a3s-gateway 0.2.1

A3S Gateway - AI-native API gateway with reverse proxy, routing, and agent orchestration
Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
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
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
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
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
//! Autoscaler — periodic decision engine that monitors metrics and emits scale decisions
//!
//! Implements a Knative-style autoscaling formula:
//! `desired = ceil((in_flight + queue_depth) / (container_concurrency * target_utilization))`
//! clamped to `[min_replicas, max_replicas]`.

use crate::config::ScalingConfig;
use crate::error::Result;
use crate::scaling::executor::{ScaleDecision, ScaleDirection, ScaleExecutor};
use std::collections::HashMap;
use std::sync::Arc;
use std::time::Instant;

/// A snapshot of metrics for a single service
#[derive(Debug, Clone)]
pub struct ServiceMetricsSnapshot {
    /// Service name
    pub service: String,
    /// Number of healthy backends
    #[allow(dead_code)]
    pub healthy_backends: usize,
    /// Total in-flight requests across all backends
    pub in_flight: usize,
    /// Requests waiting in the buffer (scale-from-zero)
    pub queue_depth: usize,
}

/// Per-service autoscaler state
struct ServiceScaleState {
    /// Scaling configuration
    config: ScalingConfig,
    /// Last time a request was observed for this service
    last_request_at: Instant,
    /// Current known replica count
    current_replicas: u32,
}

/// Autoscaler that periodically evaluates metrics and executes scaling decisions
pub struct Autoscaler {
    /// Scale executor
    executor: Arc<dyn ScaleExecutor>,
    /// Per-service state
    services: HashMap<String, ServiceScaleState>,
}

impl Autoscaler {
    /// Create a new autoscaler with the given executor and service configs
    pub fn new(executor: Arc<dyn ScaleExecutor>, configs: HashMap<String, ScalingConfig>) -> Self {
        let now = Instant::now();
        let services = configs
            .into_iter()
            .map(|(name, config)| {
                let state = ServiceScaleState {
                    config,
                    last_request_at: now,
                    current_replicas: 0,
                };
                (name, state)
            })
            .collect();

        Self { executor, services }
    }

    /// Compute the desired replica count using the Knative formula.
    ///
    /// `desired = ceil((in_flight + queue_depth) / (cc * utilization))`
    /// clamped to `[min, max]`.
    ///
    /// Special cases:
    /// - `cc == 0` (unlimited): returns current replicas (no autoscaling decision)
    /// - `in_flight + queue_depth == 0` and past cooldown: returns `min_replicas`
    pub fn compute_desired_replicas(
        config: &ScalingConfig,
        snapshot: &ServiceMetricsSnapshot,
    ) -> u32 {
        let cc = config.container_concurrency;
        if cc == 0 {
            // Unlimited concurrency — no autoscaling signal from concurrency
            return config.min_replicas.max(1);
        }

        let total_load = (snapshot.in_flight + snapshot.queue_depth) as f64;
        if total_load == 0.0 {
            return config.min_replicas;
        }

        let effective_capacity = cc as f64 * config.target_utilization;
        if effective_capacity <= 0.0 {
            return config.max_replicas;
        }

        let desired = (total_load / effective_capacity).ceil() as u32;
        desired.clamp(config.min_replicas, config.max_replicas)
    }

    /// Evaluate a metrics snapshot and return a scaling decision if needed.
    ///
    /// Returns `None` if no scaling action is required (desired == current,
    /// or scale-down is blocked by cooldown).
    pub fn evaluate(&mut self, snapshot: &ServiceMetricsSnapshot) -> Option<ScaleDecision> {
        let state = self.services.get_mut(&snapshot.service)?;

        // Update last_request_at if there's active load
        if snapshot.in_flight > 0 || snapshot.queue_depth > 0 {
            state.last_request_at = Instant::now();
        }

        let desired = Self::compute_desired_replicas(&state.config, snapshot);
        let current = state.current_replicas;

        if desired == current {
            return None;
        }

        // Scale-down cooldown: only scale down if enough time has passed since last request
        if desired < current {
            let elapsed = state.last_request_at.elapsed().as_secs();
            if elapsed < state.config.scale_down_delay_secs {
                return None;
            }
        }

        let direction = if desired > current {
            ScaleDirection::Up
        } else {
            ScaleDirection::Down
        };

        let reason = format!(
            "{}: in_flight={}, queue={}, cc={}, util={:.0}%, current={}, desired={}",
            direction,
            snapshot.in_flight,
            snapshot.queue_depth,
            state.config.container_concurrency,
            state.config.target_utilization * 100.0,
            current,
            desired,
        );

        state.current_replicas = desired;

        Some(ScaleDecision {
            service: snapshot.service.clone(),
            direction,
            current_replicas: current,
            desired_replicas: desired,
            reason,
        })
    }

