies();
217

218
219
220
221
    // Build hierarchy path as vector of strings: parent names + [basename]
    let mut hierarchy_names: Vec<String> =
        parent_hierarchies.iter().map(|p| p.basename()).collect();
    hierarchy_names.push(basename.clone());
222

223
    let metric_name = build_component_metric_name(metric_name);
224

225
    // Build updated_labels: auto-labels first, then `labels` + stored labels
226
227
    let mut updated_labels: Vec<(String, String)> = Vec::new();

228
229
230
231
232
233
234
235
236
237
238
239
    // Auto-label injection: Always add dynamo_namespace, dynamo_component, dynamo_endpoint labels
    // based on the hierarchy. Label constants defined in prometheus_names.rs labels module.
    //
    // Python counterpart: components/src/dynamo/common/utils/prometheus.py register_engine_metrics_callback()

    // Validate that user-provided labels don't conflict with auto-generated labels
    for (key, _) in labels {
        if *key == labels::NAMESPACE || *key == labels::COMPONENT || *key == labels::ENDPOINT {
            return Err(anyhow::anyhow!(
                "Label '{}' is automatically added by auto-label injection and cannot be manually set",
                key
            ));
240
        }
241
    }
242

243
244
245
246
247
248
249
250
    // Add auto-generated labels with sanitized values
    // Hierarchy: [drt, namespace, component, endpoint]
    if hierarchy_names.len() > 1 {
        let namespace = &hierarchy_names[1];
        if !namespace.is_empty() {
            let valid_namespace = sanitize_prometheus_label(namespace)?;
            if !valid_namespace.is_empty() {
                updated_labels.push((labels::NAMESPACE.to_string(), valid_namespace));
251
            }
252
        }
253
254
255
256
257
258
259
    }
    if hierarchy_names.len() > 2 {
        let component = &hierarchy_names[2];
        if !component.is_empty() {
            let valid_component = sanitize_prometheus_label(component)?;
            if !valid_component.is_empty() {
                updated_labels.push((labels::COMPONENT.to_string(), valid_component));
260
261
            }
        }
262
263
264
265
266
267
268
    }
    if hierarchy_names.len() > 3 {
        let endpoint = &hierarchy_names[3];
        if !endpoint.is_empty() {
            let valid_endpoint = sanitize_prometheus_label(endpoint)?;
            if !valid_endpoint.is_empty() {
                updated_labels.push((labels::ENDPOINT.to_string(), valid_endpoint));
269
270
271
272
273
274
275
276
277
278
            }
        }
    }

    // Add user labels
    updated_labels.extend(
        labels
            .iter()
            .map(|(k, v)| ((*k).to_string(), (*v).to_string())),
    );
279
    // Note: stored labels functionality has been removed
280
281

    // Handle different metric types
282
    let prometheus_metric = if std::any::TypeId::of::<T>()
283
        == std::any::TypeId::of::<prometheus::CounterVec>()
284
    {
285
286
287
288
289
290
291
292
293
294
295
296
297
298
        // Special handling for CounterVec with label names
        // const_labels parameter is required for CounterVec
        if buckets.is_some() {
            return Err(anyhow::anyhow!(
                "buckets parameter is not valid for CounterVec"
            ));
        }
        let mut opts = prometheus::Opts::new(&metric_name, metric_desc);
        for (key, value) in &updated_labels {
            opts = opts.const_label(key.clone(), value.clone());
        }
        let label_names = const_labels
            .ok_or_else(|| anyhow::anyhow!("CounterVec requires const_labels parameter"))?;
        T::with_opts_and_label_names(opts, label_names)?
299
300
301
    } else if std::any::TypeId::of::<T>() == std::any::TypeId::of::<prometheus::GaugeVec>() {
        // Special handling for GaugeVec with label names
        // const_labels parameter is required for GaugeVec
302
303
        if buckets.is_some() {
            return Err(anyhow::anyhow!(
304
                "buckets parameter is not valid for GaugeVec"
305
306
307
308
309
310
311
            ));
        }
        let mut opts = prometheus::Opts::new(&metric_name, metric_desc);
        for (key, value) in &updated_labels {
            opts = opts.const_label(key.clone(), value.clone());
        }
        let label_names = const_labels
312
            .ok_or_else(|| anyhow::anyhow!("GaugeVec requires const_labels parameter"))?;
313
        T::with_opts_and_label_names(opts, label_names)?
314
315
316
317
318
319
320
321
322
323
324
325
326
    } else if std::any::TypeId::of::<T>() == std::any::TypeId::of::<prometheus::Histogram>() {
        // Special handling for Histogram with custom buckets
        // buckets parameter is valid for Histogram, const_labels is not used
        if const_labels.is_some() {
            return Err(anyhow::anyhow!(
                "const_labels parameter is not valid for Histogram"
            ));
        }
        let mut opts = prometheus::HistogramOpts::new(&metric_name, metric_desc);
        for (key, value) in &updated_labels {
            opts = opts.const_label(key.clone(), value.clone());
        }
        T::with_histogram_opts_and_buckets(opts, buckets)?
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
    } else if std::any::TypeId::of::<T>() == std::any::TypeId::of::<prometheus::IntCounterVec>() {
        // Special handling for IntCounterVec with label names
        // const_labels parameter is required for IntCounterVec
        if buckets.is_some() {
            return Err(anyhow::anyhow!(
                "buckets parameter is not valid for IntCounterVec"
            ));
        }
        let mut opts = prometheus::Opts::new(&metric_name, metric_desc);
        for (key, value) in &updated_labels {
            opts = opts.const_label(key.clone(), value.clone());
        }
        let label_names = const_labels
            .ok_or_else(|| anyhow::anyhow!("IntCounterVec requires const_labels parameter"))?;
        T::with_opts_and_label_names(opts, label_names)?
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
    } else if std::any::TypeId::of::<T>() == std::any::TypeId::of::<prometheus::IntGaugeVec>() {
        // Special handling for IntGaugeVec with label names
        // const_labels parameter is required for IntGaugeVec
        if buckets.is_some() {
            return Err(anyhow::anyhow!(
                "buckets parameter is not valid for IntGaugeVec"
            ));
        }
        let mut opts = prometheus::Opts::new(&metric_name, metric_desc);
        for (key, value) in &updated_labels {
            opts = opts.const_label(key.clone(), value.clone());
        }
        let label_names = const_labels
            .ok_or_else(|| anyhow::anyhow!("IntGaugeVec requires const_labels parameter"))?;
        T::with_opts_and_label_names(opts, label_names)?
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
    } else {
        // Standard handling for Counter, IntCounter, Gauge, IntGauge
        // buckets and const_labels parameters are not valid for these types
        if buckets.is_some() {
            return Err(anyhow::anyhow!(
                "buckets parameter is not valid for Counter, IntCounter, Gauge, or IntGauge"
            ));
        }
        if const_labels.is_some() {
            return Err(anyhow::anyhow!(
                "const_labels parameter is not valid for Counter, IntCounter, Gauge, or IntGauge"
            ));
        }
        let mut opts = prometheus::Opts::new(&metric_name, metric_desc);
        for (key, value) in &updated_labels {
            opts = opts.const_label(key.clone(), value.clone());
        }
        T::with_opts(opts)?
    };

377
378
379
    let collector: Box<dyn prometheus::core::Collector> = Box::new(prometheus_metric.clone());
    hierarchy.get_metrics_registry().add_metric(collector)?;

380
    Ok(prometheus_metric)
381
382
}

383
384
385
386
387
/// Wrapper struct that provides access to metrics functionality
/// This struct is accessed via the `.metrics()` method on DistributedRuntime, Namespace, Component, and Endpoint
pub struct Metrics<H: MetricsHierarchy> {
    hierarchy: H,
}
388

