metrics.rs 64.1 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
10
pub mod prometheus_names;

11
use parking_lot::Mutex;
12
13
14
15
16
use std::collections::HashSet;
use std::sync::Arc;

use crate::component::ComponentBuilder;
use anyhow;
17
18
use once_cell::sync::Lazy;
use regex::Regex;
19
20
21
use std::any::Any;
use std::collections::HashMap;

22
23
// Import commonly used items to avoid verbose prefixes
use prometheus_names::{
24
25
    build_component_metric_name, labels, name_prefix, sanitize_prometheus_label,
    sanitize_prometheus_name, work_handler,
26
27
28
29
};

// Pipeline imports for endpoint creation
use crate::pipeline::{
30
31
    AsyncEngine, AsyncEngineContextProvider, Error, ManyOut, ResponseStream, SingleIn, async_trait,
    network::Ingress,
32
33
34
35
36
};
use crate::protocols::annotated::Annotated;
use crate::stream;
use crate::stream::StreamExt;

37
38
39
// Prometheus imports
use prometheus::Encoder;

40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
/// 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(())
}

55
56
57
/// ==============================
/// Prometheus section
/// ==============================
58
59
60
61
62
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
/// 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)
    }
}

115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
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)
    }
}

130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
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)
    }
}

145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
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)
    }
}

160
161
162
163
164
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
// 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)
    }
}

194
195
196
197
198
199
/// ==============================
/// Metrics section
/// ==============================
/// Public helper function to create metrics - accessible for Python bindings
pub fn create_metric<T: PrometheusMetric, H: MetricsHierarchy + ?Sized>(
    hierarchy: &H,
200
201
202
203
204
    metric_name: &str,
    metric_desc: &str,
    labels: &[(&str, &str)],
    buckets: Option<Vec<f64>>,
    const_labels: Option<&[&str]>,
205
) -> anyhow::Result<T> {
206
    // Validate that user-provided labels don't have duplicate keys
207
    validate_no_duplicate_label_keys(labels)?;
208
    // Note: stored labels functionality has been removed
209

210
211
    let basename = hierarchy.basename();
    let parent_hierarchies = hierarchy.parent_hierarchies();
212

213
214
215
216
    // 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());
217

218
    let metric_name = build_component_metric_name(metric_name);
219

220
    // Build updated_labels: auto-labels first, then `labels` + stored labels
221
222
    let mut updated_labels: Vec<(String, String)> = Vec::new();

223
224
225
226
227
228
229
230
231
232
233
234
    // 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
            ));
235
        }
236
    }
237

238
239
240
241
242
243
244
245
    // 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));
246
            }
247
        }
248
249
250
251
252
253
254
    }
    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));
255
256
            }
        }
257
258
259
260
261
262
263
    }
    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));
264
265
266
267
268
269
270
271
272
273
            }
        }
    }

    // Add user labels
    updated_labels.extend(
        labels
            .iter()
            .map(|(k, v)| ((*k).to_string(), (*v).to_string())),
    );
274
    // Note: stored labels functionality has been removed
275
276

    // Handle different metric types
277
    let prometheus_metric = if std::any::TypeId::of::<T>()
278
        == std::any::TypeId::of::<prometheus::CounterVec>()
279
    {
280
281
282
283
284
285
286
287
288
289
290
291
292
293
        // 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)?
294
295
296
    } 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
297
298
        if buckets.is_some() {
            return Err(anyhow::anyhow!(
299
                "buckets parameter is not valid for GaugeVec"
300
301
302
303
304
305
306
            ));
        }
        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
307
            .ok_or_else(|| anyhow::anyhow!("GaugeVec requires const_labels parameter"))?;
308
        T::with_opts_and_label_names(opts, label_names)?
309
310
311
312
313
314
315
316
317
318
319
320
321
    } 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)?
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
    } 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)?
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
    } 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)?
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
    } 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)?
    };

372
373
374
    let collector: Box<dyn prometheus::core::Collector> = Box::new(prometheus_metric.clone());
    hierarchy.get_metrics_registry().add_metric(collector)?;

