merge: Expose metrics directly to user (#80)

This pull request adds the ability to access the metrics of resources modeled
by the OpenDC Compute, Workflow, FaaS, and TensorFlow services directly from
their corresponding interfaces. Previously, users would have to interact with
OpenTelemetry to obtain these values, which is complex and provides
significant overhead.

With this pull request, users can access the metrics of all cloud resources
modeled by OpenDC via methods such as `getSchedulerStats()`, etc.

** Breaking Changes **

- `ComputeService.hostCount` removed in favour of `ComputeService.hosts.size`

2 files changed, 123 insertions, 14 deletions

diff --git a/opendc-compute/opendc-compute-simulator/src/main/kotlin/org/opendc/compute/simulator/SimHost.kt b/opendc-compute/opendc-compute-simulator/src/main/kotlin/org/opendc/compute/simulator/SimHost.kt
index 4eb6392e..323ae4fe 100644
--- a/opendc-compute/opendc-compute-simulator/src/main/kotlin/org/opendc/compute/simulator/SimHost.kt
+++ b/opendc-compute/opendc-compute-simulator/src/main/kotlin/org/opendc/compute/simulator/SimHost.kt
@@ -29,11 +29,14 @@ import io.opentelemetry.api.metrics.MeterProvider
 import io.opentelemetry.api.metrics.ObservableDoubleMeasurement
 import io.opentelemetry.api.metrics.ObservableLongMeasurement
 import kotlinx.coroutines.*
-import mu.KotlinLogging
 import org.opendc.compute.api.Flavor
 import org.opendc.compute.api.Server
 import org.opendc.compute.api.ServerState
 import org.opendc.compute.service.driver.*
+import org.opendc.compute.service.driver.telemetry.GuestCpuStats
+import org.opendc.compute.service.driver.telemetry.GuestSystemStats
+import org.opendc.compute.service.driver.telemetry.HostCpuStats
+import org.opendc.compute.service.driver.telemetry.HostSystemStats
 import org.opendc.compute.simulator.internal.Guest
 import org.opendc.compute.simulator.internal.GuestListener
 import org.opendc.simulator.compute.*
@@ -49,6 +52,8 @@ import org.opendc.simulator.compute.power.PowerDriver
 import org.opendc.simulator.compute.power.SimplePowerDriver
 import org.opendc.simulator.compute.workload.SimWorkload
 import org.opendc.simulator.flow.FlowEngine
+import java.time.Duration
+import java.time.Instant
 import java.util.*
 import kotlin.coroutines.CoroutineContext
 
@@ -81,11 +86,6 @@ public class SimHost(
     private val clock = engine.clock
 
     /**
-     * The logger instance of this server.
-     */
-    private val logger = KotlinLogging.logger {}
-
-    /**
      * The [Meter] to track metrics of the simulated host.
      */
     private val meter = meterProvider.get("org.opendc.compute.simulator")
@@ -112,6 +112,9 @@ public class SimHost(
     private val guests = HashMap<Server, Guest>()
     private val _guests = mutableListOf<Guest>()
 
+    override val instances: Set<Server>
+        get() = guests.keys
+
     override val state: HostState
         get() = _state
     private var _state: HostState = HostState.DOWN
@@ -249,6 +252,68 @@ public class SimHost(
         machine.cancel()
     }
 
+    override fun getSystemStats(): HostSystemStats {
+        updateUptime()
+
+        var terminated = 0
+        var running = 0
+        var error = 0
+        var invalid = 0
+
+        val guests = _guests.listIterator()
+        for (guest in guests) {
+            when (guest.state) {
+                ServerState.TERMINATED -> terminated++
+                ServerState.RUNNING -> running++
+                ServerState.ERROR -> error++
+                ServerState.DELETED -> {
+                    // Remove guests that have been deleted
+                    this.guests.remove(guest.server)
+                    guests.remove()
+                }
+                else -> invalid++
+            }
+        }
+
+        return HostSystemStats(
+            Duration.ofMillis(_uptime),
+            Duration.ofMillis(_downtime),
+            Instant.ofEpochMilli(_bootTime),
+            machine.powerUsage,
+            machine.energyUsage,
+            terminated,
+            running,
+            error,
+            invalid
+        )
+    }
+
+    override fun getSystemStats(server: Server): GuestSystemStats {
+        val guest = requireNotNull(guests[server]) { "Unknown server ${server.uid} at host $uid" }
+        return guest.getSystemStats()
+    }
+
+    override fun getCpuStats(): HostCpuStats {
+        val counters = hypervisor.counters
+        counters.flush()
+
+        return HostCpuStats(
+            counters.cpuActiveTime / 1000L,
+            counters.cpuIdleTime / 1000L,
+            counters.cpuStealTime / 1000L,
+            counters.cpuLostTime / 1000L,
+            hypervisor.cpuCapacity,
+            hypervisor.cpuDemand,
+            hypervisor.cpuUsage,
+            hypervisor.cpuUsage / _cpuLimit
+        )
+    }
+
+    override fun getCpuStats(server: Server): GuestCpuStats {
+        val guest = requireNotNull(guests[server]) { "Unknown server ${server.uid} at host $uid" }
+        return guest.getCpuStats()
+    }
+
     override fun hashCode(): Int = uid.hashCode()
 
     override fun equals(other: Any?): Boolean {
@@ -417,13 +482,12 @@ public class SimHost(
      * Helper function to track the CPU time of a machine.
      */
     private fun collectCpuTime(result: ObservableLongMeasurement) {
-        val counters = hypervisor.counters
-        counters.flush()
+        val stats = getCpuStats()
 
