service_discovery.rs 37 KB
Newer Older
1
2
3
4
5
use std::{
    collections::{HashMap, HashSet},
    sync::{Arc, Mutex},
    time::Duration,
};
6

7
8
9
10
use futures::{StreamExt, TryStreamExt};
use k8s_openapi::api::core::v1::Pod;
use kube::{
    api::Api,
11
12
13
14
    runtime::{
        watcher::{watcher, Config},
        WatchStreamExt,
    },
15
16
    Client,
};
17
use rustls;
18
use tokio::{task, time};
19
use tracing::{debug, error, info, warn};
20

21
22
23
24
25
use crate::{
    core::{Job, WorkerManager},
    protocols::worker_spec::WorkerConfigRequest,
    server::AppContext,
};
26

27
28
29
30
31
32
33
#[derive(Debug, Clone)]
pub struct ServiceDiscoveryConfig {
    pub enabled: bool,
    pub selector: HashMap<String, String>,
    pub check_interval: Duration,
    pub port: u16,
    pub namespace: Option<String>,
34
35
36
37
38
39
    // PD mode specific configuration
    pub pd_mode: bool,
    pub prefill_selector: HashMap<String, String>,
    pub decode_selector: HashMap<String, String>,
    // Bootstrap port annotation specific to mooncake implementation
    pub bootstrap_port_annotation: String,
40
41
42
43
44
45
46
47
}

impl Default for ServiceDiscoveryConfig {
    fn default() -> Self {
        ServiceDiscoveryConfig {
            enabled: false,
            selector: HashMap::new(),
            check_interval: Duration::from_secs(60),
48
49
            port: 8000,
            namespace: None,
50
51
52
53
            pd_mode: false,
            prefill_selector: HashMap::new(),
            decode_selector: HashMap::new(),
            bootstrap_port_annotation: "sglang.ai/bootstrap-port".to_string(),
54
55
56
57
        }
    }
}

58
59
60
61
62
63
64
#[derive(Debug, Clone, PartialEq, Eq, Hash)]
pub enum PodType {
    Prefill,
    Decode,
    Regular,
}

65
66
67
68
69
70
#[derive(Debug, Clone, PartialEq, Eq, Hash)]
pub struct PodInfo {
    pub name: String,
    pub ip: String,
    pub status: String,
    pub is_ready: bool,
71
72
    pub pod_type: Option<PodType>,
    pub bootstrap_port: Option<u16>,
73
74
75
}

impl PodInfo {
76
77
78
79
80
    fn matches_selector(pod: &Pod, selector: &HashMap<String, String>) -> bool {
        if selector.is_empty() {
            return false;
        }

81
82
83
84
        pod.metadata
            .labels
            .as_ref()
            .is_some_and(|labels| selector.iter().all(|(k, v)| labels.get(k) == Some(v)))
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
    }

    pub fn should_include(pod: &Pod, config: &ServiceDiscoveryConfig) -> bool {
        if config.pd_mode {
            if config.prefill_selector.is_empty() && config.decode_selector.is_empty() {
                warn!("PD mode enabled but both prefill_selector and decode_selector are empty");
                return false;
            }
            Self::matches_selector(pod, &config.prefill_selector)
                || Self::matches_selector(pod, &config.decode_selector)
        } else {
            if config.selector.is_empty() {
                warn!("Regular mode enabled but selector is empty");
                return false;
            }
            Self::matches_selector(pod, &config.selector)
        }
    }

    pub fn from_pod(pod: &Pod, config: Option<&ServiceDiscoveryConfig>) -> Option<Self> {
105
106
107
108
109
110
111
112
113
114
115
116
117
118
        let name = pod.metadata.name.clone()?;
        let status = pod.status.clone()?;
        let pod_ip = status.pod_ip?;

        let is_ready = if let Some(conditions) = &status.conditions {
            conditions
                .iter()
                .any(|condition| condition.type_ == "Ready" && condition.status == "True")
        } else {
            false
        };

        let pod_status = status.phase.unwrap_or_else(|| "Unknown".to_string());

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
        let pod_type = if let Some(config) = config {
            if config.pd_mode {
                if Self::matches_selector(pod, &config.prefill_selector) {
                    Some(PodType::Prefill)
                } else if Self::matches_selector(pod, &config.decode_selector) {
                    Some(PodType::Decode)
                } else {
                    Some(PodType::Regular)
                }
            } else {
                Some(PodType::Regular)
            }
        } else {
            None
        };

        let bootstrap_port = if matches!(pod_type, Some(PodType::Prefill)) {
            if let Some(config) = config {
                pod.metadata
                    .annotations
                    .as_ref()
                    .and_then(|annotations| annotations.get(&config.bootstrap_port_annotation))
                    .and_then(|port_str| port_str.parse::<u16>().ok())
            } else {
                None
            }
        } else {
            None
        };