375
    Ok(prometheus_metric)
376
377
}

378
379
380
381
382
/// 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,
}
383

384
385
386
impl<H: MetricsHierarchy> Metrics<H> {
    pub fn new(hierarchy: H) -> Self {
        Self { hierarchy }
387
388
389
390
391
392
    }

    // TODO: Add support for additional Prometheus metric types:
    // - Counter: ✅ IMPLEMENTED - create_counter()
    // - CounterVec: ✅ IMPLEMENTED - create_countervec()
    // - Gauge: ✅ IMPLEMENTED - create_gauge()
393
    // - GaugeVec: ✅ IMPLEMENTED - create_gaugevec()
394
    // - GaugeHistogram: create_gauge_histogram() - for gauge histograms
395
396
397
    // - Histogram: ✅ IMPLEMENTED - create_histogram()
    // - HistogramVec with custom buckets: create_histogram_with_buckets()
    // - Info: create_info() - for info metrics with labels
398
399
400
401
    // - IntCounter: ✅ IMPLEMENTED - create_intcounter()
    // - IntCounterVec: ✅ IMPLEMENTED - create_intcountervec()
    // - IntGauge: ✅ IMPLEMENTED - create_intgauge()
    // - IntGaugeVec: ✅ IMPLEMENTED - create_intgaugevec()
402
    // - Stateset: create_stateset() - for state-based metrics
403
404
405
    // - Summary: create_summary() - for quantiles and sum/count metrics
    // - SummaryVec: create_summary_vec() - for labeled summaries
    // - Untyped: create_untyped() - for untyped metrics
406
407
408
    //
    // 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
409
410

    /// Create a Counter metric
411
    pub fn create_counter(
412
413
414
415
        &self,
        name: &str,
        description: &str,
        labels: &[(&str, &str)],
416
    ) -> anyhow::Result<prometheus::Counter> {
417
        create_metric(&self.hierarchy, name, description, labels, None, None)
418
419
    }

420
    /// Create a CounterVec metric with label names (for dynamic labels)
421
    pub fn create_countervec(
422
423
424
        &self,
        name: &str,
        description: &str,
425
426
        const_labels: &[&str],
        const_label_values: &[(&str, &str)],
427
    ) -> anyhow::Result<prometheus::CounterVec> {
428
        create_metric(
429
            &self.hierarchy,
430
431
432
433
434
435
            name,
            description,
            const_label_values,
            None,
            Some(const_labels),
        )
436
437
    }

438
    /// Create a Gauge metric
439
    pub fn create_gauge(
440
441
442
443
        &self,
        name: &str,
        description: &str,
        labels: &[(&str, &str)],
444
    ) -> anyhow::Result<prometheus::Gauge> {
445
        create_metric(&self.hierarchy, name, description, labels, None, None)
446
447
    }

448
    /// Create a GaugeVec metric with label names (for dynamic labels)
449
    pub fn create_gaugevec(
450
451
452
453
454
455
456
        &self,
        name: &str,
        description: &str,
        const_labels: &[&str],
        const_label_values: &[(&str, &str)],
    ) -> anyhow::Result<prometheus::GaugeVec> {
        create_metric(
457
            &self.hierarchy,
458
459
460
461
462
463
464
465
            name,
            description,
            const_label_values,
            None,
            Some(const_labels),
        )
    }

466
    /// Create a Histogram metric with custom buckets
467
    pub fn create_histogram(
468
469
470
471
472
        &self,
        name: &str,
        description: &str,
        labels: &[(&str, &str)],
        buckets: Option<Vec<f64>>,
473
    ) -> anyhow::Result<prometheus::Histogram> {
474
        create_metric(&self.hierarchy, name, description, labels, buckets, None)
475
476
    }

477
    /// Create an IntCounter metric
478
    pub fn create_intcounter(
479
480
481
482
        &self,
        name: &str,
        description: &str,
        labels: &[(&str, &str)],
483
    ) -> anyhow::Result<prometheus::IntCounter> {
484
        create_metric(&self.hierarchy, name, description, labels, None, None)
485
486
487
    }