389
390
391
impl<H: MetricsHierarchy> Metrics<H> {
    pub fn new(hierarchy: H) -> Self {
        Self { hierarchy }
392
393
394
395
396
397
    }

    // TODO: Add support for additional Prometheus metric types:
    // - Counter: ✅ IMPLEMENTED - create_counter()
    // - CounterVec: ✅ IMPLEMENTED - create_countervec()
    // - Gauge: ✅ IMPLEMENTED - create_gauge()
398
    // - GaugeVec: ✅ IMPLEMENTED - create_gaugevec()
399
    // - GaugeHistogram: create_gauge_histogram() - for gauge histograms
400
401
402
    // - Histogram: ✅ IMPLEMENTED - create_histogram()
    // - HistogramVec with custom buckets: create_histogram_with_buckets()
    // - Info: create_info() - for info metrics with labels
403
404
405
406
    // - IntCounter: ✅ IMPLEMENTED - create_intcounter()
    // - IntCounterVec: ✅ IMPLEMENTED - create_intcountervec()
    // - IntGauge: ✅ IMPLEMENTED - create_intgauge()
    // - IntGaugeVec: ✅ IMPLEMENTED - create_intgaugevec()
407
    // - Stateset: create_stateset() - for state-based metrics
408
409
410
    // - Summary: create_summary() - for quantiles and sum/count metrics
    // - SummaryVec: create_summary_vec() - for labeled summaries
    // - Untyped: create_untyped() - for untyped metrics
411
412
413
    //
    // NOTE: The order of create_* methods below is mirrored in lib/bindings/python/rust/lib.rs::Metrics
    // Keep them synchronized when adding new metric types
414
415

    /// Create a Counter metric
416
    pub fn create_counter(
417
418
419
420
        &self,
        name: &str,
        description: &str,
        labels: &[(&str, &str)],
421
    ) -> anyhow::Result<prometheus::Counter> {
422
        create_metric(&self.hierarchy, name, description, labels, None, None)
423
424
    }

425
    /// Create a CounterVec metric with label names (for dynamic labels)
426
    pub fn create_countervec(
427
428
429
        &self,
        name: &str,
        description: &str,
430
431
        const_labels: &[&str],
        const_label_values: &[(&str, &str)],
432
    ) -> anyhow::Result<prometheus::CounterVec> {
433
        create_metric(
434
            &self.hierarchy,
435
436
437
438
439
440
            name,
            description,
            const_label_values,
            None,
            Some(const_labels),
        )
441
442
    }

443
    /// Create a Gauge metric
444
    pub fn create_gauge(
445
446
447
448
        &self,
        name: &str,
        description: &str,
        labels: &[(&str, &str)],
449
    ) -> anyhow::Result<prometheus::Gauge> {
450
        create_metric(&self.hierarchy, name, description, labels, None, None)
451
452
    }

453
    /// Create a GaugeVec metric with label names (for dynamic labels)
454
    pub fn create_gaugevec(
455
456
457
458
459
460
461
        &self,
        name: &str,
        description: &str,
        const_labels: &[&str],
        const_label_values: &[(&str, &str)],
    ) -> anyhow::Result<prometheus::GaugeVec> {
        create_metric(
462
            &self.hierarchy,
463
464
465
466
467
468
469
470
            name,
            description,
            const_label_values,
            None,
            Some(const_labels),
        )
    }

471
    /// Create a Histogram metric with custom buckets
472
    pub fn create_histogram(
473
474
475
476
477
        &self,
        name: &str,
        description: &str,
        labels: &[(&str, &str)],
        buckets: Option<Vec<f64>>,
478
    ) -> anyhow::Result<prometheus::Histogram> {
479
        create_metric(&self.hierarchy, name, description, labels, buckets, None)
480
481
    }

482
    /// Create an IntCounter metric
483
    pub fn create_intcounter(
484
485
486
487
        &self,
        name: &str,
        description: &str,
        labels: &[(&str, &str)],
488
    ) -> anyhow::Result<prometheus::IntCounter> {
489
        create_metric(&self.hierarchy, name, description, labels, None, None)
490
491
492
    }

    /// Create an IntCounterVec metric with label names (for dynamic labels)
493
    pub fn create_intcountervec(
494
495
496
497
498
        &self,
        name: &str,
        description: &str,
        const_labels: &[&str],
        const_label_values: &[(&str, &str)],
499
    ) -> anyhow::Result<prometheus::IntCounterVec> {
500
        create_metric(
501
            &self.hierarchy,
502
503
504
505
506
507
508
509
510
            name,
            description,
            const_label_values,
            None,
            Some(const_labels),
        )
    }

    /// Create an IntGauge metric
511
    pub fn create_intgauge(
512
513
514
515
        &self,
        name: &str,
        description: &str,
        labels: &[(&str, &str)],
516
    ) -> anyhow::Result<prometheus::IntGauge> {
517
        create_metric(&self.hierarchy, name, description, labels, None, None)
518
519
520
    }

    /// Create an IntGaugeVec metric with label names (for dynamic labels)
521
    pub fn create_intgaugevec(
522
523
524
525
526
        &self,
        name: &str,
        description: &str,
        const_labels: &[&str],
        const_label_values: &[(&str, &str)],
527
    ) -> anyhow::Result<prometheus::IntGaugeVec> {
528
        create_metric(
529
            &self.hierarchy,
530
531
532
533
534
535
536
537
538
            name,
            description,
            const_label_values,
            None,
            Some(const_labels),
        )
    }

    /// Get metrics in Prometheus text format
539
    pub fn prometheus_expfmt(&self) -> anyhow::Result<String> {
540
        self.hierarchy
541
            .get_metrics_registry()
542
            .prometheus_expfmt_combined()
543
544
545
    }
}

546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
/// This trait should be implemented by all metric registries, including Prometheus, Envy, OpenTelemetry, and others.
/// It offers a unified interface for creating and managing metrics, organizing sub-registries, and
/// generating output in Prometheus text format.
use crate::traits::DistributedRuntimeProvider;

pub trait MetricsHierarchy: Send + Sync {
    // ========================================================================
    // Required methods - must be implemented by all types
    // ========================================================================

    /// Get the name of this hierarchy (without any hierarchy prefix)
    fn basename(&self) -> String;

    /// Get the parent hierarchies as actual objects (not strings)
    /// Returns a vector of hierarchy references, ordered from root to immediate parent.
    /// For example, an Endpoint would return [DRT, Namespace, Component].
    fn parent_hierarchies(&self) -> Vec<&dyn MetricsHierarchy>;

    /// Get a reference to this hierarchy's metrics registry
    fn get_metrics_registry(&self) -> &MetricsRegistry;

    // ========================================================================
    // Provided methods - have default implementations
    // ========================================================================

    /// Access the metrics interface for this hierarchy
    /// This is a provided method that works for any type implementing MetricsHierarchy
    fn metrics(&self) -> Metrics<&Self>
    where
        Self: Sized,
    {
        Metrics::new(self)
    }
}

// Blanket implementation for references to types that implement MetricsHierarchy
impl<T: MetricsHierarchy + ?Sized> MetricsHierarchy for &T {
    fn basename(&self) -> String {
        (**self).basename()
    }

    fn parent_hierarchies(&self) -> Vec<&dyn MetricsHierarchy> {
        (**self).parent_hierarchies()
    }

    fn get_metrics_registry(&self) -> &MetricsRegistry {
        (**self).get_metrics_registry()
    }
}

/// Type alias for runtime callback functions to reduce complexity
///
/// This type represents an Arc-wrapped callback function that can be:
/// - Shared efficiently across multiple threads and contexts
/// - Cloned without duplicating the underlying closure
/// - Used in generic contexts requiring 'static lifetime
///
/// The Arc wrapper is included in the type to make sharing explicit.
pub type PrometheusUpdateCallback = Arc<dyn Fn() -> anyhow::Result<()> + Send + Sync + 'static>;

