realizar 0.8.4

Pure Rust ML inference engine built from scratch - model serving for GGUF and safetensors
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

//! Part 2: Page allocation edge cases and memory management tests
//!
//! Focus areas:
//! 1. Page allocation edge cases
//! 2. Memory management

#[cfg(test)]
mod tests {
    use crate::paged_kv::*;

    // =========================================================================
    // Page Allocation Edge Cases
    // =========================================================================

    #[test]
    fn test_alloc_exact_page_boundary() {
        // Allocate exactly block_size tokens - should use exactly 1 page
        let mut cache = PagedKvCache::new(10, 16, 8, 64);
        let seq = cache.allocate_sequence(16).expect("alloc");
        assert_eq!(cache.free_page_count(), 9);
        assert_eq!(cache.stats().pages_allocated, 1);
        cache.free_sequence(seq);
    }

    #[test]
    fn test_alloc_one_over_boundary() {
        // Allocate block_size + 1 tokens - should use 2 pages
        let mut cache = PagedKvCache::new(10, 16, 8, 64);
        let seq = cache.allocate_sequence(17).expect("alloc");
        assert_eq!(cache.free_page_count(), 8);
        assert_eq!(cache.stats().pages_allocated, 2);
        cache.free_sequence(seq);
    }

    #[test]
    fn test_alloc_all_pages_then_free() {
        let mut cache = PagedKvCache::new(5, 16, 8, 64);

        // Allocate all pages
        let seqs: Vec<SeqId> = (0..5)
            .map(|_| cache.allocate_sequence(16).expect("alloc"))
            .collect();

        assert_eq!(cache.free_page_count(), 0);
        assert_eq!(cache.stats().pages_allocated, 5);

        // Try to allocate more - should fail
        let result = cache.allocate_sequence(1);
        assert!(matches!(result, Err(PagedCacheError::OutOfMemory { .. })));

        // Free all and reallocate
        for seq in seqs {
            cache.free_sequence(seq);
        }

        assert_eq!(cache.free_page_count(), 5);
        let _new_seq = cache.allocate_sequence(80).expect("reallocate all 5");
        assert_eq!(cache.free_page_count(), 0);
    }

    #[test]
    fn test_alloc_single_token() {
        let mut cache = PagedKvCache::new(10, 16, 8, 64);
        let seq = cache.allocate_sequence(1).expect("alloc 1 token");
        assert_eq!(cache.free_page_count(), 9);
        cache.free_sequence(seq);
        assert_eq!(cache.free_page_count(), 10);
    }

    #[test]
    fn test_alloc_large_sequence_spanning_many_pages() {
        let mut cache = PagedKvCache::new(100, 16, 8, 64);
        let seq = cache.allocate_sequence(250).expect("alloc 250 tokens");
        // 250 / 16 = 15.625 -> 16 pages
        assert_eq!(cache.free_page_count(), 84);
        assert_eq!(cache.stats().pages_allocated, 16);
        cache.free_sequence(seq);
        assert_eq!(cache.free_page_count(), 100);
    }

    // =========================================================================
    // Memory Management Tests
    // =========================================================================

    #[test]
    fn test_memory_usage_increases_with_allocation() {
        let mut cache = PagedKvCache::new(100, 16, 8, 64);

        let initial = cache.memory_usage();
        assert_eq!(initial, 0);

        let _seq1 = cache.allocate_sequence(16).expect("s1");
        let usage1 = cache.memory_usage();
        assert!(usage1 > 0);

        let _seq2 = cache.allocate_sequence(32).expect("s2");
        let usage2 = cache.memory_usage();
        assert_eq!(usage2, usage1 * 3); // 1 + 2 = 3 pages total
    }

    #[test]
    fn test_utilization_percentage() {
        let mut cache = PagedKvCache::new(10, 16, 8, 64);

        assert_eq!(cache.utilization(), 0.0);

        let _seq1 = cache.allocate_sequence(16).expect("s1"); // 1 page
        assert!((cache.utilization() - 10.0).abs() < 0.01);

        let _seq2 = cache.allocate_sequence(32).expect("s2"); // 2 pages
        assert!((cache.utilization() - 30.0).abs() < 0.01);

        let _seq3 = cache.allocate_sequence(112).expect("s3"); // 7 pages = all 10
        assert!((cache.utilization() - 100.0).abs() < 0.01);
    }

    #[test]
    fn test_quantized_memory_reduction_q8() {
        let fp32_cache = QuantizedPagedKvCache::new(10, 16, 8, 64, KvQuantType::FP32);
        let q8_cache = QuantizedPagedKvCache::new(10, 16, 8, 64, KvQuantType::Q8);