149
150
151
152
153
        Some(PodInfo {
            name,
            ip: pod_ip,
            status: pod_status,
            is_ready,
154
155
            pod_type,
            bootstrap_port,
156
157
158
159
160
161
162
163
        })
    }

    pub fn is_healthy(&self) -> bool {
        self.is_ready && self.status == "Running"
    }

    pub fn worker_url(&self, port: u16) -> String {
164
        // Default to http:// prefix; workflow will detect actual protocol (HTTP vs gRPC)
165
166
167
168
169
170
        format!("http://{}:{}", self.ip, port)
    }
}

pub async fn start_service_discovery(
    config: ServiceDiscoveryConfig,
171
    app_context: Arc<AppContext>,
172
173
174
175
176
177
178
179
180
181
) -> Result<task::JoinHandle<()>, kube::Error> {
    if !config.enabled {
        return Err(kube::Error::Api(kube::error::ErrorResponse {
            status: "Disabled".to_string(),
            message: "Service discovery is disabled".to_string(),
            reason: "ConfigurationError".to_string(),
            code: 400,
        }));
    }

182
183
    let _ = rustls::crypto::ring::default_provider().install_default();

184
185
    let client = Client::try_default().await?;

186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
    // Log the appropriate selectors based on mode
    if config.pd_mode {
        let prefill_selector = config
            .prefill_selector
            .iter()
            .map(|(k, v)| format!("{}={}", k, v))
            .collect::<Vec<_>>()
            .join(",");

        let decode_selector = config
            .decode_selector
            .iter()
            .map(|(k, v)| format!("{}={}", k, v))
            .collect::<Vec<_>>()
            .join(",");
201

202
        info!(
203
            "Starting K8s service discovery | PD mode | prefill: '{}' | decode: '{}'",
204
205
206
207
208
209
210
211
212
213
214
            prefill_selector, decode_selector
        );
    } else {
        let label_selector = config
            .selector
            .iter()
            .map(|(k, v)| format!("{}={}", k, v))
            .collect::<Vec<_>>()
            .join(",");

        info!(
215
            "Starting K8s service discovery | selector: '{}'",
216
217
218
            label_selector
        );
    }
219
220
221
222
223
224
225
226
227
228

    let handle = task::spawn(async move {
        let tracked_pods = Arc::new(Mutex::new(HashSet::new()));

        let pods: Api<Pod> = if let Some(namespace) = &config.namespace {
            Api::namespaced(client, namespace)
        } else {
            Api::all(client)
        };

229
        debug!("K8s service discovery initialized");
230

231
        let config_arc = Arc::new(config.clone());
232
233
        let port = config.port;

234
        let mut retry_delay = Duration::from_secs(1);
235
        const MAX_RETRY_DELAY: Duration = Duration::from_secs(300);
236

237
238
239
240
        loop {
            let watcher_config = Config::default();
            let watcher_stream = watcher(pods.clone(), watcher_config).applied_objects();

241
            let config_clone = Arc::clone(&config_arc);
242
243
244
            let tracked_pods_clone = Arc::clone(&tracked_pods);

            let filtered_stream = watcher_stream.filter_map(move |obj_res| {
245
                let config_inner = Arc::clone(&config_clone);
246
247
248
249

                async move {
                    match obj_res {
                        Ok(pod) => {
250
                            if PodInfo::should_include(&pod, &config_inner) {
251
252
253
254
255
256
257
258
259
260
261
                                Some(Ok(pod))
                            } else {
                                None
                            }
                        }
                        Err(e) => Some(Err(e)),
                    }
                }
            });

            let tracked_pods_clone2 = Arc::clone(&tracked_pods_clone);
262
            let app_context_clone = Arc::clone(&app_context);
263
            let config_clone2 = Arc::clone(&config_arc);
264
265
266
267

            match filtered_stream
                .try_for_each(move |pod| {
                    let tracked_pods_inner = Arc::clone(&tracked_pods_clone2);
268
                    let app_context_inner = Arc::clone(&app_context_clone);
269
                    let config_inner = Arc::clone(&config_clone2);
270
271

                    async move {
272
273
274
                        let pod_info = PodInfo::from_pod(&pod, Some(&config_inner));

                        if let Some(pod_info) = pod_info {
275
276
277
278
                            if pod.metadata.deletion_timestamp.is_some() {
                                handle_pod_deletion(
                                    &pod_info,
                                    tracked_pods_inner,
279
                                    app_context_inner,
280
281
282
283
                                    port,
                                )
                                .await;
                            } else {
284
285
286
                                handle_pod_event(
                                    &pod_info,
                                    tracked_pods_inner,
287
                                    app_context_inner,
288
289
290
291
                                    port,
                                    config_inner.pd_mode,
                                )
                                .await;
292
293
294
295
296
297
298
                            }
                        }
                        Ok(())
                    }
                })
                .await
            {
299
300
301
                Ok(_) => {
                    retry_delay = Duration::from_secs(1);
                }
302
303
                Err(err) => {
                    error!("Error in Kubernetes watcher: {}", err);
304
305
306
307
308
309
310
                    warn!(
                        "Retrying in {} seconds with exponential backoff",
                        retry_delay.as_secs()
                    );
                    time::sleep(retry_delay).await;

                    retry_delay = std::cmp::min(retry_delay * 2, MAX_RETRY_DELAY);
311
312
313
314
315
                }
            }

            warn!(
                "Kubernetes watcher exited, restarting in {} seconds",
316
                config_arc.check_interval.as_secs()
317
            );
318
            time::sleep(config_arc.check_interval).await;
319
320
321
322
323
324
325
326
327
        }
    });