/// Type alias for exposition text callback functions that return Prometheus text
pub type PrometheusExpositionFormatCallback =
    Arc<dyn Fn() -> anyhow::Result<String> + Send + Sync + 'static>;

610
611
612
613
614
/// Structure to hold Prometheus registries and associated callbacks for a given hierarchy.
///
/// All fields are Arc-wrapped, so cloning shares state. This ensures metrics registered
/// on cloned instances (e.g., cloned Client/Endpoint) are visible to the original.
#[derive(Clone)]
615
pub struct MetricsRegistry {
616
617
618
    /// The Prometheus registry for this hierarchy.
    /// Arc-wrapped so clones share the same registry (metrics registered on clones are visible everywhere).
    pub prometheus_registry: Arc<std::sync::RwLock<prometheus::Registry>>,
619

620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
    /// Child registries included when emitting combined `/metrics` output.
    ///
    /// Why this exists:
    /// - Previously, `create_metric()` registered every collector into *all* parent registries
    ///   (Endpoint → Component → Namespace → DRT) so scraping the root registry included everything.
    /// - That fan-out caused Prometheus collisions when different endpoints tried to register the
    ///   same metric name with different const-labels (descriptor mismatch).
    ///
    /// We now register metrics only into the local hierarchy registry to avoid collisions.
    /// `child_registries` rebuilds “what to scrape” as a tree of registries so `/metrics` can:
    /// - traverse registries recursively,
    /// - merge metric families into one exposition payload,
    /// - warn/drop exact duplicate series, while allowing same metric name with different labels.
    child_registries: Arc<std::sync::RwLock<Vec<MetricsRegistry>>>,

635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
    /// Update callbacks invoked before metrics are scraped.
    /// Wrapped in Arc to preserve callbacks across clones (prevents callback loss when MetricsRegistry is cloned).
    pub prometheus_update_callbacks: Arc<std::sync::RwLock<Vec<PrometheusUpdateCallback>>>,

    /// Callbacks that return Prometheus exposition text appended to metrics output.
    /// Wrapped in Arc to preserve callbacks across clones (e.g., vLLM callbacks registered at Endpoint remain accessible at DRT).
    pub prometheus_expfmt_callbacks:
        Arc<std::sync::RwLock<Vec<PrometheusExpositionFormatCallback>>>,
}

impl std::fmt::Debug for MetricsRegistry {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        f.debug_struct("MetricsRegistry")
            .field("prometheus_registry", &"<RwLock<Registry>>")
            .field(
                "prometheus_update_callbacks",
                &format!(
                    "<RwLock<Vec<Callback>>> with {} callbacks",
                    self.prometheus_update_callbacks.read().unwrap().len()
                ),
            )
            .field(
                "prometheus_expfmt_callbacks",
                &format!(
                    "<RwLock<Vec<Callback>>> with {} callbacks",
                    self.prometheus_expfmt_callbacks.read().unwrap().len()
                ),
            )
            .finish()
    }
}

impl MetricsRegistry {
    /// Create a new metrics registry with an empty Prometheus registry and callback lists
    pub fn new() -> Self {
        Self {
671
            prometheus_registry: Arc::new(std::sync::RwLock::new(prometheus::Registry::new())),
672
            child_registries: Arc::new(std::sync::RwLock::new(Vec::new())),
673
674
675
676
677
            prometheus_update_callbacks: Arc::new(std::sync::RwLock::new(Vec::new())),
            prometheus_expfmt_callbacks: Arc::new(std::sync::RwLock::new(Vec::new())),
        }
    }

678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
    /// Add a child registry to be included in combined /metrics output.
    ///
    /// Dedup is by underlying Prometheus registry pointer, so repeated registration via clones is safe.
    pub fn add_child_registry(&self, child: &MetricsRegistry) {
        let child_ptr = Arc::as_ptr(&child.prometheus_registry);
        let mut guard = self.child_registries.write().unwrap();
        if guard
            .iter()
            .any(|r| Arc::as_ptr(&r.prometheus_registry) == child_ptr)
        {
            return;
        }
        guard.push(child.clone());
    }

    fn registries_for_combined_scrape(&self) -> Vec<MetricsRegistry> {
        // Traverse child registries recursively so `prometheus_expfmt()` on any hierarchy
        // (DRT/namespace/component/endpoint) includes metrics from its descendants.
        //
        // Dedup by underlying Prometheus registry pointer so multiple paths (e.g. also registering
        // directly on the root) won't duplicate output.
        fn visit(
            registry: &MetricsRegistry,
            out: &mut Vec<MetricsRegistry>,
            seen: &mut HashSet<*const std::sync::RwLock<prometheus::Registry>>,
        ) {
            let ptr = Arc::as_ptr(&registry.prometheus_registry);
            if !seen.insert(ptr) {
                return;
            }

            out.push(registry.clone());

            let children: Vec<MetricsRegistry> = registry
                .child_registries
                .read()
                .unwrap()
                .iter()
                .cloned()
                .collect();
            for child in children {
                visit(&child, out, seen);
            }
        }

        let mut out = Vec::new();
        let mut seen: HashSet<*const std::sync::RwLock<prometheus::Registry>> = HashSet::new();
        visit(self, &mut out, &mut seen);
        out
    }

    /// Combine metrics across this registry and all registered children into one Prometheus exposition output.
    ///
    /// - Families are merged by name; HELP and TYPE must match.
    /// - Multiple series for the same name are allowed if labels differ.
    /// - Exact duplicate series (same name + identical label pairs) are warned and dropped.
    pub fn prometheus_expfmt_combined(&self) -> anyhow::Result<String> {
        let registries = self.registries_for_combined_scrape();

        // Run per-registry update callbacks first.
        for registry in &registries {
            for result in registry.execute_update_callbacks() {
                if let Err(e) = result {
741
                    tracing::error!("Error executing metrics callback: {e}");
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
                }
            }
        }

        // Merge metric families.
        let mut by_name: HashMap<String, prometheus::proto::MetricFamily> = HashMap::new();
        let mut seen_series: HashSet<String> = HashSet::new();

        for (registry_idx, registry) in registries.iter().enumerate() {
            let families = registry.get_prometheus_registry().gather();
            for mut family in families {
                let name = family.name().to_string();

                let entry = by_name.entry(name.clone()).or_insert_with(|| {
                    let mut out = prometheus::proto::MetricFamily::new();
                    out.set_name(name.clone());
                    out.set_help(family.help().to_string());
                    out.set_field_type(family.get_field_type());
                    out
                });

                if entry.help() != family.help()
                    || entry.get_field_type() != family.get_field_type()
                {
                    return Err(anyhow::anyhow!(
                        "Metric family '{}' has inconsistent help/type across registries (idx={})",
                        name,
                        registry_idx
                    ));
                }

                let mut metrics = family.take_metric();
                for metric in metrics.drain(..) {
                    let mut labels: Vec<(String, String)> = metric
                        .get_label()
                        .iter()
                        .map(|lp| (lp.name().to_string(), lp.value().to_string()))
                        .collect();
                    labels.sort_by(|(ka, va), (kb, vb)| (ka, va).cmp(&(kb, vb)));

                    let key = format!(
                        "{}|{}",
                        name,
                        labels
                            .iter()
                            .map(|(k, v)| format!("{}={}", k, v))
                            .collect::<Vec<_>>()
                            .join(",")
                    );

                    if !seen_series.insert(key) {
                        tracing::warn!(
                            metric_name = %name,
                            labels = ?labels,
                            registry_idx,
                            "Duplicate Prometheus series while merging registries; dropping later sample"
                        );
                        continue;
                    }

                    entry.mut_metric().push(metric);
                }
            }
        }

        let mut merged: Vec<prometheus::proto::MetricFamily> = by_name.into_values().collect();
        merged.sort_by(|a, b| a.name().cmp(b.name()));

        let encoder = prometheus::TextEncoder::new();
        let mut buffer = Vec::new();
        encoder.encode(&merged, &mut buffer)?;
        let mut result = String::from_utf8(buffer)?;