    /// Create an IntCounterVec metric with label names (for dynamic labels)
488
    pub fn create_intcountervec(
489
490
491
492
493
        &self,
        name: &str,
        description: &str,
        const_labels: &[&str],
        const_label_values: &[(&str, &str)],
494
    ) -> anyhow::Result<prometheus::IntCounterVec> {
495
        create_metric(
496
            &self.hierarchy,
497
498
499
500
501
502
503
504
505
            name,
            description,
            const_label_values,
            None,
            Some(const_labels),
        )
    }

    /// Create an IntGauge metric
506
    pub fn create_intgauge(
507
508
509
510
        &self,
        name: &str,
        description: &str,
        labels: &[(&str, &str)],
511
    ) -> anyhow::Result<prometheus::IntGauge> {
512
        create_metric(&self.hierarchy, name, description, labels, None, None)
513
514
515
    }

    /// Create an IntGaugeVec metric with label names (for dynamic labels)
516
    pub fn create_intgaugevec(
517
518
519
520
521
        &self,
        name: &str,
        description: &str,
        const_labels: &[&str],
        const_label_values: &[(&str, &str)],
522
    ) -> anyhow::Result<prometheus::IntGaugeVec> {
523
        create_metric(
524
            &self.hierarchy,
525
526
527
528
529
530
531
532
533
            name,
            description,
            const_label_values,
            None,
            Some(const_labels),
        )
    }

    /// Get metrics in Prometheus text format
534
    pub fn prometheus_expfmt(&self) -> anyhow::Result<String> {
535
        self.hierarchy
536
            .get_metrics_registry()
537
            .prometheus_expfmt_combined()
538
539
540
    }
}

541
542
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
/// 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>;

605
606
607
608
609
/// 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)]
610
pub struct MetricsRegistry {
611
612
613
    /// 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>>,
614

615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
    /// 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>>>,

630
631
632
633
634
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
    /// 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 {
666
            prometheus_registry: Arc::new(std::sync::RwLock::new(prometheus::Registry::new())),
667
            child_registries: Arc::new(std::sync::RwLock::new(Vec::new())),
668
669
670
671
672
            prometheus_update_callbacks: Arc::new(std::sync::RwLock::new(Vec::new())),
            prometheus_expfmt_callbacks: Arc::new(std::sync::RwLock::new(Vec::new())),
        }
    }

673
674
675
676
677
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
741
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
    /// 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 {
                    tracing::error!("Error executing metrics callback: {}", e);
                }
            }
        }

        // 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)
    }

832
833
834
835
836
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
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
    /// 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) => {
                    tracing::error!("Error executing exposition text callback: {}", e);
                }
            }
        }
        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))
    }

    /// 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()
    }
}

914
#[cfg(test)]
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
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| {
936
            line.starts_with(&format!("{}_", name_prefix::COMPONENT))
937
938
939
940
941
942
943
944
945
946
947
948
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
                && !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))
    }
988
989
}

990
#[cfg(test)]
991
mod test_metricsregistry_units {
992
993
994
    use super::*;

    #[test]
995
996
997
    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");
998
        assert_eq!(result, "dynamo_component_requests");
999

1000
        let result = build_component_metric_name("counter");
1001
        assert_eq!(result, "dynamo_component_counter");
1002
1003
    }

1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
    #[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);
1033

1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
        // 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!");
    }

1057
    #[test]
1058
    fn test_metrics_registry_entry_callbacks() {
1059
        use crate::MetricsRegistry;
1060
        use std::sync::atomic::{AtomicUsize, Ordering};
1061

1062
1063
        // Test 1: Basic callback execution with counter increments
        {
1064
            let registry = MetricsRegistry::new();
1065
1066
1067
1068
1069
            let counter = Arc::new(AtomicUsize::new(0));