    Ok(handle)
}

async fn handle_pod_event(
    pod_info: &PodInfo,
    tracked_pods: Arc<Mutex<HashSet<PodInfo>>>,
328
    app_context: Arc<AppContext>,
329
    port: u16,
330
    pd_mode: bool,
331
332
333
334
) {
    let worker_url = pod_info.worker_url(port);

    if pod_info.is_healthy() {
335
336
337
338
339
340
341
342
343
344
        let should_add = {
            let mut tracker = match tracked_pods.lock() {
                Ok(tracker) => tracker,
                Err(e) => {
                    error!("Failed to acquire tracked_pods lock: {}", e);
                    return;
                }
            };

            if tracker.contains(pod_info) {
345
                false
346
347
348
349
350
351
352
            } else {
                tracker.insert(pod_info.clone());
                true
            }
        };

        if should_add {
353
            info!(
354
                "Adding pod: {} | type: {:?} | url: {}",
355
                pod_info.name, pod_info.pod_type, worker_url
356
            );
357

358
359
360
361
362
            let worker_type = if pd_mode {
                match &pod_info.pod_type {
                    Some(PodType::Prefill) => Some("prefill".to_string()),
                    Some(PodType::Decode) => Some("decode".to_string()),
                    Some(PodType::Regular) | None => None,
363
364
                }
            } else {
365
                None
366
367
            };

368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
            let bootstrap_port = if pd_mode {
                match &pod_info.pod_type {
                    Some(PodType::Prefill) => pod_info.bootstrap_port,
                    _ => None,
                }
            } else {
                None
            };

            let config = WorkerConfigRequest {
                url: worker_url.clone(),
                model_id: None,
                worker_type,
                priority: None,
                cost: None,
                labels: HashMap::new(),
                bootstrap_port,
                tokenizer_path: None,
                reasoning_parser: None,
                tool_parser: None,
                chat_template: None,
                api_key: None,
390
391
392
393
394
395
396
397
398
399
                health_check_timeout_secs: app_context.router_config.health_check.timeout_secs,
                health_check_interval_secs: app_context
                    .router_config
                    .health_check
                    .check_interval_secs,
                health_success_threshold: app_context.router_config.health_check.success_threshold,
                health_failure_threshold: app_context.router_config.health_check.failure_threshold,
                max_connection_attempts: app_context.router_config.health_check.success_threshold
                    * 20,
                dp_aware: false,
400
401
            };

402
403
404
            let job = Job::AddWorker {
                config: Box::new(config.clone()),
            };
405

406
407
408
409
410
411
412
413
414
415
416
417
418
            if let Some(job_queue) = app_context.worker_job_queue.get() {
                match job_queue.submit(job).await {
                    Ok(_) => {
                        debug!("Worker addition job submitted for: {}", worker_url);
                    }
                    Err(e) => {
                        error!(
                            "Failed to submit worker addition job for {}: {}",
                            worker_url, e
                        );
                        if let Ok(mut tracker) = tracked_pods.lock() {
                            tracker.remove(pod_info);
                        }
419
                    }
420
                }
421
422
423
424
425
            } else {
                debug!(
                    "JobQueue not initialized, skipping async worker addition for: {}",
                    worker_url
                );
426
427
428
429
430
431
432
433
            }
        }
    }
}

async fn handle_pod_deletion(
    pod_info: &PodInfo,
    tracked_pods: Arc<Mutex<HashSet<PodInfo>>>,
434
    app_context: Arc<AppContext>,
435
436
437
438
    port: u16,
) {
    let worker_url = pod_info.worker_url(port);

439
440
441
442
443
444
445
446
447
448
449
450
    let was_tracked = {
        let mut tracked = match tracked_pods.lock() {
            Ok(tracked) => tracked,
            Err(e) => {
                error!("Failed to acquire tracked_pods lock during deletion: {}", e);
                return;
            }
        };
        tracked.remove(pod_info)
    };

    if was_tracked {
451
        info!(
452
            "Removing pod: {} | type: {:?} | url: {}",
453
            pod_info.name, pod_info.pod_type, worker_url
454
        );
455