        // Append expfmt callbacks deterministically in registry order.
        let mut expfmt = String::new();
        for registry in registries {
            let text = registry.execute_expfmt_callbacks();
            if !text.is_empty() {
                if !expfmt.is_empty() && !expfmt.ends_with('\n') {
                    expfmt.push('\n');
                }
                expfmt.push_str(&text);
            }
        }

        if !expfmt.is_empty() {
            if !result.ends_with('\n') {
                result.push('\n');
            }
            result.push_str(&expfmt);
        }

        Ok(result)
    }

837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
    /// Add a callback function that receives a reference to any MetricsHierarchy
    pub fn add_update_callback(&self, callback: PrometheusUpdateCallback) {
        self.prometheus_update_callbacks
            .write()
            .unwrap()
            .push(callback);
    }

    /// Add an exposition text callback that returns Prometheus text
    pub fn add_expfmt_callback(&self, callback: PrometheusExpositionFormatCallback) {
        self.prometheus_expfmt_callbacks
            .write()
            .unwrap()
            .push(callback);
    }

    /// Execute all update callbacks and return their results
    pub fn execute_update_callbacks(&self) -> Vec<anyhow::Result<()>> {
        self.prometheus_update_callbacks
            .read()
            .unwrap()
            .iter()
            .map(|callback| callback())
            .collect()
    }

    /// Execute all exposition text callbacks and return their concatenated text
    pub fn execute_expfmt_callbacks(&self) -> String {
        let callbacks = self.prometheus_expfmt_callbacks.read().unwrap();
        let mut result = String::new();
        for callback in callbacks.iter() {
            match callback() {
                Ok(text) => {
                    if !text.is_empty() {
                        if !result.is_empty() && !result.ends_with('\n') {
                            result.push('\n');
                        }
                        result.push_str(&text);
                    }
                }
                Err(e) => {
878
                    tracing::error!("Error executing exposition text callback: {e}");
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
                }
            }
        }
        result
    }

    /// Add a Prometheus metric collector to this registry
    pub fn add_metric(
        &self,
        collector: Box<dyn prometheus::core::Collector>,
    ) -> anyhow::Result<()> {
        self.prometheus_registry
            .write()
            .unwrap()
            .register(collector)
            .map_err(|e| anyhow::anyhow!("Failed to register metric: {}", e))
    }

897
898
899
900
901
902
903
    /// Add a Prometheus metric collector, logging a warning on failure instead of returning an error.
    pub fn add_metric_or_warn(&self, collector: Box<dyn prometheus::core::Collector>, name: &str) {
        if let Err(e) = self.add_metric(collector) {
            tracing::warn!(error = %e, metric = name, "Failed to register metric");
        }
    }

904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
    /// Get a read guard to the Prometheus registry for scraping
    pub fn get_prometheus_registry(&self) -> std::sync::RwLockReadGuard<'_, prometheus::Registry> {
        self.prometheus_registry.read().unwrap()
    }

    /// Returns true if a metric with the given name already exists in the Prometheus registry
    pub fn has_metric_named(&self, metric_name: &str) -> bool {
        self.prometheus_registry
            .read()
            .unwrap()
            .gather()
            .iter()
            .any(|mf| mf.name() == metric_name)
    }
}

impl Default for MetricsRegistry {
    fn default() -> Self {
        Self::new()
    }
}

926
#[cfg(test)]
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
mod test_helpers {
    use super::prometheus_names::name_prefix;
    use super::*;

    /// Base function to filter Prometheus output lines based on a predicate.
    /// Returns lines that match the predicate, converted to String.
    fn filter_prometheus_lines<F>(input: &str, mut predicate: F) -> Vec<String>
    where
        F: FnMut(&str) -> bool,
    {
        input
            .lines()
            .filter(|line| predicate(line))
            .map(|line| line.to_string())
            .collect::<Vec<_>>()
    }

    /// Extracts all component metrics (excluding help text and type definitions).
    /// Returns only the actual metric lines with values.
    pub fn extract_metrics(input: &str) -> Vec<String> {
        filter_prometheus_lines(input, |line| {
948
            line.starts_with(&format!("{}_", name_prefix::COMPONENT))
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
                && !line.starts_with("#")
                && !line.trim().is_empty()
        })
    }

    /// Parses a Prometheus metric line and extracts the name, labels, and value.
    /// Used instead of fetching metrics directly to test end-to-end results, not intermediate state.
    ///
    /// # Example
    /// ```
    /// let line = "http_requests_total{method=\"GET\"} 1234";
    /// let (name, labels, value) = parse_prometheus_metric(line).unwrap();
    /// assert_eq!(name, "http_requests_total");
    /// assert_eq!(labels.get("method"), Some(&"GET".to_string()));
    /// assert_eq!(value, 1234.0);
    /// ```
    pub fn parse_prometheus_metric(
        line: &str,
    ) -> Option<(String, std::collections::HashMap<String, String>, f64)> {
        if line.trim().is_empty() || line.starts_with('#') {
            return None;
        }

        let parts: Vec<&str> = line.split_whitespace().collect();
        if parts.len() < 2 {
            return None;
        }

        let metric_part = parts[0];
        let value: f64 = parts[1].parse().ok()?;

        let (name, labels) = if metric_part.contains('{') {
            let brace_start = metric_part.find('{').unwrap();
            let brace_end = metric_part.rfind('}').unwrap_or(metric_part.len());
            let name = &metric_part[..brace_start];
            let labels_str = &metric_part[brace_start + 1..brace_end];

            let mut labels = std::collections::HashMap::new();
            for pair in labels_str.split(',') {
                if let Some((k, v)) = pair.split_once('=') {
                    let v = v.trim_matches('"');
                    labels.insert(k.trim().to_string(), v.to_string());
                }
            }
            (name.to_string(), labels)
        } else {
            (metric_part.to_string(), std::collections::HashMap::new())
        };

        Some((name, labels, value))
    }
1000
1001
}

1002
#[cfg(test)]
1003
mod test_metricsregistry_units {
1004
1005
1006
    use super::*;

    #[test]
1007
1008
1009
    fn test_build_component_metric_name_with_prefix() {
        // Test that build_component_metric_name correctly prepends the dynamo_component prefix
        let result = build_component_metric_name("requests");
1010
        assert_eq!(result, "dynamo_component_requests");
1011

1012
        let result = build_component_metric_name("counter");
1013
        assert_eq!(result, "dynamo_component_counter");
1014
1015
    }

1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
    #[test]
    fn test_parse_prometheus_metric() {
        use super::test_helpers::parse_prometheus_metric;
        use std::collections::HashMap;

        // Test parsing a metric with labels
        let line = "http_requests_total{method=\"GET\",status=\"200\"} 1234";
        let parsed = parse_prometheus_metric(line);
        assert!(parsed.is_some());

        let (name, labels, value) = parsed.unwrap();
        assert_eq!(name, "http_requests_total");

        let mut expected_labels = HashMap::new();
        expected_labels.insert("method".to_string(), "GET".to_string());
        expected_labels.insert("status".to_string(), "200".to_string());
        assert_eq!(labels, expected_labels);

        assert_eq!(value, 1234.0);