            // Add callbacks with different increment values
            for increment in [1, 10, 100] {
                let counter_clone = counter.clone();
1070
                registry.add_update_callback(Arc::new(move || {
1071
1072
1073
1074
                    counter_clone.fetch_add(increment, Ordering::SeqCst);
                    Ok(())
                }));
            }
1075

1076
1077
            // Verify counter starts at 0
            assert_eq!(counter.load(Ordering::SeqCst), 0);
1078

1079
            // First execution
1080
            let results = registry.execute_update_callbacks();
1081
1082
1083
            assert_eq!(results.len(), 3);
            assert!(results.iter().all(|r| r.is_ok()));
            assert_eq!(counter.load(Ordering::SeqCst), 111); // 1 + 10 + 100
1084

1085
            // Second execution - callbacks should be reusable
1086
            let results = registry.execute_update_callbacks();
1087
1088
            assert_eq!(results.len(), 3);
            assert_eq!(counter.load(Ordering::SeqCst), 222); // 111 + 111
1089

1090
1091
1092
            // Test cloning - cloned entry shares callbacks (callbacks are Arc-wrapped)
            let cloned = registry.clone();
            assert_eq!(cloned.execute_update_callbacks().len(), 3);
1093
            assert_eq!(counter.load(Ordering::SeqCst), 333); // 222 + 111
1094
1095
1096
1097

            // Original still has callbacks and shares the same Arc
            registry.execute_update_callbacks();
            assert_eq!(counter.load(Ordering::SeqCst), 444); // 333 + 111
1098
        }
1099

1100
1101
        // Test 2: Mixed success and error callbacks
        {
1102
            let registry = MetricsRegistry::new();
1103
1104
1105
1106
            let counter = Arc::new(AtomicUsize::new(0));

            // Successful callback
            let counter_clone = counter.clone();
1107
            registry.add_update_callback(Arc::new(move || {
1108
1109
1110
1111
1112
                counter_clone.fetch_add(1, Ordering::SeqCst);
                Ok(())
            }));

            // Error callback
1113
            registry.add_update_callback(Arc::new(|| Err(anyhow::anyhow!("Simulated error"))));
1114
1115
1116

            // Another successful callback
            let counter_clone = counter.clone();
1117
            registry.add_update_callback(Arc::new(move || {
1118
1119
1120
1121
1122
                counter_clone.fetch_add(10, Ordering::SeqCst);
                Ok(())
            }));

            // Execute and verify mixed results
1123
            let results = registry.execute_update_callbacks();
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
            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"
            );
1134

1135
1136
            // Verify successful callbacks still executed
            assert_eq!(counter.load(Ordering::SeqCst), 11); // 1 + 10
1137

1138
            // Execute again - errors should be consistent
1139
            let results = registry.execute_update_callbacks();
1140
1141
1142
            assert!(results[1].is_err());
            assert_eq!(counter.load(Ordering::SeqCst), 22); // 11 + 11
        }
1143

1144
1145
        // Test 3: Empty registry
        {
1146
1147
            let registry = MetricsRegistry::new();
            let results = registry.execute_update_callbacks();
1148
1149
1150
1151
            assert_eq!(results.len(), 0);
        }
    }
}
1152

1153
1154
1155
1156
#[cfg(feature = "integration")]
#[cfg(test)]
mod test_metricsregistry_prefixes {
    use super::*;
1157
    use crate::distributed::distributed_test_utils::create_test_drt_async;
1158
1159
    use prometheus::core::Collector;

1160
1161
    #[tokio::test]
    async fn test_hierarchical_prefixes_and_parent_hierarchies() {
1162
        let drt = create_test_drt_async().await;
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173

        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);
1174
1175
        assert_eq!(drt.parent_hierarchies().len(), 0);
        // DRT hierarchy is just its basename (empty string)
1176
1177
1178

        // Namespace
        assert_eq!(namespace.basename(), NAMESPACE_NAME);
1179
1180
1181
        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
1182
1183
1184

