[ci skip] feat: Add support for correlated failures

3 files changed, 28 insertions, 28 deletions

diff --git a/opendc/opendc-compute/src/main/kotlin/com/atlarge/opendc/compute/metal/driver/SimpleBareMetalDriver.kt b/opendc/opendc-compute/src/main/kotlin/com/atlarge/opendc/compute/metal/driver/SimpleBareMetalDriver.kt
index 4b9a03a6..a8f3d781 100644
--- a/opendc/opendc-compute/src/main/kotlin/com/atlarge/opendc/compute/metal/driver/SimpleBareMetalDriver.kt
+++ b/opendc/opendc-compute/src/main/kotlin/com/atlarge/opendc/compute/metal/driver/SimpleBareMetalDriver.kt
@@ -45,6 +45,7 @@ import com.atlarge.opendc.core.power.PowerModel
 import kotlinx.coroutines.CancellationException
 import kotlinx.coroutines.CoroutineScope
 import kotlinx.coroutines.Job
+import kotlinx.coroutines.cancel
 import kotlinx.coroutines.delay
 import kotlinx.coroutines.flow.Flow
 import kotlinx.coroutines.launch
@@ -82,22 +83,20 @@ public class SimpleBareMetalDriver(
      * The machine state.
      */
     private var node: Node = Node(uid, name, mapOf("driver" to this), NodeState.SHUTOFF, EmptyImage, null)
-        set(value) {
-            if (field.state != value.state) {
-                domain.launch {
-                    monitor.onUpdate(value, field.state)
-                }
-            }
 
-            if (field.server != null && value.server != null && field.server!!.state != value.server.state) {
-                domain.launch {
-                    monitor.onUpdate(value.server, field.server!!.state)
-                }
-            }
+    private suspend fun setNode(value: Node) {
+        val field = node
+        if (field.state != value.state) {
+            monitor.onUpdate(value, field.state)
+        }
 
-            field = value
+        if (field.server != null && value.server != null && field.server.state != value.server.state) {
+            monitor.onUpdate(value.server, field.server.state)
         }
 
+        node = value
+    }
+
     /**
      * The flavor that corresponds to this machine.
      */
@@ -137,7 +136,7 @@ public class SimpleBareMetalDriver(
         )
 
         server.serviceRegistry[BareMetalDriver.Key] = this@SimpleBareMetalDriver
-        node = node.copy(state = NodeState.BOOT, server = server)
+        setNode(node.copy(state = NodeState.BOOT, server = server))
         serverContext = BareMetalServerContext()
         return@withContext node
     }
@@ -151,7 +150,7 @@ public class SimpleBareMetalDriver(
         serverContext!!.cancel(fail = false)
         serverContext = null
 
-        node = node.copy(state = NodeState.SHUTOFF, server = null)
+        setNode(node.copy(state = NodeState.SHUTOFF, server = null))
         return@withContext node
     }
 
@@ -161,7 +160,7 @@ public class SimpleBareMetalDriver(
     }
 
     override suspend fun setImage(image: Image): Node = withContext(domain.coroutineContext) {
-        node = node.copy(image = image)
+        setNode(node.copy(image = image))
         return@withContext node
     }
 
@@ -190,9 +189,9 @@ public class SimpleBareMetalDriver(
          */
         suspend fun cancel(fail: Boolean) {
             if (fail)
-                job.cancel(ShutdownException(cause = Exception("Random failure")))
+                domain.cancel(ShutdownException(cause = Exception("Random failure")))
             else
-                job.cancel(ShutdownException())
+                domain.cancel(ShutdownException())
             job.join()
         }
 