        // Test parsing a metric without labels
        let line = "cpu_usage 98.5";
        let parsed = parse_prometheus_metric(line);
        assert!(parsed.is_some());

        let (name, labels, value) = parsed.unwrap();
        assert_eq!(name, "cpu_usage");
        assert!(labels.is_empty());
        assert_eq!(value, 98.5);
1045

1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
        // Test parsing a metric with float value
        let line = "response_time{service=\"api\"} 0.123";
        let parsed = parse_prometheus_metric(line);
        assert!(parsed.is_some());

        let (name, labels, value) = parsed.unwrap();
        assert_eq!(name, "response_time");

        let mut expected_labels = HashMap::new();
        expected_labels.insert("service".to_string(), "api".to_string());
        assert_eq!(labels, expected_labels);

        assert_eq!(value, 0.123);

        // Test parsing invalid lines
        assert!(parse_prometheus_metric("").is_none()); // Empty line
        assert!(parse_prometheus_metric("# HELP metric description").is_none()); // Help text
        assert!(parse_prometheus_metric("# TYPE metric counter").is_none()); // Type definition
        assert!(parse_prometheus_metric("metric_name").is_none()); // No value

        println!("✓ Prometheus metric parsing works correctly!");
    }

1069
    #[test]
1070
    fn test_metrics_registry_entry_callbacks() {
1071
        use crate::MetricsRegistry;
1072
        use std::sync::atomic::{AtomicUsize, Ordering};
1073

1074
1075
        // Test 1: Basic callback execution with counter increments
        {
1076
            let registry = MetricsRegistry::new();
1077
1078
1079
1080
1081
            let counter = Arc::new(AtomicUsize::new(0));

            // Add callbacks with different increment values
            for increment in [1, 10, 100] {
                let counter_clone = counter.clone();
1082
                registry.add_update_callback(Arc::new(move || {
1083
1084
1085
1086
                    counter_clone.fetch_add(increment, Ordering::SeqCst);
                    Ok(())
                }));
            }
1087

1088
1089
            // Verify counter starts at 0
            assert_eq!(counter.load(Ordering::SeqCst), 0);
1090

1091
            // First execution
1092
            let results = registry.execute_update_callbacks();
1093
1094
1095
            assert_eq!(results.len(), 3);
            assert!(results.iter().all(|r| r.is_ok()));
            assert_eq!(counter.load(Ordering::SeqCst), 111); // 1 + 10 + 100
1096

1097
            // Second execution - callbacks should be reusable
1098
            let results = registry.execute_update_callbacks();
1099
1100
            assert_eq!(results.len(), 3);
            assert_eq!(counter.load(Ordering::SeqCst), 222); // 111 + 111
1101

1102
1103
1104
            // Test cloning - cloned entry shares callbacks (callbacks are Arc-wrapped)
            let cloned = registry.clone();
            assert_eq!(cloned.execute_update_callbacks().len(), 3);
1105
            assert_eq!(counter.load(Ordering::SeqCst), 333); // 222 + 111
1106
1107
1108
1109

            // Original still has callbacks and shares the same Arc
            registry.execute_update_callbacks();
            assert_eq!(counter.load(Ordering::SeqCst), 444); // 333 + 111
1110
        }
1111

1112
1113
        // Test 2: Mixed success and error callbacks
        {
1114
            let registry = MetricsRegistry::new();
1115
1116
1117
1118
            let counter = Arc::new(AtomicUsize::new(0));

            // Successful callback
            let counter_clone = counter.clone();
1119
            registry.add_update_callback(Arc::new(move || {
1120
1121
1122
1123
1124
                counter_clone.fetch_add(1, Ordering::SeqCst);
                Ok(())
            }));

            // Error callback
1125
            registry.add_update_callback(Arc::new(|| Err(anyhow::anyhow!("Simulated error"))));
1126
1127
1128

            // Another successful callback
            let counter_clone = counter.clone();
1129
            registry.add_update_callback(Arc::new(move || {
1130
1131
1132
1133
1134
                counter_clone.fetch_add(10, Ordering::SeqCst);
                Ok(())
            }));

            // Execute and verify mixed results
1135
            let results = registry.execute_update_callbacks();
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
            assert_eq!(results.len(), 3);
            assert!(results[0].is_ok());
            assert!(results[1].is_err());
            assert!(results[2].is_ok());

            // Verify error message
            assert_eq!(
                results[1].as_ref().unwrap_err().to_string(),
                "Simulated error"
            );
1146

1147
1148
            // Verify successful callbacks still executed
            assert_eq!(counter.load(Ordering::SeqCst), 11); // 1 + 10
1149

1150
            // Execute again - errors should be consistent
1151
            let results = registry.execute_update_callbacks();
1152
1153
1154
            assert!(results[1].is_err());
            assert_eq!(counter.load(Ordering::SeqCst), 22); // 11 + 11
        }
1155

1156
1157
        // Test 3: Empty registry
        {
1158
1159
            let registry = MetricsRegistry::new();
            let results = registry.execute_update_callbacks();
1160
1161
1162
1163
            assert_eq!(results.len(), 0);
        }
    }
}
1164

1165
1166
1167
1168
#[cfg(feature = "integration")]
#[cfg(test)]
mod test_metricsregistry_prefixes {
    use super::*;
1169
    use crate::distributed::distributed_test_utils::create_test_drt_async;
1170
1171
    use prometheus::core::Collector;

1172
1173
    #[tokio::test]
    async fn test_hierarchical_prefixes_and_parent_hierarchies() {
1174
        let drt = create_test_drt_async().await;
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185

        const DRT_NAME: &str = "";
        const NAMESPACE_NAME: &str = "ns901";
        const COMPONENT_NAME: &str = "comp901";
        const ENDPOINT_NAME: &str = "ep901";
        let namespace = drt.namespace(NAMESPACE_NAME).unwrap();
        let component = namespace.component(COMPONENT_NAME).unwrap();
        let endpoint = component.endpoint(ENDPOINT_NAME);

        // DRT
        assert_eq!(drt.basename(), DRT_NAME);
1186
1187
        assert_eq!(drt.parent_hierarchies().len(), 0);
        // DRT hierarchy is just its basename (empty string)
1188
1189
1190

        // Namespace
        assert_eq!(namespace.basename(), NAMESPACE_NAME);
1191
1192
1193
        assert_eq!(namespace.parent_hierarchies().len(), 1);
        assert_eq!(namespace.parent_hierarchies()[0].basename(), DRT_NAME);
        // Namespace hierarchy is just its basename since parent is empty
1194
1195
1196

        // Component
        assert_eq!(component.basename(), COMPONENT_NAME);
1197
1198
1199
1200
        assert_eq!(component.parent_hierarchies().len(), 2);
        assert_eq!(component.parent_hierarchies()[0].basename(), DRT_NAME);
        assert_eq!(component.parent_hierarchies()[1].basename(), NAMESPACE_NAME);
        // Component hierarchy structure is validated by the individual assertions above
1201

1202
1203
        // Endpoint
        assert_eq!(endpoint.basename(), ENDPOINT_NAME);
1204
1205
1206
1207
1208
        assert_eq!(endpoint.parent_hierarchies().len(), 3);
        assert_eq!(endpoint.parent_hierarchies()[0].basename(), DRT_NAME);
        assert_eq!(endpoint.parent_hierarchies()[1].basename(), NAMESPACE_NAME);
        assert_eq!(endpoint.parent_hierarchies()[2].basename(), COMPONENT_NAME);
        // Endpoint hierarchy structure is validated by the individual assertions above
1209

1210
        // Relationships
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
        assert!(
            namespace
                .parent_hierarchies()
                .iter()
                .any(|h| h.basename() == drt.basename())
        );
        assert!(
            component
                .parent_hierarchies()
                .iter()
                .any(|h| h.basename() == namespace.basename())
        );
        assert!(
            endpoint
                .parent_hierarchies()
                .iter()
                .any(|h| h.basename() == component.basename())
        );
1229