456
457
        if let Err(e) = WorkerManager::remove_worker(&worker_url, &app_context) {
            error!("Failed to remove worker {}: {}", worker_url, e);
458
        }
459
460
    } else {
        debug!(
461
462
            "Pod deletion event for untracked/already removed pod: {} (type: {:?}). Worker URL: {}",
            pod_info.name, pod_info.pod_type, worker_url
463
464
465
        );
    }
}
466

467
468
#[cfg(test)]
mod tests {
469
470
471
472
473
    use k8s_openapi::{
        api::core::v1::{Pod, PodCondition, PodSpec, PodStatus},
        apimachinery::pkg::apis::meta::v1::{ObjectMeta, Time},
    };

474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
    use super::*;

    fn create_k8s_pod(
        name: Option<&str>,
        ip: Option<&str>,
        phase: Option<&str>,
        ready_status: Option<&str>,
        deletion_timestamp: Option<Time>,
    ) -> Pod {
        let mut pod = Pod {
            metadata: ObjectMeta {
                name: name.map(String::from),
                deletion_timestamp,
                ..Default::default()
            },
            spec: Some(PodSpec::default()),
            status: None,
        };

        if ip.is_some() || phase.is_some() || ready_status.is_some() {
            let mut pod_status = PodStatus {
                pod_ip: ip.map(String::from),
                phase: phase.map(String::from),
                conditions: None,
                ..Default::default()
            };

            if let Some(status_str) = ready_status {
                let condition = PodCondition {
                    type_: "Ready".to_string(),
                    status: status_str.to_string(),
                    last_probe_time: None,
                    last_transition_time: None,
                    message: None,
                    reason: None,
509
                    observed_generation: None,
510
511
512
513
                };
                pod_status.conditions = Some(vec![condition]);
            }
            pod.status = Some(pod_status);
514
        }
515
516
517
        pod
    }

518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
    fn create_pd_k8s_pod(name: &str, ip: &str, pod_type: &str, bootstrap_port: Option<u16>) -> Pod {
        let mut labels = std::collections::BTreeMap::new();
        labels.insert("app".to_string(), "sglang".to_string());
        labels.insert("component".to_string(), pod_type.to_string());

        let mut annotations = std::collections::BTreeMap::new();
        if let Some(port) = bootstrap_port {
            annotations.insert("sglang.ai/bootstrap-port".to_string(), port.to_string());
        }

        Pod {
            metadata: ObjectMeta {
                name: Some(name.to_string()),
                labels: Some(labels),
                annotations: Some(annotations),
                ..Default::default()
            },
            spec: Some(PodSpec::default()),
            status: Some(PodStatus {
                pod_ip: Some(ip.to_string()),
                phase: Some("Running".to_string()),
                conditions: Some(vec![PodCondition {
                    type_: "Ready".to_string(),
                    status: "True".to_string(),
                    last_probe_time: None,
                    last_transition_time: None,
                    message: None,
                    reason: None,
546
                    observed_generation: None,
547
548
549
550
551
552
                }]),
                ..Default::default()
            }),
        }
    }

553
    async fn create_test_app_context() -> Arc<AppContext> {
554
        use crate::{config::RouterConfig, middleware::TokenBucket};
555

556
557
558
        let router_config = RouterConfig {
            worker_startup_timeout_secs: 1,
            ..Default::default()
559
        };
560

561
562
        // Note: Using uninitialized queue for tests to avoid spawning background workers
        // Jobs submitted during tests will queue but not be processed
563
        Arc::new(AppContext {
564
            client: reqwest::Client::new(),
565
            router_config: router_config.clone(),
566
            rate_limiter: Some(Arc::new(TokenBucket::new(1000, 1000))),
567
568
569
570
            worker_registry: Arc::new(crate::core::WorkerRegistry::new()),
            policy_registry: Arc::new(crate::policies::PolicyRegistry::new(
                router_config.policy.clone(),
            )),
571
572
            tokenizer: None,
            reasoning_parser_factory: None,
573
            tool_parser_factory: None,
574
            router_manager: None,
575
            response_storage: Arc::new(crate::data_connector::MemoryResponseStorage::new()),
576
            conversation_storage: Arc::new(crate::data_connector::MemoryConversationStorage::new()),
577
578
579
            conversation_item_storage: Arc::new(
                crate::data_connector::MemoryConversationItemStorage::new(),
            ),
580
            load_monitor: None,
581
582
            configured_reasoning_parser: None,
            configured_tool_parser: None,
583
            worker_job_queue: Arc::new(std::sync::OnceLock::new()),
584
            workflow_engine: Arc::new(std::sync::OnceLock::new()),
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
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
    fn create_pd_config() -> ServiceDiscoveryConfig {
        let mut prefill_selector = HashMap::new();
        prefill_selector.insert("app".to_string(), "sglang".to_string());
        prefill_selector.insert("component".to_string(), "prefill".to_string());

        let mut decode_selector = HashMap::new();
        decode_selector.insert("app".to_string(), "sglang".to_string());
        decode_selector.insert("component".to_string(), "decode".to_string());

