ID-1276 Introduce Bard Service for sending metrics (#7434)

Co-authored-by: Tristan Garwood <[email protected]>
Co-authored-by: Adam Nichols <[email protected]>
Co-authored-by: Janet Gainer-Dewar <[email protected]>

Author: 4 people

backend/src/main/scala/cromwell/backend/BackendLifecycleActorFactory.scala

@@ -14,7 +14,7 @@ import wom.graph.CommandCallNode
 
 import scala.concurrent.ExecutionContext
 
-trait BackendLifecycleActorFactory {
+trait BackendLifecycleActorFactory extends PlatformSpecific {
 
   /**
     * Name of the backend.
@@ -166,7 +166,7 @@ trait BackendLifecycleActorFactory {
   /**
     * Allows Cromwell to self-identify which cloud it's running on for runtime attribute purposes
     */
-  def platform: Option[Platform] = None
+  override def platform: Option[Platform] = None
 }
 
 object BackendLifecycleActorFactory {

backend/src/main/scala/cromwell/backend/backend.scala

@@ -178,3 +178,18 @@ object Azure extends Platform {
 object Aws extends Platform {
   override def runtimeKey: String = "aws"
 }
+trait PlatformSpecific {
+  def platform: Option[Platform]
+
+}
+trait GcpPlatform extends PlatformSpecific {
+  override val platform: Option[Platform] = Option(Gcp)
+}
+
+trait AzurePlatform extends PlatformSpecific {
+  override val platform: Option[Platform] = Option(Azure)
+}
+
+trait AwsPlatform extends PlatformSpecific {
+  override val platform: Option[Platform] = Option(Aws)
+}

backend/src/main/scala/cromwell/backend/dummy/DummyAsyncExecutionActor.scala

@@ -31,6 +31,8 @@ class DummyAsyncExecutionActor(override val standardParams: StandardAsyncExecuti
     with StandardAsyncExecutionActor
     with CromwellInstrumentation {
 
+  override def platform: Option[Nothing] = None
+
   /** The type of the run info when a job is started. */
   override type StandardAsyncRunInfo = String
 

backend/src/main/scala/cromwell/backend/standard/StandardAsyncExecutionActor.scala

@@ -31,18 +31,23 @@ import cromwell.core.path.Path
 import cromwell.services.keyvalue.KeyValueServiceActor._
 import cromwell.services.keyvalue.KvClient
 import cromwell.services.metadata.CallMetadataKeys
+import cromwell.services.metrics.bard.BardEventing.BardEventRequest
+import cromwell.services.metrics.bard.model.TaskSummaryEvent
 import eu.timepit.refined.refineV
 import mouse.all._
 import net.ceedubs.ficus.Ficus._
 import org.apache.commons.lang3.StringUtils
 import org.apache.commons.lang3.exception.ExceptionUtils
 import shapeless.Coproduct
+import wdl4s.parser.MemoryUnit
 import wom.callable.{AdHocValue, CommandTaskDefinition, ContainerizedInputExpression}
 import wom.expression.WomExpression
 import wom.graph.LocalName
 import wom.values._
 import wom.{CommandSetupSideEffectFile, InstantiatedCommand, WomFileMapper}
 
+import java.time.OffsetDateTime
+import java.time.temporal.ChronoUnit
 import java.io.IOException
 import scala.concurrent._
 import scala.concurrent.duration._
@@ -70,6 +75,8 @@ case class DefaultStandardAsyncExecutionActorParams(
 // Override to `false` when we need the script to set an environment variable in the parent shell.
 case class ScriptPreambleData(bashString: String, executeInSubshell: Boolean = true)
 
