feat: correx-managed model lifecycle slice 4 — resource telemetry

Vendor-agnostic ResourceProbe (commons): NvidiaResourceProbe reads nvidia-smi (injectable runner) for whole-device VRAM/util and /proc/<pid>/status for the managed llama-server RSS, both fail-soft to null; UnavailableProbe for non-GPU hosts / the static path. DefaultModelManager.currentPid() + LlamaProcess.pid expose the managed pid. ServerMessage.ResourceStatus (resource.status, NonEventMessage, all-nullable) pushed every 2.5s on the global stream plus one in the initial snapshot. Live gauge only — never event-sourced (feeds no core decision), so invariants #8/#9 hold. Main wires the NVIDIA probe on the managed path when nvidia-smi is present, else UnavailableProbe. Tests: csv parsing (single/multi/malformed), gpu+rss combine, fallbacks. Plan: docs/plans/2026-05-31-model-lifecycle-management.md (slice 4 of 5).
2026-06-01 12:36:38 +04:00
parent 7341ab578e
commit 7b1df95627
10 changed files with 273 additions and 0 deletions
@@ -0,0 +1,75 @@
+package com.correx.testing.contracts.model.inference
+
+import com.correx.infrastructure.inference.commons.NvidiaResourceProbe
+import com.correx.infrastructure.inference.commons.UnavailableProbe
+import org.junit.jupiter.api.Assertions.assertEquals
+import org.junit.jupiter.api.Assertions.assertNull
+import org.junit.jupiter.api.Test
+
+class NvidiaResourceProbeTest {
+
+    @Test
+    fun `parses a single GPU csv row`() {
+        val gpu = NvidiaResourceProbe.parseGpu("8192, 16384, 37")
+        assertEquals(8192L, gpu?.memoryUsedMb)
+        assertEquals(16384L, gpu?.memoryTotalMb)
+        assertEquals(37, gpu?.utilizationPct)
+    }
+
+    @Test
+    fun `parses the first row when multiple GPUs are present`() {
+        val gpu = NvidiaResourceProbe.parseGpu("8192, 16384, 37\n1024, 16384, 5")
+        assertEquals(8192L, gpu?.memoryUsedMb)
+        assertEquals(37, gpu?.utilizationPct)
+    }
+
+    @Test
+    fun `returns null on malformed csv`() {
+        assertNull(NvidiaResourceProbe.parseGpu("not,enough"))
+        assertNull(NvidiaResourceProbe.parseGpu("abc, def, ghi"))
+        assertNull(NvidiaResourceProbe.parseGpu(""))
+    }
+
+    @Test
+    fun `probe combines gpu csv and rss reader`() {
+        val probe = NvidiaResourceProbe(
+            pidSupplier = { 4242L },
+            nvidiaSmiRunner = { "8192, 16384, 37" },
+            rssReader = { pid -> if (pid == 4242L) 1_048_576L else null },
+        )
+        val snapshot = probe.probe()
+        assertEquals(8192L, snapshot.gpu?.memoryUsedMb)
+        assertEquals(1_048_576L, snapshot.processRssBytes)
+    }
+
+    @Test
+    fun `probe reports gpu unavailable when nvidia-smi yields null`() {
+        val probe = NvidiaResourceProbe(
+            pidSupplier = { 4242L },
+            nvidiaSmiRunner = { null },
+            rssReader = { 1_048_576L },
+        )
+        val snapshot = probe.probe()
+        assertNull(snapshot.gpu)
+        assertEquals(1_048_576L, snapshot.processRssBytes)
+    }
+
+    @Test
+    fun `probe reports rss null when no pid is resident`() {
+        val probe = NvidiaResourceProbe(
+            pidSupplier = { null },
+            nvidiaSmiRunner = { "8192, 16384, 37" },
+            rssReader = { error("must not be called when pid is null") },
+        )
+        val snapshot = probe.probe()
+        assertEquals(8192L, snapshot.gpu?.memoryUsedMb)
+        assertNull(snapshot.processRssBytes)
+    }
+
+    @Test
+    fun `unavailable probe is fully unavailable`() {
+        val snapshot = UnavailableProbe.probe()
+        assertNull(snapshot.gpu)
+        assertNull(snapshot.processRssBytes)
+    }
+}