        ServiceDiscoveryConfig {
            enabled: true,
            selector: HashMap::new(),
            check_interval: Duration::from_secs(60),
            port: 8080,
            namespace: None,
            pd_mode: true,
            prefill_selector,
            decode_selector,
            bootstrap_port_annotation: "sglang.ai/bootstrap-port".to_string(),
        }
    }

    #[test]
    fn test_pod_info_should_include() {
        let config = create_pd_config();

        let prefill_pod = create_pd_k8s_pod("prefill-pod", "10.0.0.1", "prefill", Some(8081));
        assert!(PodInfo::should_include(&prefill_pod, &config));

        let decode_pod = create_pd_k8s_pod("decode-pod", "10.0.0.2", "decode", None);
        assert!(PodInfo::should_include(&decode_pod, &config));

        let unmatched_pod = create_pd_k8s_pod("other-pod", "10.0.0.3", "other", None);
        assert!(!PodInfo::should_include(&unmatched_pod, &config));

        let mut regular_config = ServiceDiscoveryConfig::default();
        regular_config
            .selector
            .insert("app".to_string(), "sglang".to_string());
        regular_config.pd_mode = false;

        let regular_pod = create_pd_k8s_pod("worker-pod", "10.0.0.4", "worker", None);
        assert!(PodInfo::should_include(&regular_pod, &regular_config));
    }

633
634
635
636
637
638
    #[test]
    fn test_service_discovery_config_default() {
        let config = ServiceDiscoveryConfig::default();
        assert!(!config.enabled);
        assert!(config.selector.is_empty());
        assert_eq!(config.check_interval, Duration::from_secs(60));
639
        assert_eq!(config.port, 8000);
640
        assert!(config.namespace.is_none());
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
        assert!(!config.pd_mode);
        assert!(config.prefill_selector.is_empty());
        assert!(config.decode_selector.is_empty());
        assert_eq!(config.bootstrap_port_annotation, "sglang.ai/bootstrap-port");
    }

    #[test]
    fn test_pod_type_enum() {
        let prefill = PodType::Prefill;
        let decode = PodType::Decode;
        let regular = PodType::Regular;

        assert_eq!(format!("{:?}", prefill), "Prefill");
        assert_eq!(format!("{:?}", decode), "Decode");
        assert_eq!(format!("{:?}", regular), "Regular");
656
657
658
659
660
661
662
663
664
665
666
    }

    #[test]
    fn test_pod_info_from_pod_valid() {
        let k8s_pod = create_k8s_pod(
            Some("test-pod"),
            Some("10.0.0.1"),
            Some("Running"),
            Some("True"),
            None,
        );
667
        let pod_info = PodInfo::from_pod(&k8s_pod, None).unwrap();
668
669
670
671
        assert_eq!(pod_info.name, "test-pod");
        assert_eq!(pod_info.ip, "10.0.0.1");
        assert_eq!(pod_info.status, "Running");
        assert!(pod_info.is_ready);
672
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
        assert!(pod_info.pod_type.is_none());
        assert!(pod_info.bootstrap_port.is_none());
    }

    #[test]
    fn test_pod_info_from_pod_with_pd_config_prefill() {
        let k8s_pod = create_pd_k8s_pod("prefill-pod", "10.0.0.1", "prefill", Some(8081));
        let config = create_pd_config();

        let pod_info = PodInfo::from_pod(&k8s_pod, Some(&config)).unwrap();
        assert_eq!(pod_info.name, "prefill-pod");
        assert_eq!(pod_info.ip, "10.0.0.1");
        assert_eq!(pod_info.status, "Running");
        assert!(pod_info.is_ready);
        assert_eq!(pod_info.pod_type, Some(PodType::Prefill));
        assert_eq!(pod_info.bootstrap_port, Some(8081));
    }

    #[test]
    fn test_pod_info_from_pod_with_pd_config_decode() {
        let k8s_pod = create_pd_k8s_pod("decode-pod", "10.0.0.2", "decode", None);
        let config = create_pd_config();

        let pod_info = PodInfo::from_pod(&k8s_pod, Some(&config)).unwrap();
        assert_eq!(pod_info.name, "decode-pod");
        assert_eq!(pod_info.ip, "10.0.0.2");
        assert_eq!(pod_info.status, "Running");
        assert!(pod_info.is_ready);
        assert_eq!(pod_info.pod_type, Some(PodType::Decode));
        assert!(pod_info.bootstrap_port.is_none());
    }

    #[test]
    fn test_pod_info_from_pod_with_pd_config_regular_mode() {
        let k8s_pod = create_pd_k8s_pod("regular-pod", "10.0.0.3", "worker", None);
        let mut config = create_pd_config();
708
        config.pd_mode = false;
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

        let pod_info = PodInfo::from_pod(&k8s_pod, Some(&config)).unwrap();
        assert_eq!(pod_info.name, "regular-pod");
        assert_eq!(pod_info.ip, "10.0.0.3");
        assert_eq!(pod_info.status, "Running");
        assert!(pod_info.is_ready);
        assert_eq!(pod_info.pod_type, Some(PodType::Regular));
        assert!(pod_info.bootstrap_port.is_none());
    }