        // Component
        assert_eq!(component.basename(), COMPONENT_NAME);
1185
1186
1187
1188
        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
1189

1190
1191
        // Endpoint
        assert_eq!(endpoint.basename(), ENDPOINT_NAME);
1192
1193
1194
1195
1196
        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
1197

1198
        // Relationships
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
        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())
        );
1217

1218
        // Depth
1219
1220
1221
1222
        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);
1223

1224
1225
1226
        // 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.
1227
        let invalid_namespace = drt.namespace("@@123").unwrap();
1228
1229
1230
1231
        let result =
            invalid_namespace
                .metrics()
                .create_counter("test_counter", "A test counter", &[]);
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
        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");
1242
        }
1243

1244
1245
        // Valid namespace works
        let valid_namespace = drt.namespace("ns567").unwrap();
1246
1247
        assert!(
            valid_namespace
1248
                .metrics()
1249
1250
1251
                .create_counter("test_counter", "A test counter", &[])
                .is_ok()
        );
1252
    }
1253

1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
    #[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
        );
    }

1285
1286
    #[tokio::test]
    async fn test_recursive_namespace() {
1287
        // Create a distributed runtime for testing
1288
        let drt = create_test_drt_async().await;
1289

1290
1291
1292
1293
        // 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();
1294

1295
1296
        // Create a component in the deepest namespace
        let component = ns3.component("test-component").unwrap();
1297

1298
1299
        // Verify the hierarchy structure
        assert_eq!(ns1.basename(), "ns1");
1300
1301
1302
        assert_eq!(ns1.parent_hierarchies().len(), 1);
        assert_eq!(ns1.parent_hierarchies()[0].basename(), "");
        // ns1 hierarchy is just its basename since parent is empty
1303
1304

        assert_eq!(ns2.basename(), "ns2");
1305
1306
1307
1308
        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
1309

1310
        assert_eq!(ns3.basename(), "ns3");
1311
1312
1313
1314
1315
        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
1316

1317
        assert_eq!(component.basename(), "test-component");
1318
1319
1320
1321
1322
1323
        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
1324

1325
        println!("✓ Chained namespace test passed - all prefixes correct");
1326
1327
1328
1329
1330
    }
}

#[cfg(feature = "integration")]
#[cfg(test)]
1331
1332
mod test_metricsregistry_prometheus_fmt_outputs {
    use super::prometheus_names::name_prefix;
1333
    use super::*;
1334
    use crate::distributed::distributed_test_utils::create_test_drt_async;
1335
1336
1337
    use prometheus::Counter;
    use std::sync::Arc;

1338
1339
    #[tokio::test]
    async fn test_prometheusfactory_using_metrics_registry_trait() {
1340
        // Setup real DRT and registry using the test-friendly constructor
1341
        let drt = create_test_drt_async().await;
1342

1343
        // Use a simple constant namespace name
1344
        let namespace_name = "ns345";
1345

1346
        let namespace = drt.namespace(namespace_name).unwrap();
1347
1348
        let component = namespace.component("comp345").unwrap();
        let endpoint = component.endpoint("ep345");
1349
1350
1351

        // Test Counter creation
        let counter = endpoint
1352
            .metrics()
1353
            .create_counter("testcounter", "A test counter", &[])
1354
1355
1356
1357
1358
            .unwrap();
        counter.inc_by(123.456789);
        let epsilon = 0.01;
        assert!((counter.get() - 123.456789).abs() < epsilon);

1359
        let endpoint_output_raw = endpoint.metrics().prometheus_expfmt().unwrap();
1360
        println!("Endpoint output:");
1361
1362
        println!("{}", endpoint_output_raw);

1363
        let expected_endpoint_output = r#"# HELP dynamo_component_testcounter A test counter
1364
# TYPE dynamo_component_testcounter counter
1365
dynamo_component_testcounter{dynamo_component="comp345",dynamo_endpoint="ep345",dynamo_namespace="ns345"} 123.456789"#.to_string();
1366
1367

        assert_eq!(
1368
1369
            endpoint_output_raw.trim_end_matches('\n'),
            expected_endpoint_output.trim_end_matches('\n'),
1370
1371
            "\n=== ENDPOINT COMPARISON FAILED ===\n\
             Actual:\n{}\n\
1372
             Expected:\n{}\n\
1373
             ==============================",
1374
1375
            endpoint_output_raw,
            expected_endpoint_output
1376
1377
1378
1379
        );

