companion-runtime-manager-performance.test.mjs
file-level
1
files
1
commits
0
hotspots
0
π§ dead
0
π₯ blast risk
| 1 | /** |
| 2 | * Tier 6 β PERFORMANCE: lib/companion-runtime-manager.mjs |
| 3 | * |
| 4 | * Latency bounds for all decision functions. The runtime manager sits on the hot path |
| 5 | * for every inference request. Decisions must complete in sub-millisecond time to avoid |
| 6 | * adding measurable overhead to inference throughput. |
| 7 | * |
| 8 | * Bounds (conservative; well within what node:crypto + pure-JS can deliver): |
| 9 | * - Per-decision calls (evaluateRuntimeRequest, evaluateAdmission, etc.): mean < 0.1ms |
| 10 | * - 10k evaluateRuntimeRequest calls: total < 500ms |
| 11 | * - 10k lifecycle transitions: total < 200ms |
| 12 | * - Integrity accumulator: 100KB with 1-byte chunks: total < 500ms |
| 13 | * |
| 14 | * Reference: docs/COMPANION-APP-PHASE-4-RUNTIME-MANAGER.md Β§4 (backpressure). |
| 15 | */ |
| 16 | |
| 17 | import { describe, it } from 'node:test'; |
| 18 | import assert from 'node:assert/strict'; |
| 19 | import crypto from 'node:crypto'; |
| 20 | |
| 21 | import { |
| 22 | LIFECYCLE_STATES, |
| 23 | LIFECYCLE_EVENTS, |
| 24 | createLifecycleState, |
| 25 | transitionLifecycle, |
| 26 | canServeInference, |
| 27 | createAdmissionState, |
| 28 | evaluateAdmission, |
| 29 | recordInFlight, |
| 30 | recordCompletion, |
| 31 | createResourceLimits, |
| 32 | evaluateResourceLimits, |
| 33 | evaluateRuntimeRequest, |
| 34 | createIntegrityAccumulator, |
| 35 | verifyModelBytes, |
| 36 | validateSourceUrl, |
| 37 | } from '../lib/companion-runtime-manager.mjs'; |
| 38 | |
| 39 | function makeDigest(data) { |
| 40 | return crypto.createHash('sha256').update(data).digest('hex'); |
| 41 | } |
| 42 | |
| 43 | const ALLOWED_URLS = ['https://models.example.com/']; |
| 44 | const VALID_URL = 'https://models.example.com/model.gguf'; |
| 45 | const READY = { state: LIFECYCLE_STATES.READY }; |
| 46 | const VALID_LIMITS = createResourceLimits({ maxRamBytes: 8e9, maxVramBytes: 4e9, maxCpuPercent: 80 }); |
| 47 | const VALID_OBS = { ramBytes: 1e9, vramBytes: 0.5e9, cpuPercent: 10 }; |
| 48 | const VALID_ADMISSION = createAdmissionState({ maxInFlight: 100, queueBound: 200 }); |
| 49 | |
| 50 | // ββ evaluateRuntimeRequest ββββββββββββββββββββββββββββββββββββββββββββββββββββ |
| 51 | |
| 52 | describe('performance: evaluateRuntimeRequest', () => { |
| 53 | it('10k calls complete in < 500ms', () => { |
| 54 | const N = 10_000; |
| 55 | const start = performance.now(); |
| 56 | for (let i = 0; i < N; i++) { |
| 57 | evaluateRuntimeRequest({ |
| 58 | lifecycleState: READY, |
| 59 | admissionState: VALID_ADMISSION, |
| 60 | resourceObservation: VALID_OBS, |
| 61 | resourceLimits: VALID_LIMITS, |
| 62 | }); |
| 63 | } |
| 64 | const elapsed = performance.now() - start; |
| 65 | assert.ok(elapsed < 500, `10k evaluateRuntimeRequest took ${elapsed.toFixed(1)}ms, expected < 500ms`); |
| 66 | }); |
| 67 | |
| 68 | it('mean per-call < 0.05ms (50ΞΌs)', () => { |
| 69 | const N = 10_000; |
| 70 | const start = performance.now(); |
| 71 | for (let i = 0; i < N; i++) { |
| 72 | evaluateRuntimeRequest({ |
| 73 | lifecycleState: READY, |
| 74 | admissionState: VALID_ADMISSION, |
| 75 | resourceObservation: VALID_OBS, |
| 76 | resourceLimits: VALID_LIMITS, |
| 77 | }); |
