metadata.rs 12.5 KB
Newer Older
1
// SPDX-FileCopyrightText: Copyright (c) 2024-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
2
3
4
// SPDX-License-Identifier: Apache-2.0

use anyhow::Result;
5
use std::collections::{HashMap, HashSet};
6
7
use std::sync::Arc;

8
use super::{DiscoveryInstance, DiscoveryInstanceId, DiscoveryQuery};
9
10
11
12
13

/// Metadata stored on each pod and exposed via HTTP endpoint
/// This struct holds all discovery registrations for this pod instance
#[derive(Debug, Clone, serde::Serialize, serde::Deserialize)]
pub struct DiscoveryMetadata {
14
    /// Registered endpoint instances (key: path string from EndpointInstanceId::to_path())
15
    endpoints: HashMap<String, DiscoveryInstance>,
16
    /// Registered model card instances (key: path string from ModelCardInstanceId::to_path())
17
18
19
20
21
22
23
24
25
26
27
28
29
30
    model_cards: HashMap<String, DiscoveryInstance>,
}

impl DiscoveryMetadata {
    /// Create a new empty metadata store
    pub fn new() -> Self {
        Self {
            endpoints: HashMap::new(),
            model_cards: HashMap::new(),
        }
    }

    /// Register an endpoint instance
    pub fn register_endpoint(&mut self, instance: DiscoveryInstance) -> Result<()> {
31
32
33
34
35
36
37
38
        match instance.id() {
            DiscoveryInstanceId::Endpoint(key) => {
                self.endpoints.insert(key.to_path(), instance);
                Ok(())
            }
            DiscoveryInstanceId::Model(_) => {
                anyhow::bail!("Cannot register non-endpoint instance as endpoint")
            }
39
40
41
42
43
        }
    }

    /// Register a model card instance
    pub fn register_model_card(&mut self, instance: DiscoveryInstance) -> Result<()> {
44
45
46
47
48
49
50
51
        match instance.id() {
            DiscoveryInstanceId::Model(key) => {
                self.model_cards.insert(key.to_path(), instance);
                Ok(())
            }
            DiscoveryInstanceId::Endpoint(_) => {
                anyhow::bail!("Cannot register non-model-card instance as model card")
            }
52
53
54
55
56
        }
    }

    /// Unregister an endpoint instance
    pub fn unregister_endpoint(&mut self, instance: &DiscoveryInstance) -> Result<()> {
57
58
59
60
61
62
63
64
        match instance.id() {
            DiscoveryInstanceId::Endpoint(key) => {
                self.endpoints.remove(&key.to_path());
                Ok(())
            }
            DiscoveryInstanceId::Model(_) => {
                anyhow::bail!("Cannot unregister non-endpoint instance as endpoint")
            }
65
66
67
68
69
        }
    }

    /// Unregister a model card instance
    pub fn unregister_model_card(&mut self, instance: &DiscoveryInstance) -> Result<()> {
70
71
72
73
74
75
76
77
        match instance.id() {
            DiscoveryInstanceId::Model(key) => {
                self.model_cards.remove(&key.to_path());
                Ok(())
            }
            DiscoveryInstanceId::Endpoint(_) => {
                anyhow::bail!("Cannot unregister non-model-card instance as model card")
            }
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
        }
    }

    /// Get all registered endpoints
    pub fn get_all_endpoints(&self) -> Vec<DiscoveryInstance> {
        self.endpoints.values().cloned().collect()
    }

    /// Get all registered model cards
    pub fn get_all_model_cards(&self) -> Vec<DiscoveryInstance> {
        self.model_cards.values().cloned().collect()
    }

    /// Get all registered instances (endpoints and model cards)
    pub fn get_all(&self) -> Vec<DiscoveryInstance> {
        self.endpoints
            .values()
            .chain(self.model_cards.values())
            .cloned()
            .collect()
    }

    /// Filter this metadata by query
    pub fn filter(&self, query: &DiscoveryQuery) -> Vec<DiscoveryInstance> {
        let all_instances = match query {
            DiscoveryQuery::AllEndpoints
            | DiscoveryQuery::NamespacedEndpoints { .. }
            | DiscoveryQuery::ComponentEndpoints { .. }
            | DiscoveryQuery::Endpoint { .. } => self.get_all_endpoints(),