1230
        // Depth
1231
1232
1233
1234
        assert_eq!(drt.parent_hierarchies().len(), 0);
        assert_eq!(namespace.parent_hierarchies().len(), 1);
        assert_eq!(component.parent_hierarchies().len(), 2);
        assert_eq!(endpoint.parent_hierarchies().len(), 3);
1235

1236
1237
1238
        // Invalid namespace behavior - sanitizes to "_123" and succeeds
        // @ryanolson intended to enable validation (see TODO comment in component.rs) but didn't turn it on,
        // so invalid characters are sanitized in MetricsRegistry rather than rejected.
1239
        let invalid_namespace = drt.namespace("@@123").unwrap();
1240
1241
1242
1243
        let result =
            invalid_namespace
                .metrics()
                .create_counter("test_counter", "A test counter", &[]);
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
        assert!(result.is_ok());
        if let Ok(counter) = &result {
            // Verify the namespace was sanitized to "_123" in the label
            let desc = counter.desc();
            let namespace_label = desc[0]
                .const_label_pairs
                .iter()
                .find(|l| l.name() == "dynamo_namespace")
                .expect("Should have dynamo_namespace label");
            assert_eq!(namespace_label.value(), "_123");
1254
        }
1255

1256
1257
        // Valid namespace works
        let valid_namespace = drt.namespace("ns567").unwrap();
1258
1259
        assert!(
            valid_namespace
1260
                .metrics()
1261
1262
1263
                .create_counter("test_counter", "A test counter", &[])
                .is_ok()
        );
1264
    }
1265

1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
    #[tokio::test]
    async fn test_expfmt_callback_only_registered_on_endpoint_is_included_once() {
        // Sanity test: if an expfmt callback is registered only on the endpoint registry,
        // scraping from the root (DRT) should still include it exactly once via the
        // child-registry traversal.
        let drt = create_test_drt_async().await;
        let namespace = drt.namespace("ns_expfmt_ep_only").unwrap();
        let component = namespace.component("comp_expfmt_ep_only").unwrap();
        let endpoint = component.endpoint("ep_expfmt_ep_only");

        let metric_line = "dynamo_component_active_decode_blocks{dp_rank=\"0\"} 0\n";
        let callback: PrometheusExpositionFormatCallback =
            Arc::new(move || Ok(metric_line.to_string()));

        endpoint
            .get_metrics_registry()
            .add_expfmt_callback(callback);

        let output = drt.metrics().prometheus_expfmt().unwrap();
        let occurrences = output
            .lines()
            .filter(|line| line == &metric_line.trim_end_matches('\n'))
            .count();

        assert_eq!(
            occurrences, 1,
            "endpoint-registered exposition callback should appear once, got {} occurrences\n\n{}",
            occurrences, output
        );
    }

1297
1298
    #[tokio::test]
    async fn test_recursive_namespace() {
1299
        // Create a distributed runtime for testing
1300
        let drt = create_test_drt_async().await;
1301

1302
1303
1304
1305
        // Create a deeply chained namespace: ns1.ns2.ns3
        let ns1 = drt.namespace("ns1").unwrap();
        let ns2 = ns1.namespace("ns2").unwrap();
        let ns3 = ns2.namespace("ns3").unwrap();
1306

1307
1308
        // Create a component in the deepest namespace
        let component = ns3.component("test-component").unwrap();
1309

1310
1311
        // Verify the hierarchy structure
        assert_eq!(ns1.basename(), "ns1");
1312
1313
1314
        assert_eq!(ns1.parent_hierarchies().len(), 1);
        assert_eq!(ns1.parent_hierarchies()[0].basename(), "");
        // ns1 hierarchy is just its basename since parent is empty
1315
1316

        assert_eq!(ns2.basename(), "ns2");
1317
1318
1319
1320
        assert_eq!(ns2.parent_hierarchies().len(), 2);
        assert_eq!(ns2.parent_hierarchies()[0].basename(), "");
        assert_eq!(ns2.parent_hierarchies()[1].basename(), "ns1");
        // ns2 hierarchy structure validated by parent assertions above
1321

1322
        assert_eq!(ns3.basename(), "ns3");
1323
1324
1325
1326
1327
        assert_eq!(ns3.parent_hierarchies().len(), 3);
        assert_eq!(ns3.parent_hierarchies()[0].basename(), "");
        assert_eq!(ns3.parent_hierarchies()[1].basename(), "ns1");
        assert_eq!(ns3.parent_hierarchies()[2].basename(), "ns2");
        // ns3 hierarchy structure validated by parent assertions above
1328

1329
        assert_eq!(component.basename(), "test-component");
1330
1331
1332
1333
1334
1335
        assert_eq!(component.parent_hierarchies().len(), 4);
        assert_eq!(component.parent_hierarchies()[0].basename(), "");
        assert_eq!(component.parent_hierarchies()[1].basename(), "ns1");
        assert_eq!(component.parent_hierarchies()[2].basename(), "ns2");
        assert_eq!(component.parent_hierarchies()[3].basename(), "ns3");
        // component hierarchy structure validated by parent assertions above
1336

1337
        println!("✓ Chained namespace test passed - all prefixes correct");
1338
1339
1340
1341
1342
    }
}

#[cfg(feature = "integration")]
#[cfg(test)]
1343
1344
mod test_metricsregistry_prometheus_fmt_outputs {
    use super::prometheus_names::name_prefix;
1345
    use super::*;
1346
    use crate::distributed::distributed_test_utils::create_test_drt_async;
1347
1348
1349
    use prometheus::Counter;
    use std::sync::Arc;

1350
1351
    #[tokio::test]
    async fn test_prometheusfactory_using_metrics_registry_trait() {
1352
        // Setup real DRT and registry using the test-friendly constructor
1353
        let drt = create_test_drt_async().await;
1354

1355
        // Use a simple constant namespace name
1356
        let namespace_name = "ns345";
1357

1358
        let namespace = drt.namespace(namespace_name).unwrap();
1359
1360
        let component = namespace.component("comp345").unwrap();
        let endpoint = component.endpoint("ep345");
1361
1362
1363

        // Test Counter creation
        let counter = endpoint
1364
            .metrics()
1365
            .create_counter("testcounter", "A test counter", &[])
1366
1367
1368
1369
1370
            .unwrap();
        counter.inc_by(123.456789);
        let epsilon = 0.01;
        assert!((counter.get() - 123.456789).abs() < epsilon);

1371
        let endpoint_output_raw = endpoint.metrics().prometheus_expfmt().unwrap();
1372
        println!("Endpoint output:");
1373
1374
        println!("{}", endpoint_output_raw);

1375
        let expected_endpoint_output = r#"# HELP dynamo_component_testcounter A test counter
1376
# TYPE dynamo_component_testcounter counter
1377
dynamo_component_testcounter{dynamo_component="comp345",dynamo_endpoint="ep345",dynamo_namespace="ns345"} 123.456789"#.to_string();
1378
1379

        assert_eq!(
1380
1381
            endpoint_output_raw.trim_end_matches('\n'),
            expected_endpoint_output.trim_end_matches('\n'),
1382
1383
            "\n=== ENDPOINT COMPARISON FAILED ===\n\
             Actual:\n{}\n\
1384
             Expected:\n{}\n\
1385
             ==============================",
1386
1387
            endpoint_output_raw,
            expected_endpoint_output
1388
1389
1390
1391
        );

        // Test Gauge creation
        let gauge = component
1392
            .metrics()
1393
            .create_gauge("testgauge", "A test gauge", &[])
1394
1395
1396
1397
1398
            .unwrap();
        gauge.set(50000.0);
        assert_eq!(gauge.get(), 50000.0);