    #[test]
    fn test_pod_info_from_pod_with_pd_config_unmatched_labels() {
        let k8s_pod = create_pd_k8s_pod("unknown-pod", "10.0.0.4", "unknown", None);
        let config = create_pd_config();

        let pod_info = PodInfo::from_pod(&k8s_pod, Some(&config)).unwrap();
        assert_eq!(pod_info.name, "unknown-pod");
        assert_eq!(pod_info.ip, "10.0.0.4");
        assert_eq!(pod_info.status, "Running");
        assert!(pod_info.is_ready);
        assert_eq!(pod_info.pod_type, Some(PodType::Regular));
        assert!(pod_info.bootstrap_port.is_none());
    }

    #[test]
    fn test_pod_info_from_pod_with_pd_config_invalid_bootstrap_port() {
        let mut pod = create_pd_k8s_pod("prefill-pod", "10.0.0.1", "prefill", None);
        pod.metadata.annotations.as_mut().unwrap().insert(
            "sglang.ai/bootstrap-port".to_string(),
            "invalid".to_string(),
        );
        let config = create_pd_config();

        let pod_info = PodInfo::from_pod(&pod, Some(&config)).unwrap();
        assert_eq!(pod_info.pod_type, Some(PodType::Prefill));
744
        assert!(pod_info.bootstrap_port.is_none());
745
746
747
748
749
750
751
752
753
754
755
    }

    #[test]
    fn test_pod_info_from_pod_not_ready() {
        let k8s_pod = create_k8s_pod(
            Some("test-pod"),
            Some("10.0.0.1"),
            Some("Running"),
            Some("False"),
            None,
        );
756
        let pod_info = PodInfo::from_pod(&k8s_pod, None).unwrap();
757
758
759
760
761
762
763
764
765
766
767
768
        assert!(!pod_info.is_ready);
    }

    #[test]
    fn test_pod_info_from_pod_no_conditions() {
        let k8s_pod = create_k8s_pod(
            Some("test-pod"),
            Some("10.0.0.1"),
            Some("Running"),
            None,
            None,
        );
769
        let pod_info = PodInfo::from_pod(&k8s_pod, None).unwrap();
770
771
772
773
774
775
        assert!(!pod_info.is_ready);
    }

    #[test]
    fn test_pod_info_from_pod_missing_name() {
        let k8s_pod = create_k8s_pod(None, Some("10.0.0.1"), Some("Running"), Some("True"), None);
776
        assert!(PodInfo::from_pod(&k8s_pod, None).is_none());
777
778
779
780
781
    }

    #[test]
    fn test_pod_info_from_pod_missing_ip() {
        let k8s_pod = create_k8s_pod(Some("test-pod"), None, Some("Running"), Some("True"), None);
782
        assert!(PodInfo::from_pod(&k8s_pod, None).is_none());
783
784
785
786
787
    }

    #[test]
    fn test_pod_info_from_pod_missing_status_phase() {
        let k8s_pod = create_k8s_pod(Some("test-pod"), Some("10.0.0.1"), None, Some("True"), None);
788
        let pod_info = PodInfo::from_pod(&k8s_pod, None).unwrap();
789
790
791
792
793
794
795
        assert_eq!(pod_info.status, "Unknown");
    }

    #[test]
    fn test_pod_info_from_pod_no_status_object() {
        let mut k8s_pod = create_k8s_pod(Some("test-pod"), None, None, None, None);
        k8s_pod.status = None;
796
        assert!(PodInfo::from_pod(&k8s_pod, None).is_none());
797
798
799
800
801
802
803
804
805
    }

    #[test]
    fn test_pod_info_is_healthy() {
        let healthy_pod = PodInfo {
            name: "p1".into(),
            ip: "1.1.1.1".into(),
            status: "Running".into(),
            is_ready: true,
806
807
            pod_type: None,
            bootstrap_port: None,
808
809
810
811
812
813
814
815
        };
        assert!(healthy_pod.is_healthy());

        let not_ready_pod = PodInfo {
            name: "p2".into(),
            ip: "1.1.1.2".into(),
            status: "Running".into(),
            is_ready: false,
816
817
            pod_type: None,
            bootstrap_port: None,
818
819
820
821
822
823
824
825
        };
        assert!(!not_ready_pod.is_healthy());

        let not_running_pod = PodInfo {
            name: "p3".into(),
            ip: "1.1.1.3".into(),
            status: "Pending".into(),
            is_ready: true,
826
827
            pod_type: None,
            bootstrap_port: None,
828
829
830
831
        };
        assert!(!not_running_pod.is_healthy());
    }