@@ -200,7 +199,7 @@ public class SimpleBareMetalDriver(
             assert(!finalized) { "Machine is already finalized" }
 
             val server = server.copy(state = ServerState.ACTIVE)
-            node = node.copy(state = NodeState.ACTIVE, server = server)
+            setNode(node.copy(state = NodeState.ACTIVE, server = server))
         }
 
         override suspend fun exit(cause: Throwable?) {
@@ -217,7 +216,7 @@ public class SimpleBareMetalDriver(
                 else
                     NodeState.ERROR
             val server = server.copy(state = serverState)
-            node = node.copy(state = nodeState, server = server)
+            setNode(node.copy(state = nodeState, server = server))
         }
 
         private var flush: Job? = null
diff --git a/opendc/opendc-compute/src/main/kotlin/com/atlarge/opendc/compute/virt/service/SimpleVirtProvisioningService.kt b/opendc/opendc-compute/src/main/kotlin/com/atlarge/opendc/compute/virt/service/SimpleVirtProvisioningService.kt
index 0f38eca1..a50292a7 100644
--- a/opendc/opendc-compute/src/main/kotlin/com/atlarge/opendc/compute/virt/service/SimpleVirtProvisioningService.kt
+++ b/opendc/opendc-compute/src/main/kotlin/com/atlarge/opendc/compute/virt/service/SimpleVirtProvisioningService.kt
@@ -14,7 +14,6 @@ import com.atlarge.opendc.compute.virt.driver.hypervisor.InsufficientMemoryOnSer
 import com.atlarge.opendc.compute.virt.monitor.HypervisorMonitor
 import com.atlarge.opendc.compute.virt.service.allocation.AllocationPolicy
 import kotlinx.coroutines.launch
-import kotlinx.coroutines.yield
 
 class SimpleVirtProvisioningService(
     public override val allocationPolicy: AllocationPolicy,
@@ -56,13 +55,6 @@ class SimpleVirtProvisioningService(
                     server.flavor.memorySize
                 )
                 hypervisors[server] = hvView
-                yield()
-                server.serviceRegistry[VirtDriver.Key].addMonitor(object : VirtDriverMonitor {
-                    override suspend fun onUpdate(numberOfActiveServers: Int, availableMemory: Long) {
-                        hvView.numberOfActiveServers = numberOfActiveServers
-                        hvView.availableMemory = availableMemory
-                    }
-                })
             }
         }
     }
@@ -100,10 +92,19 @@ class SimpleVirtProvisioningService(
     }
 
     override suspend fun onUpdate(server: Server, previousState: ServerState) {
+        println("${server.uid} ${server.state} ${hypervisors[server]}")
         when (server.state) {
             ServerState.ACTIVE -> {
                 val hv = hypervisors[server] ?: return
                 availableHypervisors += hv
+
+                server.serviceRegistry[VirtDriver.Key].addMonitor(object : VirtDriverMonitor {
+                    override suspend fun onUpdate(numberOfActiveServers: Int, availableMemory: Long) {
+                        hv.numberOfActiveServers = numberOfActiveServers
+                        hv.availableMemory = availableMemory
+                    }
+                })
+
                 requestCycle()
             }
             ServerState.SHUTOFF, ServerState.ERROR -> {
diff --git a/opendc/opendc-compute/src/test/kotlin/com/atlarge/opendc/compute/metal/service/SimpleProvisioningServiceTest.kt b/opendc/opendc-compute/src/test/kotlin/com/atlarge/opendc/compute/metal/service/SimpleProvisioningServiceTest.kt
index a837130d..ef19427e 100644
--- a/opendc/opendc-compute/src/test/kotlin/com/atlarge/opendc/compute/metal/service/SimpleProvisioningServiceTest.kt
+++ b/opendc/opendc-compute/src/test/kotlin/com/atlarge/opendc/compute/metal/service/SimpleProvisioningServiceTest.kt
@@ -63,7 +63,7 @@ internal class SimpleProvisioningServiceTest {
 
             val cpuNode = ProcessingNode("Intel", "Xeon", "amd64", 4)
             val cpus = List(4) { ProcessingUnit(cpuNode, it, 2400.0) }
-            val driver = SimpleBareMetalDriver(dom, UUID.randomUUID(), "test", cpus, emptyList())
+            val driver = SimpleBareMetalDriver(dom.newDomain(), UUID.randomUUID(), "test", cpus, emptyList())
 
             val provisioner = SimpleProvisioningService(dom)
             provisioner.create(driver)