            DiscoveryQuery::AllModels
            | DiscoveryQuery::NamespacedModels { .. }
            | DiscoveryQuery::ComponentModels { .. }
            | DiscoveryQuery::EndpointModels { .. } => self.get_all_model_cards(),
        };

        filter_instances(all_instances, query)
    }
}

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

/// Filter instances by query predicate
fn filter_instances(
    instances: Vec<DiscoveryInstance>,
    query: &DiscoveryQuery,
) -> Vec<DiscoveryInstance> {
    match query {
        DiscoveryQuery::AllEndpoints | DiscoveryQuery::AllModels => instances,

        DiscoveryQuery::NamespacedEndpoints { namespace } => instances
            .into_iter()
            .filter(|inst| match inst {
                DiscoveryInstance::Endpoint(i) => &i.namespace == namespace,
                _ => false,
            })
            .collect(),

        DiscoveryQuery::ComponentEndpoints {
            namespace,
            component,
        } => instances
            .into_iter()
            .filter(|inst| match inst {
                DiscoveryInstance::Endpoint(i) => {
                    &i.namespace == namespace && &i.component == component
                }
                _ => false,
            })
            .collect(),

        DiscoveryQuery::Endpoint {
            namespace,
            component,
            endpoint,
        } => instances
            .into_iter()
            .filter(|inst| match inst {
                DiscoveryInstance::Endpoint(i) => {
                    &i.namespace == namespace
                        && &i.component == component
                        && &i.endpoint == endpoint
                }
                _ => false,
            })
            .collect(),

        DiscoveryQuery::NamespacedModels { namespace } => instances
            .into_iter()
            .filter(|inst| match inst {
                DiscoveryInstance::Model { namespace: ns, .. } => ns == namespace,
                _ => false,
            })
            .collect(),

        DiscoveryQuery::ComponentModels {
            namespace,
            component,
        } => instances
            .into_iter()
            .filter(|inst| match inst {
                DiscoveryInstance::Model {
                    namespace: ns,
                    component: comp,
                    ..
                } => ns == namespace && comp == component,
                _ => false,
            })
            .collect(),

        DiscoveryQuery::EndpointModels {
            namespace,
            component,
            endpoint,
        } => instances
            .into_iter()
            .filter(|inst| match inst {
                DiscoveryInstance::Model {
                    namespace: ns,
                    component: comp,
                    endpoint: ep,
                    ..
                } => ns == namespace && comp == component && ep == endpoint,
                _ => false,
            })
            .collect(),
    }
}

/// Snapshot of all discovered instances and their metadata
#[derive(Clone, Debug)]
pub struct MetadataSnapshot {
    /// Map of instance_id -> metadata
    pub instances: HashMap<u64, Arc<DiscoveryMetadata>>,
216
217
218
    /// Map of instance_id -> CR generation for change detection
    /// Keys match `instances` keys exactly - only ready pods with CRs are included
    pub generations: HashMap<u64, i64>,
219
220
221
222
223
224
225
226
227
228
    /// Sequence number for debugging
    pub sequence: u64,
    /// Timestamp for observability
    pub timestamp: std::time::Instant,
}

impl MetadataSnapshot {
    pub fn empty() -> Self {
        Self {
            instances: HashMap::new(),
229
            generations: HashMap::new(),
230
231
232
233
234
            sequence: 0,
            timestamp: std::time::Instant::now(),
        }
    }

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
    /// Compare with previous snapshot and return true if changed.
    /// Logs diagnostic info about what changed.
    /// This is done on the basis of the generation of the DynamoWorkerMetadata CRs that are owned by ready workers
    pub fn has_changes_from(&self, prev: &MetadataSnapshot) -> bool {
        if self.generations == prev.generations {
            tracing::trace!(
                "Snapshot (seq={}): no changes, {} instances",
                self.sequence,
                self.instances.len()
            );
            return false;
        }