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
    #[test]
    fn test_pod_info_equality_with_pod_type() {
        let pod1 = PodInfo {
            name: "pod1".into(),
            ip: "1.2.3.4".into(),
            status: "Running".into(),
            is_ready: true,
            pod_type: Some(PodType::Prefill),
            bootstrap_port: Some(8081),
        };

        let pod2 = PodInfo {
            name: "pod1".into(),
            ip: "1.2.3.4".into(),
            status: "Running".into(),
            is_ready: true,
            pod_type: Some(PodType::Prefill),
            bootstrap_port: Some(8081),
        };

        let pod3 = PodInfo {
            name: "pod1".into(),
            ip: "1.2.3.4".into(),
            status: "Running".into(),
            is_ready: true,
            pod_type: Some(PodType::Decode),
            bootstrap_port: None,
        };

        assert_eq!(pod1, pod2);
        assert_ne!(pod1, pod3);
    }

865
866
    #[tokio::test]
    async fn test_handle_pod_event_add_unhealthy_pod() {
867
        let app_context = create_test_app_context().await;
868
869
870
871
872
873
        let tracked_pods = Arc::new(Mutex::new(HashSet::new()));
        let pod_info = PodInfo {
            name: "pod1".into(),
            ip: "1.2.3.4".into(),
            status: "Pending".into(),
            is_ready: false,
874
875
            pod_type: None,
            bootstrap_port: None,
876
877
878
879
880
881
        };
        let port = 8080u16;

        handle_pod_event(
            &pod_info,
            Arc::clone(&tracked_pods),
882
            Arc::clone(&app_context),
883
            port,
884
            false, // pd_mode = false
885
886
887
888
889
890
891
892
        )
        .await;

        assert!(!tracked_pods.lock().unwrap().contains(&pod_info));
    }

    #[tokio::test]
    async fn test_handle_pod_deletion_non_existing_pod() {
893
        let app_context = create_test_app_context().await;
894
895
896
897
898
899
        let tracked_pods = Arc::new(Mutex::new(HashSet::new()));
        let pod_info = PodInfo {
            name: "pod1".into(),
            ip: "1.2.3.4".into(),
            status: "Running".into(),
            is_ready: true,
900
901
            pod_type: None,
            bootstrap_port: None,
902
903
904
905
906
907
        };
        let port = 8080u16;

        handle_pod_deletion(
            &pod_info,
            Arc::clone(&tracked_pods),
908
            Arc::clone(&app_context),
909
910
911
912
913
            port,
        )
        .await;

        assert!(tracked_pods.lock().unwrap().is_empty());
914
    }
915
916
917

    #[tokio::test]
    async fn test_handle_pd_pod_event_prefill_pod() {
918
        let app_context = create_test_app_context().await;
919
920
921
922
923
924
925
926
927
928
929
930
931
932
        let tracked_pods = Arc::new(Mutex::new(HashSet::new()));
        let pod_info = PodInfo {
            name: "prefill-pod".into(),
            ip: "1.2.3.4".into(),
            status: "Running".into(),
            is_ready: true,
            pod_type: Some(PodType::Prefill),
            bootstrap_port: Some(8081),
        };
        let port = 8080u16;

        handle_pod_event(
            &pod_info,
            Arc::clone(&tracked_pods),
933
            Arc::clone(&app_context),
934
            port,
935
            true, // pd_mode = true for PD pod
936
937
938
        )
        .await;

939
940
941
942
943
944
        // With fully async control plane, pod is tracked and job is queued
        // Worker registration and validation happen in background job
        assert!(tracked_pods.lock().unwrap().contains(&pod_info));

        // Note: In tests with uninitialized queue, background jobs don't process
        // Worker won't appear in registry until background job runs (in production)
945
946
947
948
    }

    #[tokio::test]
    async fn test_handle_pd_pod_event_decode_pod() {
949
        let app_context = create_test_app_context().await;
950
951
952
953
954
955
956
957
958
959
960
961
962
963
        let tracked_pods = Arc::new(Mutex::new(HashSet::new()));
        let pod_info = PodInfo {
            name: "decode-pod".into(),
            ip: "1.2.3.5".into(),
            status: "Running".into(),
            is_ready: true,
            pod_type: Some(PodType::Decode),
            bootstrap_port: None,
        };
        let port = 8080u16;

        handle_pod_event(
            &pod_info,
            Arc::clone(&tracked_pods),
964
            Arc::clone(&app_context),
965
            port,
966
            true, // pd_mode = true for PD pod
967
968
969
        )
        .await;

970
971
972
973
974
975
        // With fully async control plane, pod is tracked and job is queued
        // Worker registration and validation happen in background job
        assert!(tracked_pods.lock().unwrap().contains(&pod_info));

        // Note: In tests with uninitialized queue, background jobs don't process
        // Worker won't appear in registry until background job runs (in production)
976
977
978
979
    }