        // Compare page memory bytes
        let fp32_page = QuantizedKvPage::new(PageId::new(0), KvQuantType::FP32, 16, 8, 64);
        let q8_page = QuantizedKvPage::new(PageId::new(0), KvQuantType::Q8, 16, 8, 64);

        assert!(q8_page.memory_bytes() < fp32_page.memory_bytes());
        assert_eq!(fp32_cache.quant_type(), KvQuantType::FP32);
        assert_eq!(q8_cache.quant_type(), KvQuantType::Q8);
    }

    #[test]
    fn test_quantized_memory_reduction_q4() {
        let q8_page = QuantizedKvPage::new(PageId::new(0), KvQuantType::Q8, 16, 8, 64);
        let q4_page = QuantizedKvPage::new(PageId::new(0), KvQuantType::Q4, 16, 8, 64);

        // Q4 should use less memory than Q8
        assert!(q4_page.memory_bytes() < q8_page.memory_bytes());
    }

    #[test]
    fn test_fp32_equivalent_memory_calculation() {
        let mut cache = QuantizedPagedKvCache::new(100, 16, 8, 64, KvQuantType::Q8);
        let _seq = cache.allocate_sequence(32).expect("alloc"); // 2 pages

        let fp32_equiv = cache.fp32_equivalent_memory();
        let actual = cache.memory_usage();

        // Q8 actual should be less than FP32 equivalent
        assert!(actual < fp32_equiv);
    }

    #[test]
    fn test_memory_savings_empty_cache() {
        let cache = QuantizedPagedKvCache::new(100, 16, 8, 64, KvQuantType::Q8);
        // No pages used, returns 1.0 (no savings)
        assert_eq!(cache.memory_savings(), 1.0);
    }

    // =========================================================================
    // Error Handling Tests
    // =========================================================================

    #[test]
    fn test_oom_error_details() {
        let mut cache = PagedKvCache::new(2, 16, 8, 64);
        let _seq = cache.allocate_sequence(16).expect("first alloc");

        // Try to allocate more than available
        let result = cache.allocate_sequence(32); // needs 2 pages, only 1 free
        match result {
            Err(PagedCacheError::OutOfMemory { needed, available }) => {
                assert_eq!(needed, 2);
                assert_eq!(available, 1);
            },
            _ => panic!("Expected OutOfMemory error"),
        }
    }

    #[test]
    fn test_sequence_not_found_in_get_tokens() {
        let cache = PagedKvCache::new(10, 16, 8, 64);
        let fake = SeqId::new();

        let result = cache.get_sequence_tokens(fake);
        match result {
            Err(PagedCacheError::SequenceNotFound(id)) => {
                assert_eq!(id, fake.value());
            },
            _ => panic!("Expected SequenceNotFound error"),
        }
    }

    #[test]
    fn test_invalid_page_access_deep_position() {
        let mut cache = PagedKvCache::new(10, 16, 8, 64);
        let seq = cache.allocate_sequence(16).expect("alloc"); // 1 page only

        // Try to access position beyond allocated pages
        let result = cache.get_page(seq, 1000);
        match result {
            Err(PagedCacheError::InvalidPageAccess { page_id, offset }) => {
                assert_eq!(page_id, 62); // 1000 / 16 = 62
                assert_eq!(offset, 1000);
            },
            _ => panic!("Expected InvalidPageAccess error"),
        }
    }

    #[test]
    fn test_cow_oom_during_get_page_mut() {
        let mut cache = PagedKvCache::new(2, 16, 8, 64);
        let parent = cache.allocate_sequence(16).expect("parent"); // 1 page
        cache.update_tokens(parent, 16).expect("update");

        // Fork to create shared page (uses COW)
        let child = cache.fork_sequence(parent).expect("fork");

        // Use remaining free page
        let _other = cache.allocate_sequence(16).expect("other"); // takes last free

        // Now try get_page_mut on child - COW needs a new page but none available
        let result = cache.get_page_mut(child, 0);
        assert!(matches!(result, Err(PagedCacheError::OutOfMemory { .. })));
    }

    #[test]
    fn test_extend_oom() {
        let mut cache = PagedKvCache::new(2, 16, 8, 64);
        let seq = cache.allocate_sequence(16).expect("alloc");
        cache.update_tokens(seq, 16).expect("update");

        // Use remaining page
        let _other = cache.allocate_sequence(16).expect("other");

        // Try to extend original - needs new page but none available
        let result = cache.extend(seq, 32);
        match result {
            Err(PagedCacheError::OutOfMemory { needed, available }) => {
                assert!(needed > 0);
                assert_eq!(available, 0);
            },
            _ => panic!("Expected OutOfMemory error"),
        }
    }

    #[test]
    fn test_quantized_cache_sequence_not_found() {
        let cache = QuantizedPagedKvCache::new(10, 16, 8, 64, KvQuantType::Q8);
        let fake = SeqId::new();

        let result = cache.get_page(fake, 0);
        assert!(matches!(result, Err(PagedCacheError::SequenceNotFound(_))));
    }