        // Compute diff for logging
        let curr_ids: HashSet<u64> = self.generations.keys().copied().collect();
        let prev_ids: HashSet<u64> = prev.generations.keys().copied().collect();

        let added: Vec<_> = curr_ids
            .difference(&prev_ids)
            .map(|id| format!("{:x}", id))
            .collect();
        let removed: Vec<_> = prev_ids
            .difference(&curr_ids)
            .map(|id| format!("{:x}", id))
            .collect();
        let updated: Vec<_> = self
            .generations
            .iter()
            .filter(|(k, v)| prev.generations.get(*k).is_some_and(|pv| pv != *v))
            .map(|(k, _)| format!("{:x}", k))
            .collect();

        tracing::info!(
            "Snapshot (seq={}): {} instances, added={:?}, removed={:?}, updated={:?}",
            self.sequence,
            self.instances.len(),
            added,
            removed,
            updated
        );

        true
    }

279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
    /// Filter all instances in the snapshot by query
    pub fn filter(&self, query: &DiscoveryQuery) -> Vec<DiscoveryInstance> {
        self.instances
            .values()
            .flat_map(|metadata| metadata.filter(query))
            .collect()
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::component::{Instance, TransportType};

    #[test]
    fn test_metadata_serde() {
        let mut metadata = DiscoveryMetadata::new();

        // Add an endpoint
        let instance = DiscoveryInstance::Endpoint(Instance {
            namespace: "test".to_string(),
            component: "comp1".to_string(),
            endpoint: "ep1".to_string(),
            instance_id: 123,
303
            transport: TransportType::Nats("nats://localhost:4222".to_string()),
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
        });

        metadata.register_endpoint(instance).unwrap();

        // Serialize
        let json = serde_json::to_string(&metadata).unwrap();

        // Deserialize
        let deserialized: DiscoveryMetadata = serde_json::from_str(&json).unwrap();

        assert_eq!(deserialized.endpoints.len(), 1);
        assert_eq!(deserialized.model_cards.len(), 0);
    }

    #[tokio::test]
    async fn test_concurrent_registration() {
        use tokio::sync::RwLock;

        let metadata = Arc::new(RwLock::new(DiscoveryMetadata::new()));

        // Spawn multiple tasks registering concurrently
        let handles: Vec<_> = (0..10)
            .map(|i| {
                let metadata = metadata.clone();
                tokio::spawn(async move {
                    let mut meta = metadata.write().await;
                    let instance = DiscoveryInstance::Endpoint(Instance {
                        namespace: "test".to_string(),
                        component: "comp1".to_string(),
                        endpoint: format!("ep{}", i),
                        instance_id: i,
335
                        transport: TransportType::Nats("nats://localhost:4222".to_string()),
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
                    });
                    meta.register_endpoint(instance).unwrap();
                })
            })
            .collect();

        // Wait for all to complete
        for handle in handles {
            handle.await.unwrap();
        }

        // Verify all registrations succeeded
        let meta = metadata.read().await;
        assert_eq!(meta.endpoints.len(), 10);
    }

    #[tokio::test]
    async fn test_metadata_accessors() {
        let mut metadata = DiscoveryMetadata::new();

        // Register endpoints
        for i in 0..3 {
            let instance = DiscoveryInstance::Endpoint(Instance {
                namespace: "test".to_string(),
                component: "comp1".to_string(),
                endpoint: format!("ep{}", i),
                instance_id: i,
363
                transport: TransportType::Nats("nats://localhost:4222".to_string()),
364
365
366
367
368
369
370
371
372
373
374
375
            });
            metadata.register_endpoint(instance).unwrap();
        }

        // Register model cards
        for i in 0..2 {
            let instance = DiscoveryInstance::Model {
                namespace: "test".to_string(),
                component: "comp1".to_string(),
                endpoint: format!("ep{}", i),
                instance_id: i,
                card_json: serde_json::json!({"model": "test"}),
376
                model_suffix: None,
377
378
379
380
381
382
383
384
385
            };
            metadata.register_model_card(instance).unwrap();
        }

        assert_eq!(metadata.get_all_endpoints().len(), 3);
        assert_eq!(metadata.get_all_model_cards().len(), 2);
        assert_eq!(metadata.get_all().len(), 5);
    }
}