-        result.record(counters.cpuActiveTime / 1000L, _activeState)
-        result.record(counters.cpuIdleTime / 1000L, _idleState)
-        result.record(counters.cpuStealTime / 1000L, _stealState)
-        result.record(counters.cpuLostTime / 1000L, _lostState)
+        result.record(stats.activeTime, _activeState)
+        result.record(stats.idleTime, _idleState)
+        result.record(stats.stealTime, _stealState)
+        result.record(stats.lostTime, _lostState)
 
         val guests = _guests
         for (i in guests.indices) {
@@ -450,7 +514,7 @@ public class SimHost(
 
         val guests = _guests
         for (i in guests.indices) {
-            guests[i].updateUptime(duration)
+            guests[i].updateUptime()
         }
     }
 
diff --git a/opendc-compute/opendc-compute-simulator/src/main/kotlin/org/opendc/compute/simulator/internal/Guest.kt b/opendc-compute/opendc-compute-simulator/src/main/kotlin/org/opendc/compute/simulator/internal/Guest.kt
index bb378ee3..0d4c550d 100644
--- a/opendc-compute/opendc-compute-simulator/src/main/kotlin/org/opendc/compute/simulator/internal/Guest.kt
+++ b/opendc-compute/opendc-compute-simulator/src/main/kotlin/org/opendc/compute/simulator/internal/Guest.kt
@@ -32,6 +32,8 @@ import kotlinx.coroutines.*
 import mu.KotlinLogging
 import org.opendc.compute.api.Server
 import org.opendc.compute.api.ServerState
+import org.opendc.compute.service.driver.telemetry.GuestCpuStats
+import org.opendc.compute.service.driver.telemetry.GuestSystemStats
 import org.opendc.compute.simulator.SimHost
 import org.opendc.compute.simulator.SimWorkloadMapper
 import org.opendc.simulator.compute.kernel.SimHypervisor
@@ -39,6 +41,8 @@ import org.opendc.simulator.compute.kernel.SimVirtualMachine
 import org.opendc.simulator.compute.runWorkload
 import org.opendc.simulator.compute.workload.SimWorkload
 import java.time.Clock
+import java.time.Duration
+import java.time.Instant
 import kotlin.coroutines.CoroutineContext
 
 /**
@@ -146,6 +150,37 @@ internal class Guest(
     }
 
     /**
+     * Obtain the system statistics of this guest.
+     */
+    fun getSystemStats(): GuestSystemStats {
+        updateUptime()
+
+        return GuestSystemStats(
+            Duration.ofMillis(_uptime),
+            Duration.ofMillis(_downtime),
+            Instant.ofEpochMilli(_bootTime)
+        )
+    }
+
+    /**
+     * Obtain the CPU statistics of this guest.
+     */
+    fun getCpuStats(): GuestCpuStats {
+        val counters = machine.counters
+        counters.flush()
+
+        return GuestCpuStats(
+            counters.cpuActiveTime / 1000L,
+            counters.cpuIdleTime / 1000L,
+            counters.cpuStealTime / 1000L,
+            counters.cpuLostTime / 1000L,
+            machine.cpuCapacity,
+            machine.cpuUsage,
+            machine.cpuUsage / _cpuLimit
+        )
+    }
+
+    /**
      * The [Job] representing the current active virtual machine instance or `null` if no virtual machine is active.
      */
     private var job: Job? = null
@@ -209,6 +244,8 @@ internal class Guest(
      * This method is invoked when the guest stopped.
      */
     private fun onStop(target: ServerState) {
+        updateUptime()
+
         state = target
         listener.onStop(this)
     }
@@ -224,10 +261,16 @@ internal class Guest(
         .put(STATE_KEY, "down")
         .build()
 
+    private var _lastReport = clock.millis()
+
     /**
      * Helper function to track the uptime and downtime of the guest.
      */
-    fun updateUptime(duration: Long) {
+    fun updateUptime() {
+        val now = clock.millis()
+        val duration = now - _lastReport
+        _lastReport = now
+
         if (state == ServerState.RUNNING) {
             _uptime += duration
         } else if (state == ServerState.ERROR) {
@@ -239,6 +282,8 @@ internal class Guest(
      * Helper function to track the uptime of the guest.
      */
     fun collectUptime(result: ObservableLongMeasurement) {
+        updateUptime()
+
         result.record(_uptime, _upState)
         result.record(_downtime, _downState)
     }