        // Test Prometheus format output for Component (gauge + histogram)
1399
        let component_output_raw = component.metrics().prometheus_expfmt().unwrap();
1400
        println!("Component output:");
1401
1402
        println!("{}", component_output_raw);

1403
        let expected_component_output = r#"# HELP dynamo_component_testcounter A test counter
1404
# TYPE dynamo_component_testcounter counter
1405
dynamo_component_testcounter{dynamo_component="comp345",dynamo_endpoint="ep345",dynamo_namespace="ns345"} 123.456789
1406
1407
# HELP dynamo_component_testgauge A test gauge
# TYPE dynamo_component_testgauge gauge
1408
dynamo_component_testgauge{dynamo_component="comp345",dynamo_namespace="ns345"} 50000"#.to_string();
1409
1410

        assert_eq!(
1411
1412
            component_output_raw.trim_end_matches('\n'),
            expected_component_output.trim_end_matches('\n'),
1413
1414
            "\n=== COMPONENT COMPARISON FAILED ===\n\
             Actual:\n{}\n\
1415
             Expected:\n{}\n\
1416
             ==============================",
1417
1418
            component_output_raw,
            expected_component_output
1419
1420
1421
        );

        let intcounter = namespace
1422
            .metrics()
1423
            .create_intcounter("testintcounter", "A test int counter", &[])
1424
1425
1426
1427
1428
            .unwrap();
        intcounter.inc_by(12345);
        assert_eq!(intcounter.get(), 12345);

        // Test Prometheus format output for Namespace (int_counter + gauge + histogram)
1429
        let namespace_output_raw = namespace.metrics().prometheus_expfmt().unwrap();
1430
        println!("Namespace output:");
1431
1432
        println!("{}", namespace_output_raw);

1433
        let expected_namespace_output = r#"# HELP dynamo_component_testcounter A test counter
1434
# TYPE dynamo_component_testcounter counter
1435
dynamo_component_testcounter{dynamo_component="comp345",dynamo_endpoint="ep345",dynamo_namespace="ns345"} 123.456789
1436
1437
# HELP dynamo_component_testgauge A test gauge
# TYPE dynamo_component_testgauge gauge
1438
dynamo_component_testgauge{dynamo_component="comp345",dynamo_namespace="ns345"} 50000
1439
1440
# HELP dynamo_component_testintcounter A test int counter
# TYPE dynamo_component_testintcounter counter
1441
dynamo_component_testintcounter{dynamo_namespace="ns345"} 12345"#.to_string();
1442
1443

        assert_eq!(
1444
1445
            namespace_output_raw.trim_end_matches('\n'),
            expected_namespace_output.trim_end_matches('\n'),
1446
1447
            "\n=== NAMESPACE COMPARISON FAILED ===\n\
             Actual:\n{}\n\
1448
             Expected:\n{}\n\
1449
             ==============================",
1450
1451
            namespace_output_raw,
            expected_namespace_output
1452
1453
1454
        );

        // Test IntGauge creation
1455
        let intgauge = namespace
1456
            .metrics()
1457
            .create_intgauge("testintgauge", "A test int gauge", &[])
1458
1459
1460
1461
1462
            .unwrap();
        intgauge.set(42);
        assert_eq!(intgauge.get(), 42);

        // Test IntGaugeVec creation
1463
        let intgaugevec = namespace
1464
            .metrics()
1465
            .create_intgaugevec(
1466
                "testintgaugevec",
1467
                "A test int gauge vector",
1468
                &["instance", "status"],
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
                &[("service", "api")],
            )
            .unwrap();
        intgaugevec
            .with_label_values(&["server1", "active"])
            .set(10);
        intgaugevec
            .with_label_values(&["server2", "inactive"])
            .set(0);

1479
1480
        // Test CounterVec creation
        let countervec = endpoint
1481
            .metrics()
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
            .create_countervec(
                "testcountervec",
                "A test counter vector",
                &["method", "status"],
                &[("service", "api")],
            )
            .unwrap();
        countervec.with_label_values(&["GET", "200"]).inc_by(10.0);
        countervec.with_label_values(&["POST", "201"]).inc_by(5.0);

        // Test Histogram creation
        let histogram = component
1494
            .metrics()
1495
1496
1497
1498
1499
1500
1501
            .create_histogram("testhistogram", "A test histogram", &[], None)
            .unwrap();
        histogram.observe(1.0);
        histogram.observe(2.5);
        histogram.observe(4.0);

        // Test Prometheus format output for DRT (all metrics combined)
1502
        let drt_output_raw = drt.metrics().prometheus_expfmt().unwrap();
1503
        println!("DRT output:");
1504
1505
        println!("{}", drt_output_raw);

1506
1507
1508
        // The uptime_seconds value is dynamic (depends on elapsed wall-clock time),
        // so we check all other lines exactly and validate uptime separately.
        let expected_drt_output_without_uptime = r#"# HELP dynamo_component_testcounter A test counter
1509
# TYPE dynamo_component_testcounter counter
1510
dynamo_component_testcounter{dynamo_component="comp345",dynamo_endpoint="ep345",dynamo_namespace="ns345"} 123.456789
1511
1512
# HELP dynamo_component_testcountervec A test counter vector
# TYPE dynamo_component_testcountervec counter
1513
1514
dynamo_component_testcountervec{dynamo_component="comp345",dynamo_endpoint="ep345",dynamo_namespace="ns345",method="GET",service="api",status="200"} 10
dynamo_component_testcountervec{dynamo_component="comp345",dynamo_endpoint="ep345",dynamo_namespace="ns345",method="POST",service="api",status="201"} 5
1515
1516
# HELP dynamo_component_testgauge A test gauge
# TYPE dynamo_component_testgauge gauge
1517
dynamo_component_testgauge{dynamo_component="comp345",dynamo_namespace="ns345"} 50000
1518
1519
# HELP dynamo_component_testhistogram A test histogram
# TYPE dynamo_component_testhistogram histogram
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
dynamo_component_testhistogram_bucket{dynamo_component="comp345",dynamo_namespace="ns345",le="0.005"} 0
dynamo_component_testhistogram_bucket{dynamo_component="comp345",dynamo_namespace="ns345",le="0.01"} 0
dynamo_component_testhistogram_bucket{dynamo_component="comp345",dynamo_namespace="ns345",le="0.025"} 0
dynamo_component_testhistogram_bucket{dynamo_component="comp345",dynamo_namespace="ns345",le="0.05"} 0
dynamo_component_testhistogram_bucket{dynamo_component="comp345",dynamo_namespace="ns345",le="0.1"} 0
dynamo_component_testhistogram_bucket{dynamo_component="comp345",dynamo_namespace="ns345",le="0.25"} 0
dynamo_component_testhistogram_bucket{dynamo_component="comp345",dynamo_namespace="ns345",le="0.5"} 0
dynamo_component_testhistogram_bucket{dynamo_component="comp345",dynamo_namespace="ns345",le="1"} 1
dynamo_component_testhistogram_bucket{dynamo_component="comp345",dynamo_namespace="ns345",le="2.5"} 2
dynamo_component_testhistogram_bucket{dynamo_component="comp345",dynamo_namespace="ns345",le="5"} 3
dynamo_component_testhistogram_bucket{dynamo_component="comp345",dynamo_namespace="ns345",le="10"} 3
dynamo_component_testhistogram_bucket{dynamo_component="comp345",dynamo_namespace="ns345",le="+Inf"} 3
dynamo_component_testhistogram_sum{dynamo_component="comp345",dynamo_namespace="ns345"} 7.5
dynamo_component_testhistogram_count{dynamo_component="comp345",dynamo_namespace="ns345"} 3
1534
1535
# HELP dynamo_component_testintcounter A test int counter
# TYPE dynamo_component_testintcounter counter
1536
dynamo_component_testintcounter{dynamo_namespace="ns345"} 12345
1537
1538
# HELP dynamo_component_testintgauge A test int gauge
# TYPE dynamo_component_testintgauge gauge
1539
dynamo_component_testintgauge{dynamo_namespace="ns345"} 42
1540
1541
# HELP dynamo_component_testintgaugevec A test int gauge vector
# TYPE dynamo_component_testintgaugevec gauge
1542
dynamo_component_testintgaugevec{dynamo_namespace="ns345",instance="server1",service="api",status="active"} 10
1543
1544
dynamo_component_testintgaugevec{dynamo_namespace="ns345",instance="server2",service="api",status="inactive"} 0"#;

