|
| 1 | + /* |
| 2 | +
|
| 3 | +Copyright 2025 The llm-d Authors. |
| 4 | +
|
| 5 | +
|
| 6 | +Licensed under the Apache License, Version 2.0 (the "License"); |
| 7 | +
|
| 8 | +you may not use this file except in compliance with the License. |
| 9 | +
|
| 10 | +You may obtain a copy of the License at |
| 11 | +
|
| 12 | +
|
| 13 | +http://www.apache.org/licenses/LICENSE-2.0 |
| 14 | +
|
| 15 | +
|
| 16 | +Unless required by applicable law or agreed to in writing, software |
| 17 | +
|
| 18 | +distributed under the License is distributed on an "AS IS" BASIS, |
| 19 | +
|
| 20 | +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 21 | +
|
| 22 | +See the License for the specific language governing permissions and |
| 23 | +
|
| 24 | +limitations under the License. |
| 25 | +
|
| 26 | +*/ |
| 27 | + |
| 28 | +package main |
| 29 | + |
| 30 | +import ( |
| 31 | + "context" |
| 32 | + "math/rand/v2" |
| 33 | + "testing" |
| 34 | + |
| 35 | + "github.com/alicebob/miniredis/v2" |
| 36 | + "github.com/llm-d/llm-d-kv-cache/pkg/kvcache/kvblock" |
| 37 | + "k8s.io/apimachinery/pkg/util/sets" |
| 38 | +) |
| 39 | + |
| 40 | +const ( |
| 41 | + modelName = "bert-base-uncased" |
| 42 | + // Default number of keys for benchmarking |
| 43 | + benchNumKeys = 10000 |
| 44 | +) |
| 45 | + |
| 46 | +// generateWorkloadKeys creates a slice of keys with random chunk hashes. |
| 47 | +func generateWorkloadKeys(numKeys int) []kvblock.Key { |
| 48 | + // Use a fixed seed to ensure the exact same keys are generated for all profiling sessions. |
| 49 | + randGen := rand.New(rand.NewPCG(42, 1024)) |
| 50 | + |
| 51 | + keys := make([]kvblock.Key, numKeys) |
| 52 | + for i := range numKeys { |
| 53 | + keys[i] = kvblock.Key{ |
| 54 | + ModelName: modelName, |
| 55 | + ChunkHash: randGen.Uint64(), |
| 56 | + } |
| 57 | + } |
| 58 | + return keys |
| 59 | +} |
| 60 | + |
| 61 | +// helper to initialize specific index types. |
| 62 | +// redisAddr is optional; only used if indexType is "redis". |
| 63 | +func getIndexConfig(indexType string, redisAddr string) *kvblock.IndexConfig { |
| 64 | + switch indexType { |
| 65 | + case "redis": |
| 66 | + cfg := kvblock.DefaultRedisIndexConfig() |
| 67 | + |
| 68 | + // TODO: Verify the field name for the address in RedisIndexConfig! |
| 69 | + // It might be 'Addr', 'Address', 'Endpoint', or 'Url'. |
| 70 | + // The error 'cfg.Addr undefined' means 'Addr' is wrong. |
| 71 | + // I am guessing 'Address' here. |
| 72 | + cfg.Address = redisAddr |
| 73 | + |
| 74 | + return &kvblock.IndexConfig{ |
| 75 | + RedisConfig: cfg, |
| 76 | + EnableMetrics: false, |
| 77 | + } |
| 78 | + case "cost": |
| 79 | + return &kvblock.IndexConfig{ |
| 80 | + CostAwareMemoryConfig: kvblock.DefaultCostAwareMemoryIndexConfig(), |
| 81 | + EnableMetrics: false, |
| 82 | + } |
| 83 | + case "memory": |
| 84 | + return kvblock.DefaultIndexConfig() |
| 85 | + default: |
| 86 | + return kvblock.DefaultIndexConfig() |
| 87 | + } |
| 88 | +} |
| 89 | + |
| 90 | +// setupMiniredis starts a purely in-memory redis instance. |
| 91 | +// Returns the instance and a cleanup function. |
| 92 | +func setupMiniredis(b *testing.B) (*miniredis.Miniredis, func()) { |
| 93 | + s, err := miniredis.Run() |
| 94 | + if err != nil { |
| 95 | + b.Fatalf("failed to start miniredis: %v", err) |
| 96 | + } |
| 97 | + return s, func() { s.Close() } |
| 98 | +} |
| 99 | + |
| 100 | +// benchmarkAdd measures the performance of Adding keys to the index. |
