feat(router): capture latency + tokens on ROUTER ChatTurnEvent
ChatTurnEvent gains nullable latencyMs and tokensUsed fields (defaults preserve backward-compat; legacy JSON without them deserializes cleanly). emitChatTurn accepts optional metrics; the ROUTER emit site passes inferenceResponse values through; the USER emit site leaves both null.
This commit is contained in:
@@ -52,9 +52,11 @@ import com.correx.testing.fixtures.inference.MockTokenizer
|
||||
import kotlinx.coroutines.flow.Flow
|
||||
import kotlinx.coroutines.runBlocking
|
||||
import kotlinx.datetime.Instant
|
||||
import kotlinx.serialization.decodeFromString
|
||||
import org.junit.jupiter.api.Assertions.assertEquals
|
||||
import org.junit.jupiter.api.Assertions.assertFalse
|
||||
import org.junit.jupiter.api.Assertions.assertNotNull
|
||||
import org.junit.jupiter.api.Assertions.assertNull
|
||||
import org.junit.jupiter.api.Assertions.assertTrue
|
||||
import org.junit.jupiter.api.Test
|
||||
|
||||
@@ -544,6 +546,75 @@ class RouterFacadeTest {
|
||||
assertEquals(2, hits.size)
|
||||
}
|
||||
|
||||
// --------------------------------------------------------------------------
|
||||
// Task 2.1 — latency + token metrics on ROUTER ChatTurnEvent
|
||||
// --------------------------------------------------------------------------
|
||||
|
||||
@Test
|
||||
fun `ROUTER ChatTurnEvent carries latencyMs and tokensUsed from inference and USER turn has both null`(): Unit = runBlocking {
|
||||
val capturedRequests = mutableListOf<InferenceRequest>()
|
||||
val knownLatencyMs = 42L
|
||||
val knownTokenUsage = TokenUsage(promptTokens = 10, completionTokens = 20)
|
||||
val metricsProvider = object : InferenceProvider {
|
||||
override val id = ProviderId("mock")
|
||||
override val name = "Mock"
|
||||
override val tokenizer = MockTokenizer()
|
||||
override suspend fun infer(request: InferenceRequest): InferenceResponse {
|
||||
capturedRequests.add(request)
|
||||
return InferenceResponse(
|
||||
requestId = request.requestId,
|
||||
text = "metrics response",
|
||||
finishReason = FinishReason.Stop,
|
||||
tokensUsed = knownTokenUsage,
|
||||
latencyMs = knownLatencyMs,
|
||||
)
|
||||
}
|
||||
override suspend fun healthCheck(): ProviderHealth = ProviderHealth.Healthy
|
||||
override fun capabilities(): Set<CapabilityScore> = setOf(CapabilityScore(ModelCapability.General, 1.0))
|
||||
}
|
||||
val eventStore = mockEventStore()
|
||||
val facade = DefaultRouterFacade(
|
||||
routerRepository = object : RouterRepository {
|
||||
override suspend fun getRouterState(sessionId: SessionId): RouterState = RouterState()
|
||||
},
|
||||
routerContextBuilder = object : RouterContextBuilder {
|
||||
override suspend fun build(state: RouterState, budget: TokenBudget): ContextPack = emptyContextPack()
|
||||
},
|
||||
inferenceRouter = object : InferenceRouter {
|
||||
override suspend fun route(stageId: StageId, requiredCapabilities: Set<ModelCapability>): InferenceProvider =
|
||||
metricsProvider
|
||||
},
|
||||
eventStore = eventStore,
|
||||
config = RouterConfig(tokenBudget = TokenBudget(limit = 5000)),
|
||||
embedder = NoopEmbedder(dimension = 8),
|
||||
l3MemoryStore = InMemoryL3MemoryStore(),
|
||||
)
|
||||
|
||||
facade.onUserInput(sessionId = SessionId("metrics-session"), input = "hello metrics")
|
||||
|
||||
val chatEvents = eventStore.appendedEvents.map { it.payload }.filterIsInstance<ChatTurnEvent>()
|
||||
assertEquals(2, chatEvents.size)
|
||||
val userEvent = chatEvents.first { it.role == ChatTurnRole.USER }
|
||||
val routerEvent = chatEvents.first { it.role == ChatTurnRole.ROUTER }
|
||||
|
||||
assertNull(userEvent.latencyMs, "USER ChatTurnEvent must have null latencyMs")
|
||||
assertNull(userEvent.tokensUsed, "USER ChatTurnEvent must have null tokensUsed")
|
||||
|
||||
assertEquals(knownLatencyMs, routerEvent.latencyMs)
|
||||
assertNotNull(routerEvent.tokensUsed)
|
||||
assertEquals(30, routerEvent.tokensUsed!!.totalTokens)
|
||||
}
|
||||
|
||||
@Test
|
||||
fun `ChatTurnEvent backward-compat - legacy JSON without latencyMs and tokensUsed deserializes with both null`() {
|
||||
val legacyJson = """{"sessionId":"s1","turnId":"t1","role":"USER","content":"hello","timestampMs":1000}"""
|
||||
val event = com.correx.core.events.serialization.eventJson.decodeFromString<ChatTurnEvent>(legacyJson)
|
||||
assertEquals(SessionId("s1"), event.sessionId)
|
||||
assertEquals("hello", event.content)
|
||||
assertNull(event.latencyMs)
|
||||
assertNull(event.tokensUsed)
|
||||
}
|
||||
|
||||
// --------------------------------------------------------------------------
|
||||
// Helpers
|
||||
// --------------------------------------------------------------------------
|
||||
|
||||
Reference in New Issue
Block a user