| 78 | } |
| 79 | const mean = (performance.now() - start) / N; |
| 80 | assert.ok(mean < 0.05, `mean per-call ${mean.toFixed(4)}ms, expected < 0.05ms`); |
| 81 | }); |
| 82 | }); |
| 83 | |
| 84 | // ββ evaluateAdmission βββββββββββββββββββββββββββββββββββββββββββββββββββββββββ |
| 85 | |
| 86 | describe('performance: evaluateAdmission', () => { |
| 87 | it('50k calls complete in < 500ms', () => { |
| 88 | const N = 50_000; |
| 89 | const s = { ...VALID_ADMISSION, inFlight: 1 }; |
| 90 | const start = performance.now(); |
| 91 | for (let i = 0; i < N; i++) evaluateAdmission(s); |
| 92 | const elapsed = performance.now() - start; |
| 93 | assert.ok(elapsed < 500, `50k evaluateAdmission took ${elapsed.toFixed(1)}ms, expected < 500ms`); |
| 94 | }); |
| 95 | }); |
| 96 | |
| 97 | // ββ evaluateResourceLimits ββββββββββββββββββββββββββββββββββββββββββββββββββββ |
| 98 | |
| 99 | describe('performance: evaluateResourceLimits', () => { |
| 100 | it('50k calls complete in < 300ms', () => { |
| 101 | const N = 50_000; |
| 102 | const start = performance.now(); |
| 103 | for (let i = 0; i < N; i++) evaluateResourceLimits(VALID_OBS, VALID_LIMITS); |
| 104 | const elapsed = performance.now() - start; |
| 105 | assert.ok(elapsed < 300, `50k evaluateResourceLimits took ${elapsed.toFixed(1)}ms, expected < 300ms`); |
| 106 | }); |
| 107 | }); |
| 108 | |
| 109 | // ββ Lifecycle transitions ββββββββββββββββββββββββββββββββββββββββββββββββββββ |
| 110 | |
| 111 | describe('performance: lifecycle transitions', () => { |
| 112 | it('10k full round-trips (stoppedβstartingβreadyβdrainβstopped) < 200ms', () => { |
| 113 | const N = 10_000; |
| 114 | let lifecycle = createLifecycleState(); |
| 115 | const start = performance.now(); |
| 116 | for (let i = 0; i < N; i++) { |
| 117 | lifecycle = transitionLifecycle(lifecycle, LIFECYCLE_EVENTS.START).newState; |
| 118 | lifecycle = transitionLifecycle(lifecycle, LIFECYCLE_EVENTS.HEALTH_OK).newState; |
| 119 | lifecycle = transitionLifecycle(lifecycle, LIFECYCLE_EVENTS.DRAIN).newState; |
| 120 | lifecycle = transitionLifecycle(lifecycle, LIFECYCLE_EVENTS.STOPPED).newState; |
| 121 | } |
| 122 | const elapsed = performance.now() - start; |
| 123 | assert.ok(elapsed < 200, `10k lifecycle round-trips took ${elapsed.toFixed(1)}ms, expected < 200ms`); |
| 124 | }); |
| 125 | |
| 126 | it('canServeInference: 100k calls < 100ms', () => { |
| 127 | const N = 100_000; |
| 128 | const start = performance.now(); |
| 129 | for (let i = 0; i < N; i++) canServeInference(READY); |
| 130 | const elapsed = performance.now() - start; |
| 131 | assert.ok(elapsed < 100, `100k canServeInference took ${elapsed.toFixed(1)}ms, expected < 100ms`); |
| 132 | }); |
| 133 | }); |
| 134 | |
| 135 | // ββ Integrity accumulator ββββββββββββββββββββββββββββββββββββββββββββββββββββ |
| 136 | |
| 137 | describe('performance: integrity accumulator', () => { |
| 138 | it('100KB model data in 1-byte chunks completes in < 500ms', () => { |
| 139 | const data = crypto.randomBytes(100_000); |
| 140 | const digest = makeDigest(data); |
| 141 | |
| 142 | const acc = createIntegrityAccumulator({ |
| 143 | expectedDigest: digest, expectedSizeBytes: data.length, |
| 144 | sourceUrl: VALID_URL, allowedSourceUrls: ALLOWED_URLS, |
| 145 | }); |
| 146 | |
| 147 | const start = performance.now(); |