1545
        // Split actual output into non-uptime lines and validate the uptime value line.
1546
        let mut non_uptime_lines = Vec::new();
1547
        let mut saw_uptime_value = false;
1548
1549
1550
1551
1552
        for line in drt_output_raw.trim_end_matches('\n').lines() {
            if line.starts_with("dynamo_component_uptime_seconds ") {
                let val_str = line
                    .strip_prefix("dynamo_component_uptime_seconds ")
                    .unwrap();
1553
1554
                val_str.parse::<f64>().expect("uptime should be a float");
                saw_uptime_value = true;
1555
1556
1557
1558
1559
1560
1561
1562
            } else if line.starts_with("# HELP dynamo_component_uptime_seconds")
                || line.starts_with("# TYPE dynamo_component_uptime_seconds")
            {
                // Skip HELP/TYPE lines for uptime (we just verify it exists via the value)
            } else {
                non_uptime_lines.push(line);
            }
        }
1563
1564
1565
1566
        assert!(
            saw_uptime_value,
            "uptime_seconds metric should be present in initial scrape"
        );
1567

1568
        let actual_without_uptime = non_uptime_lines.join("\n");
1569
        assert_eq!(
1570
1571
1572
            actual_without_uptime,
            expected_drt_output_without_uptime.trim_end_matches('\n'),
            "\n=== DRT COMPARISON FAILED (excluding uptime) ===\n\
1573
             Expected:\n{}\n\
1574
             Actual:\n{}\n\
1575
             ==============================",
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
            expected_drt_output_without_uptime,
            actual_without_uptime
        );

        // Wait briefly so the uptime gauge is clearly positive on the next scrape.
        tokio::time::sleep(std::time::Duration::from_millis(10)).await;
        let drt_output_after = drt.metrics().prometheus_expfmt().unwrap();
        let uptime_after: f64 = drt_output_after
            .lines()
            .find(|l| l.starts_with("dynamo_component_uptime_seconds "))
            .expect("uptime_seconds metric should be present after sleep")
            .strip_prefix("dynamo_component_uptime_seconds ")
            .unwrap()
            .parse()
            .expect("uptime should be a float");
        assert!(
            uptime_after > 0.0,
            "uptime_seconds should be > 0 after 10ms sleep, got {}",
            uptime_after
1595
1596
1597
1598
        );

        println!("✓ All Prometheus format outputs verified successfully!");
    }
1599
1600
1601

    #[test]
    fn test_refactored_filter_functions() {
1602
        // Test data with component metrics
1603
1604
1605
1606
1607
1608
1609
        let test_input = r#"# HELP dynamo_component_requests Total requests
# TYPE dynamo_component_requests counter
dynamo_component_requests 42
# HELP dynamo_component_latency Response latency
# TYPE dynamo_component_latency histogram
dynamo_component_latency_bucket{le="0.1"} 10
dynamo_component_latency_bucket{le="0.5"} 25
1610
dynamo_component_errors_total 5"#;
1611
1612
1613

        // Test extract_metrics (only actual metric lines, excluding help/type)
        let metrics_only = super::test_helpers::extract_metrics(test_input);
1614
        assert_eq!(metrics_only.len(), 4); // 4 actual metric lines (excluding help/type)
1615
1616
1617
1618
1619
        assert!(
            metrics_only
                .iter()
                .all(|line| line.starts_with("dynamo_component") && !line.starts_with("#"))
        );
1620
1621
1622

        println!("✓ All refactored filter functions work correctly!");
    }
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714

    #[tokio::test]
    async fn test_same_metric_name_different_endpoints() {
        // Test that the same metric name can exist in different endpoints without collision.
        // This validates the multi-registry approach: each endpoint has its own registry,
        // and metrics are merged at scrape time with distinct labels.
        let drt = create_test_drt_async().await;
        let namespace = drt.namespace("ns_test").unwrap();
        let component = namespace.component("comp_test").unwrap();

        // Create two endpoints with the same metric name
        let ep1 = component.endpoint("ep1");
        let ep2 = component.endpoint("ep2");

        let counter1 = ep1
            .metrics()
            .create_counter("requests_total", "Total requests", &[])
            .unwrap();
        counter1.inc_by(100.0);

        let counter2 = ep2
            .metrics()
            .create_counter("requests_total", "Total requests", &[])
            .unwrap();
        counter2.inc_by(200.0);

        // Get merged Prometheus output from component level
        let output = component.metrics().prometheus_expfmt().unwrap();

        let expected_output = r#"# HELP dynamo_component_requests_total Total requests
# TYPE dynamo_component_requests_total counter
dynamo_component_requests_total{dynamo_component="comp_test",dynamo_endpoint="ep1",dynamo_namespace="ns_test"} 100
dynamo_component_requests_total{dynamo_component="comp_test",dynamo_endpoint="ep2",dynamo_namespace="ns_test"} 200"#;

        assert_eq!(
            output.trim_end_matches('\n'),
            expected_output.trim_end_matches('\n'),
            "\n=== MULTI-REGISTRY COMPARISON FAILED ===\n\
             Actual:\n{}\n\
             Expected:\n{}\n\
             ==============================",
            output,
            expected_output
        );

        println!("✓ Multi-registry prevents Prometheus collisions!");
    }

    #[tokio::test]
    async fn test_duplicate_series_warning() {
        // Test that duplicate series (same metric name + same labels) are detected and deduplicated.
        // This should log a warning and keep only one of the duplicate series.
        let drt = create_test_drt_async().await;
        let namespace = drt.namespace("ns_dup").unwrap();
        let component = namespace.component("comp_dup").unwrap();

        // Create two endpoints with counters that will have identical labels when scraped
        let ep1 = component.endpoint("ep_same");
        let ep2 = component.endpoint("ep_same"); // Same endpoint name = duplicate labels

        let counter1 = ep1
            .metrics()
            .create_counter("dup_metric", "Duplicate metric test", &[])
            .unwrap();
        counter1.inc_by(50.0);

        let counter2 = ep2
            .metrics()
            .create_counter("dup_metric", "Duplicate metric test", &[])
            .unwrap();
        counter2.inc_by(75.0);

        // Get merged output - duplicates should be deduplicated
        let output = component.metrics().prometheus_expfmt().unwrap();

        let expected_output = r#"# HELP dynamo_component_dup_metric Duplicate metric test
# TYPE dynamo_component_dup_metric counter
dynamo_component_dup_metric{dynamo_component="comp_dup",dynamo_endpoint="ep_same",dynamo_namespace="ns_dup"} 50"#;

        assert_eq!(
            output.trim_end_matches('\n'),
            expected_output.trim_end_matches('\n'),
            "\n=== DEDUPLICATION COMPARISON FAILED ===\n\
             Actual:\n{}\n\
             Expected:\n{}\n\
             ==============================",
            output,
            expected_output
        );

        println!("✓ Duplicate series detection and deduplication works!");
    }
1715
}