+case class StartAndEndTimes(jobStart: OffsetDateTime, cpuStart: Option[OffsetDateTime], jobEnd: OffsetDateTime)
+
 /**
   * An extension of the generic AsyncBackendJobExecutionActor providing a standard abstract implementation of an
   * asynchronous polling backend.
@@ -85,7 +92,8 @@ trait StandardAsyncExecutionActor
     with StandardCachingActorHelper
     with AsyncIoActorClient
     with KvClient
-    with SlowJobWarning {
+    with SlowJobWarning
+    with PlatformSpecific {
   this: Actor with ActorLogging with BackendJobLifecycleActor =>
 
   override lazy val ioCommandBuilder: IoCommandBuilder = DefaultIoCommandBuilder
@@ -894,6 +902,14 @@ trait StandardAsyncExecutionActor
     */
   def getTerminalEvents(runStatus: StandardAsyncRunState): Seq[ExecutionEvent] = Seq.empty
 
+  /**
+    * Get the min and max event times from a terminal run status
+    *
+    * @param runStatus The terminal run status, as defined by isTerminal.
+    * @return The min and max event times, if events exist.
+    */
+  def getStartAndEndTimes(runStatus: StandardAsyncRunState): Option[StartAndEndTimes] = None
+
   /**
     * Returns true if the status represents a completion.
     *
@@ -1316,6 +1332,7 @@ trait StandardAsyncExecutionActor
         val metadata = getTerminalMetadata(state)
         onTaskComplete(state, oldHandle)
         tellMetadata(metadata)
+        tellBard(state)
         handleExecutionResult(state, oldHandle)
       case s =>
         Future.successful(
@@ -1509,6 +1526,40 @@ trait StandardAsyncExecutionActor
     serviceRegistryActor.putMetadata(jobDescriptor.workflowDescriptor.id, Option(jobDescriptor.key), metadataKeyValues)
   }
 
+  def tellBard(state: StandardAsyncRunState): Unit =
+    getStartAndEndTimes(state) match {
+      case Some(startAndEndTimes: StartAndEndTimes) =>
+        val dockerImage =
+          RuntimeAttributesValidation.extractOption(DockerValidation.instance, validatedRuntimeAttributes)
+        val cpus = RuntimeAttributesValidation.extract(CpuValidation.instance, validatedRuntimeAttributes).value
+        val memory = RuntimeAttributesValidation
+          .extract(MemoryValidation.instance(), validatedRuntimeAttributes)
+          .to(MemoryUnit.Bytes)
+          .amount
+        serviceRegistryActor ! BardEventRequest(
+          TaskSummaryEvent(
+            workflowDescriptor.id.id,
+            workflowDescriptor.possibleParentWorkflowId.map(_.id),
+            workflowDescriptor.rootWorkflowId.id,
+            jobDescriptor.key.tag,
+            jobDescriptor.key.call.fullyQualifiedName,
+            jobDescriptor.key.index,
+            jobDescriptor.key.attempt,
+            state.getClass.getSimpleName,
+            platform.map(_.runtimeKey),
+            dockerImage,
+            cpus,
+            memory,
+            startAndEndTimes.jobStart.toString,
+            startAndEndTimes.cpuStart.map(_.toString),
+            startAndEndTimes.jobEnd.toString,
+            startAndEndTimes.jobStart.until(startAndEndTimes.jobEnd, ChronoUnit.SECONDS),
+            startAndEndTimes.cpuStart.map(_.until(startAndEndTimes.jobEnd, ChronoUnit.SECONDS))
+          )
+        )
+      case _ => ()
+    }
+
   implicit override protected lazy val ec: ExecutionContextExecutor = context.dispatcher
 }
 

build.sbt

@@ -339,7 +339,7 @@ lazy val languageFactoryCore = (project in languageFactoryRoot / "language-facto
   .dependsOn(common % "test->test")
 
 lazy val wdlDraft2LanguageFactory = (project in languageFactoryRoot / "wdl-draft2")
-  .withLibrarySettings("wdl-draft2", draft2LanguageFactoryDependencies)
+  .withLibrarySettings("wdl-draft2", mockServerDependencies)
   .dependsOn(languageFactoryCore)
   .dependsOn(common % "test->test")
   .dependsOn(wdlModelDraft2)

core/src/main/resources/reference.conf

@@ -36,6 +36,10 @@ akka {
     mailbox-type = "akka.dispatch.UnboundedControlAwareMailbox"
   }
 