| 148 | for (let i = 0; i < data.length; i++) { |
| 149 | acc.update(data.subarray(i, i + 1)); |
| 150 | } |
| 151 | acc.finalize(); |
| 152 | const elapsed = performance.now() - start; |
| 153 | assert.ok(elapsed < 500, `100KB 1-byte-chunk accumulation took ${elapsed.toFixed(1)}ms, expected < 500ms`); |
| 154 | }); |
| 155 | |
| 156 | it('1MB model data in 4KB chunks completes in < 200ms', () => { |
| 157 | const CHUNK = 4096; |
| 158 | const data = crypto.randomBytes(1024 * 1024); |
| 159 | const digest = makeDigest(data); |
| 160 | |
| 161 | const acc = createIntegrityAccumulator({ |
| 162 | expectedDigest: digest, expectedSizeBytes: data.length, |
| 163 | sourceUrl: VALID_URL, allowedSourceUrls: ALLOWED_URLS, |
| 164 | }); |
| 165 | |
| 166 | const start = performance.now(); |
| 167 | for (let i = 0; i < data.length; i += CHUNK) { |
| 168 | acc.update(data.subarray(i, i + CHUNK)); |
| 169 | } |
| 170 | acc.finalize(); |
| 171 | const elapsed = performance.now() - start; |
| 172 | assert.ok(elapsed < 200, `1MB 4KB-chunk accumulation took ${elapsed.toFixed(1)}ms, expected < 200ms`); |
| 173 | }); |
| 174 | }); |
| 175 | |
| 176 | // ββ verifyModelBytes βββββββββββββββββββββββββββββββββββββββββββββββββββββββββ |
| 177 | |
| 178 | describe('performance: verifyModelBytes', () => { |
| 179 | it('1000 calls on a 1KB buffer complete in < 300ms', () => { |
| 180 | const data = crypto.randomBytes(1024); |
| 181 | const digest = makeDigest(data); |
| 182 | const N = 1000; |
| 183 | |
| 184 | const start = performance.now(); |
| 185 | for (let i = 0; i < N; i++) { |
| 186 | verifyModelBytes({ |
| 187 | fileData: data, expectedDigest: digest, expectedSizeBytes: data.length, |
| 188 | sourceUrl: VALID_URL, allowedSourceUrls: ALLOWED_URLS, |
| 189 | }); |
| 190 | } |
| 191 | const elapsed = performance.now() - start; |
| 192 | assert.ok(elapsed < 300, `1000 verifyModelBytes took ${elapsed.toFixed(1)}ms, expected < 300ms`); |
| 193 | }); |
| 194 | }); |
| 195 | |
| 196 | // ββ validateSourceUrl (hot path for spec validation) ββββββββββββββββββββββββ |
| 197 | |
| 198 | describe('performance: validateSourceUrl', () => { |
| 199 | it('100k calls complete in < 500ms', () => { |
| 200 | const N = 100_000; |
| 201 | const start = performance.now(); |
| 202 | for (let i = 0; i < N; i++) validateSourceUrl(VALID_URL, ALLOWED_URLS); |
| 203 | const elapsed = performance.now() - start; |
| 204 | assert.ok(elapsed < 500, `100k validateSourceUrl took ${elapsed.toFixed(1)}ms, expected < 500ms`); |
| 205 | }); |
| 206 | }); |
| 207 | |
| 208 | // ββ Admission record cycling ββββββββββββββββββββββββββββββββββββββββββββββββββ |
| 209 | |
| 210 | describe('performance: recordInFlight/recordCompletion cycling', () => { |
| 211 | it('10k recordInFlight + recordCompletion pairs < 100ms', () => { |
| 212 | const N = 10_000; |
| 213 | let s = createAdmissionState({ maxInFlight: N + 1, queueBound: N + 1 }); |
| 214 | const start = performance.now(); |
| 215 | for (let i = 0; i < N; i++) { |
| 216 | s = recordInFlight(s); |
| 217 | s = recordCompletion(s); |
| 218 | } |
| 219 | const elapsed = performance.now() - start; |
| 220 | assert.ok(elapsed < 100, `10k in-flight cycles took ${elapsed.toFixed(1)}ms, expected < 100ms`); |
| 221 | assert.equal(s.inFlight, 0); // net zero |
| 222 | }); |
| 223 | }); |