        // Test Gauge creation
        let gauge = component
1380
            .metrics()
1381
            .create_gauge("testgauge", "A test gauge", &[])
1382
1383
1384
1385
1386
            .unwrap();
        gauge.set(50000.0);
        assert_eq!(gauge.get(), 50000.0);

        // Test Prometheus format output for Component (gauge + histogram)
1387
        let component_output_raw = component.metrics().prometheus_expfmt().unwrap();
1388
        println!("Component output:");
1389
1390
        println!("{}", component_output_raw);

1391
        let expected_component_output = r#"# HELP dynamo_component_testcounter A test counter
1392
# TYPE dynamo_component_testcounter counter
1393
dynamo_component_testcounter{dynamo_component="comp345",dynamo_endpoint="ep345",dynamo_namespace="ns345"} 123.456789
1394
1395
# HELP dynamo_component_testgauge A test gauge
# TYPE dynamo_component_testgauge gauge
1396
dynamo_component_testgauge{dynamo_component="comp345",dynamo_namespace="ns345"} 50000"#.to_string();
1397
1398

        assert_eq!(
1399
1400
            component_output_raw.trim_end_matches('\n'),
            expected_component_output.trim_end_matches('\n'),
1401
1402
            "\n=== COMPONENT COMPARISON FAILED ===\n\
             Actual:\n{}\n\
1403
             Expected:\n{}\n\
1404
             ==============================",
1405
1406
            component_output_raw,
            expected_component_output
1407
1408
1409
        );

        let intcounter = namespace
1410
            .metrics()
1411
            .create_intcounter("testintcounter", "A test int counter", &[])
1412
1413
1414
1415
1416
            .unwrap();
        intcounter.inc_by(12345);
        assert_eq!(intcounter.get(), 12345);

        // Test Prometheus format output for Namespace (int_counter + gauge + histogram)
1417
        let namespace_output_raw = namespace.metrics().prometheus_expfmt().unwrap();
1418
        println!("Namespace output:");
1419
1420
        println!("{}", namespace_output_raw);

1421
        let expected_namespace_output = r#"# HELP dynamo_component_testcounter A test counter
1422
# TYPE dynamo_component_testcounter counter
1423
dynamo_component_testcounter{dynamo_component="comp345",dynamo_endpoint="ep345",dynamo_namespace="ns345"} 123.456789
1424
1425
# HELP dynamo_component_testgauge A test gauge
# TYPE dynamo_component_testgauge gauge
1426
dynamo_component_testgauge{dynamo_component="comp345",dynamo_namespace="ns345"} 50000
1427
1428
# HELP dynamo_component_testintcounter A test int counter
# TYPE dynamo_component_testintcounter counter
1429
dynamo_component_testintcounter{dynamo_namespace="ns345"} 12345"#.to_string();
1430
1431

        assert_eq!(
1432
1433
            namespace_output_raw.trim_end_matches('\n'),
            expected_namespace_output.trim_end_matches('\n'),
1434
1435
            "\n=== NAMESPACE COMPARISON FAILED ===\n\
             Actual:\n{}\n\
1436
             Expected:\n{}\n\
1437
             ==============================",
1438
1439
            namespace_output_raw,
            expected_namespace_output
1440
1441
1442
        );

        // Test IntGauge creation
1443
        let intgauge = namespace
1444
            .metrics()
1445
            .create_intgauge("testintgauge", "A test int gauge", &[])
1446
1447
1448
1449
1450
            .unwrap();
        intgauge.set(42);
        assert_eq!(intgauge.get(), 42);