    #[tokio::test]
    async fn test_handle_pd_pod_deletion_tracked_pod() {
980
        let app_context = create_test_app_context().await;
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
        let tracked_pods = Arc::new(Mutex::new(HashSet::new()));
        let pod_info = PodInfo {
            name: "test-pod".into(),
            ip: "1.2.3.4".into(),
            status: "Running".into(),
            is_ready: true,
            pod_type: Some(PodType::Prefill),
            bootstrap_port: Some(8081),
        };

        // Add pod to tracked set first
        {
            let mut tracked = tracked_pods.lock().unwrap();
            tracked.insert(pod_info.clone());
        }

        let port = 8080u16;

        handle_pod_deletion(
            &pod_info,
            Arc::clone(&tracked_pods),
1002
            Arc::clone(&app_context),
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
            port,
        )
        .await;

        // Pod should be removed from tracking
        assert!(!tracked_pods.lock().unwrap().contains(&pod_info));
    }

    #[tokio::test]
    async fn test_handle_pd_pod_deletion_untracked_pod() {
1013
        let app_context = create_test_app_context().await;
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
        let tracked_pods = Arc::new(Mutex::new(HashSet::new()));
        let pod_info = PodInfo {
            name: "untracked-pod".into(),
            ip: "1.2.3.4".into(),
            status: "Running".into(),
            is_ready: true,
            pod_type: Some(PodType::Decode),
            bootstrap_port: None,
        };
        let port = 8080u16;

        // Don't add pod to tracked set

        handle_pod_deletion(
            &pod_info,
            Arc::clone(&tracked_pods),
1030
            Arc::clone(&app_context),
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
            port,
        )
        .await;

        // Tracked set should remain empty
        assert!(tracked_pods.lock().unwrap().is_empty());
    }

    #[tokio::test]
    async fn test_unified_handler_regular_mode() {
1041
        let app_context = create_test_app_context().await;
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
        let tracked_pods = Arc::new(Mutex::new(HashSet::new()));
        let pod_info = PodInfo {
            name: "regular-pod".into(),
            ip: "1.2.3.4".into(),
            status: "Running".into(),
            is_ready: true,
            pod_type: Some(PodType::Regular),
            bootstrap_port: None,
        };
        let port = 8080u16;

        handle_pod_event(
            &pod_info,
            Arc::clone(&tracked_pods),
1056
            Arc::clone(&app_context),
1057
1058
1059
1060
1061
            port,
            false, // pd_mode = false
        )
        .await;

1062
1063
1064
1065
1066
1067
1068
        // With fully async control plane, pod is tracked and job is queued
        // In regular mode (pd_mode=false), worker_type defaults to Regular
        // Worker registration and validation happen in background job
        assert!(tracked_pods.lock().unwrap().contains(&pod_info));

        // Note: In tests with uninitialized queue, background jobs don't process
        // Worker won't appear in registry until background job runs (in production)
1069
1070
1071
1072
    }

    #[tokio::test]
    async fn test_unified_handler_pd_mode_with_prefill() {
1073
        let app_context = create_test_app_context().await;
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
        let tracked_pods = Arc::new(Mutex::new(HashSet::new()));
        let pod_info = PodInfo {
            name: "prefill-pod".into(),
            ip: "1.2.3.4".into(),
            status: "Running".into(),
            is_ready: true,
            pod_type: Some(PodType::Prefill),
            bootstrap_port: Some(8081),
        };
        let port = 8080u16;

        handle_pod_event(
            &pod_info,
            Arc::clone(&tracked_pods),
1088
            Arc::clone(&app_context),
1089
1090
1091
1092
1093
            port,
            true, // pd_mode = true
        )
        .await;

1094
1095
1096
1097
1098
1099
        // With fully async control plane, pod is tracked and job is queued
        // Worker registration and validation happen in background job
        assert!(tracked_pods.lock().unwrap().contains(&pod_info));

        // Note: In tests with uninitialized queue, background jobs don't process
        // Worker won't appear in registry until background job runs (in production)
1100
1101
1102
1103
    }

    #[tokio::test]
    async fn test_unified_handler_deletion_with_pd_mode() {
1104
        let app_context = create_test_app_context().await;
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
        let tracked_pods = Arc::new(Mutex::new(HashSet::new()));
        let pod_info = PodInfo {
            name: "decode-pod".into(),
            ip: "1.2.3.4".into(),
            status: "Running".into(),
            is_ready: true,
            pod_type: Some(PodType::Decode),
            bootstrap_port: None,
        };

        // Add pod to tracked set first
        {
            let mut tracked = tracked_pods.lock().unwrap();
            tracked.insert(pod_info.clone());
        }

        let port = 8080u16;

        handle_pod_deletion(
            &pod_info,
            Arc::clone(&tracked_pods),
1126
            Arc::clone(&app_context),
1127
1128
1129
1130
1131
1132
1133
            port,
        )
        .await;

        // Pod should be removed from tracking
        assert!(!tracked_pods.lock().unwrap().contains(&pod_info));
    }
1134
}