    /// Execute a single evaluation cycle for all services using the provided metrics function.
    pub async fn tick<F>(&mut self, metrics_fn: F) -> Vec<Result<()>>
    where
        F: Fn(&str) -> Option<ServiceMetricsSnapshot>,
    {
        let service_names: Vec<String> = self.services.keys().cloned().collect();
        let mut results = Vec::new();

        for name in &service_names {
            if let Some(snapshot) = metrics_fn(name) {
                if let Some(decision) = self.evaluate(&snapshot) {
                    tracing::info!(
                        service = decision.service,
                        direction = %decision.direction,
                        from = decision.current_replicas,
                        to = decision.desired_replicas,
                        reason = decision.reason,
                        "Autoscaler decision"
                    );
                    let result = self.executor.execute(&decision).await;
                    results.push(result.map(|_| ()));
                }
            }
        }

        results
    }

    /// Get the executor reference
    #[allow(dead_code)]
    pub fn executor(&self) -> &Arc<dyn ScaleExecutor> {
        &self.executor
    }

    /// Check if a service is registered with the autoscaler
    #[allow(dead_code)]
    pub fn has_service(&self, name: &str) -> bool {
        self.services.contains_key(name)
    }

    /// Number of services being autoscaled
    pub fn service_count(&self) -> usize {
        self.services.len()
    }

    /// Update the current replica count for a service (e.g., after querying the executor)
    #[allow(dead_code)]
    pub fn set_current_replicas(&mut self, service: &str, replicas: u32) {
        if let Some(state) = self.services.get_mut(service) {
            state.current_replicas = replicas;
        }
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::scaling::executor::MockScaleExecutor;

    fn default_config() -> ScalingConfig {
        ScalingConfig {
            min_replicas: 0,
            max_replicas: 10,
            container_concurrency: 10,
            target_utilization: 0.7,
            scale_down_delay_secs: 300,
            ..ScalingConfig::default()
        }
    }

    fn snapshot(service: &str, in_flight: usize, queue_depth: usize) -> ServiceMetricsSnapshot {
        ServiceMetricsSnapshot {
            service: service.into(),
            healthy_backends: 2,
            in_flight,
            queue_depth,
        }
    }

    // --- compute_desired_replicas ---

    #[test]
    fn test_formula_basic() {
        let config = default_config();
        // 10 in-flight, cc=10, util=0.7 → ceil(10 / 7.0) = ceil(1.43) = 2
        let snap = snapshot("svc", 10, 0);
        assert_eq!(Autoscaler::compute_desired_replicas(&config, &snap), 2);
    }

    #[test]
    fn test_formula_includes_queue_depth() {
        let config = default_config();
        // 5 in-flight + 5 queue = 10, cc=10, util=0.7 → ceil(10/7) = 2
        let snap = snapshot("svc", 5, 5);
        assert_eq!(Autoscaler::compute_desired_replicas(&config, &snap), 2);
    }

    #[test]
    fn test_formula_high_load() {
        let config = default_config();
        // 70 in-flight, cc=10, util=0.7 → ceil(70/7) = 10
        let snap = snapshot("svc", 70, 0);
        assert_eq!(Autoscaler::compute_desired_replicas(&config, &snap), 10);
    }

    #[test]
    fn test_formula_clamped_to_max() {
        let config = default_config();
        // 100 in-flight, cc=10, util=0.7 → ceil(100/7) = 15, clamped to 10
        let snap = snapshot("svc", 100, 0);
        assert_eq!(Autoscaler::compute_desired_replicas(&config, &snap), 10);
    }

    #[test]
    fn test_formula_clamped_to_min() {
        let config = ScalingConfig {
            min_replicas: 2,
            ..default_config()
        };
        // 1 in-flight, cc=10, util=0.7 → ceil(1/7) = 1, clamped to min=2
        let snap = snapshot("svc", 1, 0);
        assert_eq!(Autoscaler::compute_desired_replicas(&config, &snap), 2);
    }

    #[test]
    fn test_formula_zero_load_returns_min() {
        let config = default_config();
        let snap = snapshot("svc", 0, 0);
        assert_eq!(Autoscaler::compute_desired_replicas(&config, &snap), 0);
    }

    #[test]
    fn test_formula_zero_load_with_min() {
        let config = ScalingConfig {
            min_replicas: 1,
            ..default_config()
        };
        let snap = snapshot("svc", 0, 0);
        assert_eq!(Autoscaler::compute_desired_replicas(&config, &snap), 1);
    }

    #[test]
    fn test_formula_cc_zero_unlimited() {
        let config = ScalingConfig {
            container_concurrency: 0,
            min_replicas: 1,
            ..default_config()
        };
        let snap = snapshot("svc", 50, 0);
        // cc=0 means unlimited, returns max(min_replicas, 1)
        assert_eq!(Autoscaler::compute_desired_replicas(&config, &snap), 1);
    }