        // Test IntGaugeVec creation
1451
        let intgaugevec = namespace
1452
            .metrics()
1453
            .create_intgaugevec(
1454
                "testintgaugevec",
1455
                "A test int gauge vector",
1456
                &["instance", "status"],
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
                &[("service", "api")],
            )
            .unwrap();
        intgaugevec
            .with_label_values(&["server1", "active"])
            .set(10);
        intgaugevec
            .with_label_values(&["server2", "inactive"])
            .set(0);

1467
1468
        // Test CounterVec creation
        let countervec = endpoint
1469
            .metrics()
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
            .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
1482
            .metrics()
1483
1484
1485
1486
1487
1488
1489
            .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)
1490
        let drt_output_raw = drt.metrics().prometheus_expfmt().unwrap();
1491
        println!("DRT output:");
1492
1493
        println!("{}", drt_output_raw);

1494
        let expected_drt_output = r#"# HELP dynamo_component_testcounter A test counter
1495
# TYPE dynamo_component_testcounter counter
1496
dynamo_component_testcounter{dynamo_component="comp345",dynamo_endpoint="ep345",dynamo_namespace="ns345"} 123.456789
1497
1498
# HELP dynamo_component_testcountervec A test counter vector
# TYPE dynamo_component_testcountervec counter
1499
1500
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
1501
1502
# HELP dynamo_component_testgauge A test gauge
# TYPE dynamo_component_testgauge gauge
1503
dynamo_component_testgauge{dynamo_component="comp345",dynamo_namespace="ns345"} 50000
1504
1505
# HELP dynamo_component_testhistogram A test histogram
# TYPE dynamo_component_testhistogram histogram
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
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
1520
1521
# HELP dynamo_component_testintcounter A test int counter
# TYPE dynamo_component_testintcounter counter
1522
dynamo_component_testintcounter{dynamo_namespace="ns345"} 12345
1523
1524
# HELP dynamo_component_testintgauge A test int gauge
# TYPE dynamo_component_testintgauge gauge
1525
dynamo_component_testintgauge{dynamo_namespace="ns345"} 42
1526
1527
# HELP dynamo_component_testintgaugevec A test int gauge vector
# TYPE dynamo_component_testintgaugevec gauge
1528
dynamo_component_testintgaugevec{dynamo_namespace="ns345",instance="server1",service="api",status="active"} 10
1529
1530
1531
1532
dynamo_component_testintgaugevec{dynamo_namespace="ns345",instance="server2",service="api",status="inactive"} 0
# HELP dynamo_component_uptime_seconds Total uptime of the DistributedRuntime in seconds
# TYPE dynamo_component_uptime_seconds gauge
dynamo_component_uptime_seconds 0"#.to_string();
1533
1534

        assert_eq!(
1535
1536
            drt_output_raw.trim_end_matches('\n'),
            expected_drt_output.trim_end_matches('\n'),
1537
1538
            "\n=== DRT COMPARISON FAILED ===\n\
             Expected:\n{}\n\
1539
             Actual (filtered):\n{}\n\
1540
             ==============================",
1541
1542
            expected_drt_output,
            drt_output_raw
1543
1544
1545
1546
        );

        println!("✓ All Prometheus format outputs verified successfully!");
    }
1547
1548
1549

    #[test]
    fn test_refactored_filter_functions() {
1550
        // Test data with component metrics
1551
1552
1553
1554
1555
1556
1557
        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
1558
dynamo_component_errors_total 5"#;
1559
1560
1561

        // Test extract_metrics (only actual metric lines, excluding help/type)
        let metrics_only = super::test_helpers::extract_metrics(test_input);
1562
        assert_eq!(metrics_only.len(), 4); // 4 actual metric lines (excluding help/type)
1563
1564
1565
1566
1567
        assert!(
            metrics_only
                .iter()
                .all(|line| line.starts_with("dynamo_component") && !line.starts_with("#"))
        );
1568
1569
1570

        println!("✓ All refactored filter functions work correctly!");
    }
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
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

    #[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!");
    }
1663
}