feat: Track VM failures in SC20 experiment

3 files changed, 15 insertions, 19 deletions

diff --git a/opendc/opendc-compute/src/test/kotlin/com/atlarge/opendc/compute/metal/driver/SimpleBareMetalDriverTest.kt b/opendc/opendc-compute/src/test/kotlin/com/atlarge/opendc/compute/metal/driver/SimpleBareMetalDriverTest.kt
index 1b7a47d5..166e93b8 100644
--- a/opendc/opendc-compute/src/test/kotlin/com/atlarge/opendc/compute/metal/driver/SimpleBareMetalDriverTest.kt
+++ b/opendc/opendc-compute/src/test/kotlin/com/atlarge/opendc/compute/metal/driver/SimpleBareMetalDriverTest.kt
@@ -34,8 +34,6 @@ import com.atlarge.opendc.compute.core.image.FlopsApplicationImage
 import com.atlarge.opendc.compute.metal.Node
 import com.atlarge.opendc.compute.metal.NodeState
 import com.atlarge.opendc.compute.metal.monitor.NodeMonitor
-import com.atlarge.opendc.core.failure.FaultInjector
-import com.atlarge.opendc.core.failure.UncorrelatedFaultInjector
 import kotlinx.coroutines.launch
 import kotlinx.coroutines.runBlocking
 import kotlinx.coroutines.withContext
@@ -78,9 +76,6 @@ internal class SimpleBareMetalDriverTest {
                 driver.setImage(image)
                 driver.start()
             }
-
-            val injector = UncorrelatedFaultInjector()
-            injector.enqueue(driver)
         }
 
         runBlocking {
diff --git a/opendc/opendc-experiments-sc20/src/main/kotlin/com/atlarge/opendc/experiments/sc20/Sc20HypervisorMonitor.kt b/opendc/opendc-experiments-sc20/src/main/kotlin/com/atlarge/opendc/experiments/sc20/Sc20Monitor.kt
index 9e8f0fa8..6ce9cefa 100644
--- a/opendc/opendc-experiments-sc20/src/main/kotlin/com/atlarge/opendc/experiments/sc20/Sc20HypervisorMonitor.kt
+++ b/opendc/opendc-experiments-sc20/src/main/kotlin/com/atlarge/opendc/experiments/sc20/Sc20Monitor.kt
@@ -1,6 +1,8 @@
 package com.atlarge.opendc.experiments.sc20
 
 import com.atlarge.opendc.compute.core.Server
+import com.atlarge.opendc.compute.core.ServerState
+import com.atlarge.opendc.compute.core.monitor.ServerMonitor
 import com.atlarge.opendc.compute.metal.driver.BareMetalDriver
 import com.atlarge.opendc.compute.virt.monitor.HypervisorMonitor
 import kotlinx.coroutines.flow.first
@@ -8,15 +10,22 @@ import java.io.BufferedWriter
 import java.io.Closeable
 import java.io.FileWriter
 
-class Sc20HypervisorMonitor(
+class Sc20Monitor(
     destination: String
-) : HypervisorMonitor, Closeable {
+) : HypervisorMonitor, ServerMonitor, Closeable {
     private val outputFile = BufferedWriter(FileWriter(destination))
+    private var failed: Int = 0
 
     init {
         outputFile.write("time,requestedBurst,grantedBurst,numberOfDeployedImages,server,hostUsage,powerDraw\n")
     }
 
+    override suspend fun onUpdate(server: Server, previousState: ServerState) {
+        if (server.state == ServerState.ERROR) {
+            failed++
+        }
+    }
+
     override suspend fun onSliceFinish(
         time: Long,
         requestedBurst: Long,
@@ -29,7 +38,7 @@ class Sc20HypervisorMonitor(
         val usage = driver.usage.first()
         val powerDraw = driver.powerDraw.first()
 
-        outputFile.write("$time,$requestedBurst,$grantedBurst,$numberOfDeployedImages,${hostServer.uid},$usage,$powerDraw")
+        outputFile.write("$time,$requestedBurst,$grantedBurst,$numberOfDeployedImages,${hostServer.uid},$usage,$powerDraw,$failed")
         outputFile.newLine()
     }
 
diff --git a/opendc/opendc-experiments-sc20/src/main/kotlin/com/atlarge/opendc/experiments/sc20/TestExperiment.kt b/opendc/opendc-experiments-sc20/src/main/kotlin/com/atlarge/opendc/experiments/sc20/TestExperiment.kt
index f0d3fc8d..e47438f0 100644
--- a/opendc/opendc-experiments-sc20/src/main/kotlin/com/atlarge/opendc/experiments/sc20/TestExperiment.kt
+++ b/opendc/opendc-experiments-sc20/src/main/kotlin/com/atlarge/opendc/experiments/sc20/TestExperiment.kt
@@ -27,9 +27,6 @@ package com.atlarge.opendc.experiments.sc20
 import com.atlarge.odcsim.SimulationEngineProvider
 import com.atlarge.odcsim.simulationContext
 import com.atlarge.opendc.compute.core.Flavor
-import com.atlarge.opendc.compute.core.Server
-import com.atlarge.opendc.compute.core.ServerState
-import com.atlarge.opendc.compute.core.monitor.ServerMonitor
 import com.atlarge.opendc.compute.metal.service.ProvisioningService
 import com.atlarge.opendc.compute.virt.service.SimpleVirtProvisioningService
 import com.atlarge.opendc.compute.virt.service.allocation.AvailableMemoryAllocationPolicy
@@ -85,12 +82,7 @@ class ExperimentParameters(parser: ArgParser) {
  */
 fun main(args: Array<String>) {
     ArgParser(args).parseInto(::ExperimentParameters).run {
-        val hypervisorMonitor = Sc20HypervisorMonitor(outputFile)
-        val monitor = object : ServerMonitor {
-            override suspend fun onUpdate(server: Server, previousState: ServerState) {
-                println(server)
-            }
-        }
+        val monitor = Sc20Monitor(outputFile)
 
         val provider = ServiceLoader.load(SimulationEngineProvider::class.java).first()
         val system = provider("test")
@@ -115,7 +107,7 @@ fun main(args: Array<String>) {
                 AvailableMemoryAllocationPolicy(),
                 simulationContext,
                 environment.platforms[0].zones[0].services[ProvisioningService.Key],
-                hypervisorMonitor
+                monitor
             )
 
             val reader = Sc20TraceReader(File(traceDirectory), performanceInterferenceModel, getSelectedVmList())
@@ -134,6 +126,6 @@ fun main(args: Array<String>) {
         }
 
         // Explicitly close the monitor to flush its buffer
-        hypervisorMonitor.close()
+        monitor.close()
     }
 }