    #[test]
    fn test_formula_utilization_100_percent() {
        let config = ScalingConfig {
            target_utilization: 1.0,
            ..default_config()
        };
        // 10 in-flight, cc=10, util=1.0 → ceil(10/10) = 1
        let snap = snapshot("svc", 10, 0);
        assert_eq!(Autoscaler::compute_desired_replicas(&config, &snap), 1);
    }

    // --- evaluate ---

    #[test]
    fn test_evaluate_scale_up() {
        let mock = Arc::new(MockScaleExecutor::new());
        let mut configs = HashMap::new();
        configs.insert("svc".into(), default_config());
        let mut autoscaler = Autoscaler::new(mock, configs);

        let snap = snapshot("svc", 20, 0);
        let decision = autoscaler.evaluate(&snap).unwrap();
        assert_eq!(decision.direction, ScaleDirection::Up);
        assert_eq!(decision.current_replicas, 0);
        assert_eq!(decision.desired_replicas, 3); // ceil(20/7) = 3
    }

    #[test]
    fn test_evaluate_no_change() {
        let mock = Arc::new(MockScaleExecutor::new());
        let mut configs = HashMap::new();
        configs.insert("svc".into(), default_config());
        let mut autoscaler = Autoscaler::new(mock, configs);

        // Set current to match desired
        autoscaler.set_current_replicas("svc", 3);
        // 20 in-flight → desired=3, current=3 → no change
        let snap = snapshot("svc", 20, 0);
        assert!(autoscaler.evaluate(&snap).is_none());
    }

    #[test]
    fn test_evaluate_scale_down_blocked_by_cooldown() {
        let mock = Arc::new(MockScaleExecutor::new());
        let mut configs = HashMap::new();
        configs.insert("svc".into(), default_config());
        let mut autoscaler = Autoscaler::new(mock, configs);

        // Set current replicas high
        autoscaler.set_current_replicas("svc", 5);

        // Zero load → desired=0, but cooldown blocks it (last_request_at is recent)
        let snap = snapshot("svc", 0, 0);
        assert!(autoscaler.evaluate(&snap).is_none());
    }

    #[test]
    fn test_evaluate_unknown_service() {
        let mock = Arc::new(MockScaleExecutor::new());
        let mut autoscaler = Autoscaler::new(mock, HashMap::new());

        let snap = snapshot("unknown", 10, 0);
        assert!(autoscaler.evaluate(&snap).is_none());
    }

    #[test]
    fn test_evaluate_reason_formatting() {
        let mock = Arc::new(MockScaleExecutor::new());
        let mut configs = HashMap::new();
        configs.insert("svc".into(), default_config());
        let mut autoscaler = Autoscaler::new(mock, configs);

        let snap = snapshot("svc", 15, 5);
        let decision = autoscaler.evaluate(&snap).unwrap();
        assert!(decision.reason.contains("in_flight=15"));
        assert!(decision.reason.contains("queue=5"));
        assert!(decision.reason.contains("cc=10"));
    }

    // --- tick ---

    #[tokio::test]
    async fn test_tick_executes_decisions() {
        let mock = Arc::new(MockScaleExecutor::new());
        let mut configs = HashMap::new();
        configs.insert("svc".into(), default_config());
        let mut autoscaler = Autoscaler::new(mock.clone(), configs);

        let results = autoscaler
            .tick(|name| {
                if name == "svc" {
                    Some(snapshot("svc", 20, 0))
                } else {
                    None
                }
            })
            .await;

        assert_eq!(results.len(), 1);
        assert!(results[0].is_ok());
        assert_eq!(mock.decisions().len(), 1);
    }

    #[tokio::test]
    async fn test_tick_no_metrics_no_decision() {
        let mock = Arc::new(MockScaleExecutor::new());
        let mut configs = HashMap::new();
        configs.insert("svc".into(), default_config());
        let mut autoscaler = Autoscaler::new(mock.clone(), configs);

        let results = autoscaler.tick(|_| None).await;
        assert!(results.is_empty());
        assert!(mock.decisions().is_empty());
    }

    // --- construction ---

    #[test]
    fn test_autoscaler_has_service() {
        let mock = Arc::new(MockScaleExecutor::new());
        let mut configs = HashMap::new();
        configs.insert("api".into(), default_config());
        let autoscaler = Autoscaler::new(mock, configs);

        assert!(autoscaler.has_service("api"));
        assert!(!autoscaler.has_service("web"));
        assert_eq!(autoscaler.service_count(), 1);
    }

    #[test]
    fn test_autoscaler_executor_name() {
        let mock = Arc::new(MockScaleExecutor::new());
        let autoscaler = Autoscaler::new(mock, HashMap::new());
        assert_eq!(autoscaler.executor().name(), "mock");
    }
}