| 101 | +func benchmarkAdd(b *testing.B, indexType string) { |
| 102 | + ctx := context.Background() |
| 103 | + podEntries := []kvblock.PodEntry{{PodIdentifier: "pod1", DeviceTier: "gpu"}} |
| 104 | + keys := generateWorkloadKeys(benchNumKeys) |
| 105 | + |
| 106 | + var redisAddr string |
| 107 | + |
| 108 | + // Clean setup for Miniredis specifically |
| 109 | + if indexType == "redis" { |
| 110 | + mr, cleanup := setupMiniredis(b) |
| 111 | + defer cleanup() |
| 112 | + redisAddr = mr.Addr() |
| 113 | + } |
| 114 | + |
| 115 | + b.ResetTimer() |
| 116 | + |
| 117 | + for i := 0; i < b.N; i++ { |
| 118 | + b.StopTimer() |
| 119 | + // Create a fresh index client, but connect to the SAME background redis server |
| 120 | + cfg := getIndexConfig(indexType, redisAddr) |
| 121 | + index, err := kvblock.NewIndex(ctx, cfg) |
| 122 | + if err != nil { |
| 123 | + b.Fatalf("failed to create index: %v", err) |
| 124 | + } |
| 125 | + |
| 126 | + b.StartTimer() |
| 127 | + |
| 128 | + // Pass 'keys' for both engineKeys and requestKeys |
| 129 | + err = index.Add(ctx, keys, keys, podEntries) |
| 130 | + if err != nil { |
| 131 | + b.Fatalf("failed to add entries: %v", err) |
| 132 | + } |
| 133 | + } |
| 134 | +} |
| 135 | + |
| 136 | +// benchmarkLookup measures the performance of Looking up keys. |
| 137 | +func benchmarkLookup(b *testing.B, indexType string) { |
| 138 | + ctx := context.Background() |
| 139 | + podEntries := []kvblock.PodEntry{{PodIdentifier: "pod1", DeviceTier: "gpu"}} |
| 140 | + |
| 141 | + // Intentionally use an empty podIdentifierSet to return all pods during lookup, |
| 142 | + // as documented in the Index interface. |
| 143 | + podIdentifierSet := sets.Set[string]{} |
| 144 | + |
| 145 | + keys := generateWorkloadKeys(benchNumKeys) |
| 146 | + |
| 147 | + var redisAddr string |
| 148 | + if indexType == "redis" { |
| 149 | + mr, cleanup := setupMiniredis(b) |
| 150 | + defer cleanup() |
| 151 | + redisAddr = mr.Addr() |
| 152 | + } |
| 153 | + |
| 154 | + // Setup: Create index and populate it |
| 155 | + cfg := getIndexConfig(indexType, redisAddr) |
| 156 | + index, err := kvblock.NewIndex(ctx, cfg) |
| 157 | + if err != nil { |
| 158 | + b.Fatalf("failed to create index: %v", err) |
| 159 | + } |
| 160 | + |
| 161 | + if err := index.Add(ctx, keys, keys, podEntries); err != nil { |
| 162 | + b.Fatalf("failed to populate index: %v", err) |
| 163 | + } |
| 164 | + |
| 165 | + b.ResetTimer() |
| 166 | + |
| 167 | + for i := 0; i < b.N; i++ { |
| 168 | + _, err = index.Lookup(ctx, keys, podIdentifierSet) |
| 169 | + if err != nil { |
| 170 | + b.Fatalf("failed to lookup entries: %v", err) |
| 171 | + } |
| 172 | + } |
| 173 | +} |
| 174 | + |
| 175 | +// --- Benchmark Entry Points --- |
| 176 | + |
| 177 | +func BenchmarkInMemory_Add(b *testing.B) { |
| 178 | + benchmarkAdd(b, "memory") |
| 179 | +} |
| 180 | + |
| 181 | +func BenchmarkInMemory_Lookup(b *testing.B) { |
| 182 | + benchmarkLookup(b, "memory") |
| 183 | +} |
| 184 | + |
| 185 | +func BenchmarkRedis_Add(b *testing.B) { |
| 186 | + benchmarkAdd(b, "redis") |
| 187 | +} |
| 188 | + |
| 189 | +func BenchmarkRedis_Lookup(b *testing.B) { |
| 190 | + benchmarkLookup(b, "redis") |
| 191 | +} |
| 192 | + |
| 193 | +func BenchmarkCostAware_Add(b *testing.B) { |
| 194 | + benchmarkAdd(b, "cost") |
| 195 | +} |
| 196 | + |
| 197 | +func BenchmarkCostAware_Lookup(b *testing.B) { |
| 198 | + benchmarkLookup(b, "cost") |
| 199 | +} |
0 commit comments