    #[test]
    fn test_quantized_cache_invalid_page_access() {
        let mut cache = QuantizedPagedKvCache::new(10, 16, 8, 64, KvQuantType::Q8);
        let seq = cache.allocate_sequence(16).expect("alloc"); // 1 page

        let result = cache.get_page(seq, 100); // beyond allocated
        assert!(matches!(
            result,
            Err(PagedCacheError::InvalidPageAccess { .. })
        ));
    }

    // =========================================================================
    // COW (Copy-on-Write) Edge Cases
    // =========================================================================

    #[test]
    fn test_cow_multiple_forks() {
        let mut cache = PagedKvCache::new(10, 16, 8, 64);
        let parent = cache.allocate_sequence(16).expect("parent");
        cache.update_tokens(parent, 8).expect("update");

        // Fork multiple times
        let child1 = cache.fork_sequence(parent).expect("fork1");
        let child2 = cache.fork_sequence(parent).expect("fork2");
        let child3 = cache.fork_sequence(child1).expect("fork3");

        assert_eq!(cache.stats().active_sequences, 4);
        assert_eq!(cache.stats().cow_operations, 3);

        // Each child has own SeqId
        assert_ne!(parent, child1);
        assert_ne!(parent, child2);
        assert_ne!(child1, child3);

        // Writing to child should trigger COW
        let _page = cache.get_page_mut(child2, 0).expect("get mut");
        assert_eq!(cache.stats().cow_operations, 4);
    }

    #[test]
    fn test_cow_ref_count_after_free() {
        let mut cache = PagedKvCache::new(10, 16, 8, 64);
        let parent = cache.allocate_sequence(16).expect("parent");
        cache.update_tokens(parent, 16).expect("update");

        let child = cache.fork_sequence(parent).expect("fork");

        // Free child first - parent's page should still be valid
        cache.free_sequence(child);
        assert_eq!(cache.stats().active_sequences, 1);

        // Parent still has valid page
        let page = cache.get_page(parent, 0).expect("get");
        assert_eq!(page.ref_count, 1); // Back to single owner
    }

    #[test]
    fn test_cow_preserves_data() {
        let mut cache = PagedKvCache::new(10, 16, 8, 64);
        let parent = cache.allocate_sequence(16).expect("parent");
        cache.update_tokens(parent, 16).expect("update");

        // Write data to parent
        {
            let page = cache.get_page_mut(parent, 0).expect("mut");
            page.keys[0] = 123.0;
            page.values[0] = 456.0;
        }

        // Fork
        let child = cache.fork_sequence(parent).expect("fork");

        // Modify child (triggers COW)
        {
            let page = cache.get_page_mut(child, 0).expect("mut child");
            page.keys[0] = 789.0;
        }

        // Parent should have original data
        let parent_page = cache.get_page(parent, 0).expect("get parent");
        assert_eq!(parent_page.keys[0], 123.0);
        assert_eq!(parent_page.values[0], 456.0);

        // Child should have modified data
        let child_page = cache.get_page(child, 0).expect("get child");
        assert_eq!(child_page.keys[0], 789.0);
    }

    // =========================================================================
    // Fragmentation & Defragmentation Edge Cases
    // =========================================================================

    #[test]
    fn test_fragmentation_with_zero_pages() {
        let cache = PagedKvCache::new(0, 16, 8, 64);
        let stats = cache.fragmentation_stats();
        assert_eq!(stats.holes, 0);
        assert_eq!(stats.fragmentation_ratio, 0.0);
    }

    #[test]
    fn test_should_defrag_low_free_ratio() {
        let mut cache = PagedKvCache::new(10, 16, 8, 64);

        // Fill 9 of 10 pages (90% full)
        let seq = cache.allocate_sequence(144).expect("alloc 144");
        cache.update_tokens(seq, 100).expect("update");

        // At 90% utilization, free_ratio = 0.1
        // No holes since single sequence, so shouldn't trigger
        let should = cache.should_defragment();
        assert!(!should);
    }

    #[test]
    fn test_compact_empty_page_list() {
        let mut cache = PagedKvCache::new(10, 16, 8, 64);
        // compact_sequence with nonexistent seq returns 0
        let fake = SeqId::new();
        let moved = cache.compact_sequence(fake);
        assert_eq!(moved, 0);
    }

    #[test]
    fn test_defrag_increments_stats_only_when_pages_moved() {
        let mut cache = PagedKvCache::new(10, 16, 8, 64);
        let _seq = cache.allocate_sequence(32).expect("alloc");

        // Already contiguous, no moves
        let moved = cache.defragment();
        assert_eq!(moved, 0);
        assert_eq!(cache.stats().defrag_operations, 0);
        assert_eq!(cache.stats().pages_moved, 0);
    }
}