+  bard-actor-mailbox {
+    mailbox-type = "akka.dispatch.UnboundedControlAwareMailbox"
+  }
+
   dispatchers {
     # A dispatcher for actors performing blocking io operations
     # Prevents the whole system from being slowed down when waiting for responses from external resources for instance
@@ -582,6 +586,17 @@ services {
       # ecm.base-url = ""
     }
   }
+  // Bard is used for metrics collection in the Terra SaaS offering and is not applicable outside of it.
+  Bard {
+    class = "cromwell.services.metrics.bard.impl.BardEventingActor"
+    config {
+      enabled = false
+      bard {
+        base-url = ""
+        connection-pool-size = 0
+      }
+    }
+  }
 }
 
 include required(classpath("reference_database.inc.conf"))

project/Dependencies.scala

@@ -6,6 +6,7 @@ object Dependencies {
   private val akkaV = "2.5.32" // scala-steward:off (CROM-6637)
   private val ammoniteOpsV = "2.4.1"
   private val apacheHttpClientV = "4.5.13"
+  private val apacheHttpClient5V = "5.3.1"
   private val awsSdkV = "2.17.265"
   // We would like to use the BOM to manage Azure SDK versions, but SBT doesn't support it.
   // https://github.com/Azure/azure-sdk-for-java/tree/main/sdk/boms/azure-sdk-bom
@@ -21,6 +22,7 @@ object Dependencies {
   private val azureAppInsightsLogbackV = "2.6.4"
   private val betterFilesV = "3.9.1"
   private val jsonSmartV = "2.4.10"
+  private val bardClientV = "1.0.4"
   /*
   cats-effect, fs2, http4s, and sttp (also to v3) should all be upgraded at the same time to use cats-effect 3.x.
    */
@@ -421,8 +423,8 @@ object Dependencies {
       exclude("org.apache.httpcomponents", "httpclient"),
     "org.broadinstitute.dsde.workbench" %% "workbench-google" % workbenchGoogleV
       exclude("com.google.apis", "google-api-services-genomics"),
-    "org.apache.httpcomponents" % "httpclient" % apacheHttpClientV,
-    "com.google.apis" % "google-api-services-cloudkms" % googleCloudKmsV
+    "org.apache.httpcomponents.client5" % "httpclient5" % apacheHttpClient5V,
+  "com.google.apis" % "google-api-services-cloudkms" % googleCloudKmsV
       exclude("com.google.guava", "guava-jdk5"),
     "org.glassfish.hk2.external" % "jakarta.inject" % jakartaInjectV,
   ) ++ googleGenomicsV2Alpha1Dependency ++ googleLifeSciencesV2BetaDependency ++ googleBatchv1Dependency
@@ -529,7 +531,7 @@ object Dependencies {
     "jakarta.activation" % "jakarta.activation-api" % jakartaActivationV,
   )
 
-  val draft2LanguageFactoryDependencies = List(
+  val mockServerDependencies = List(
     "org.mock-server" % "mockserver-netty" % mockserverNettyV % Test
   )
 
@@ -593,7 +595,12 @@ object Dependencies {
   val servicesDependencies: List[ModuleID] = List(
     "com.google.api" % "gax-grpc" % googleGaxGrpcV,
     "org.apache.commons" % "commons-csv" % commonsCsvV,
-  ) ++ testDatabaseDependencies ++ akkaHttpDependencies
+    "bio.terra" % "bard-client-resttemplate-javax" % bardClientV
+      exclude("org.springframework", "spring-aop")
+      exclude("org.springframework", "spring-jcl"),
+    "org.apache.httpcomponents.client5" % "httpclient5" % apacheHttpClient5V // Needed for rest-template connection pooling
+
+  ) ++ testDatabaseDependencies ++ akkaHttpDependencies ++ mockServerDependencies
 
   val serverDependencies: List[ModuleID] = slf4jBindingDependencies
 
@@ -664,7 +671,7 @@ object Dependencies {
       cromwellApiClientDependencies ++
       databaseMigrationDependencies ++
       databaseSqlDependencies ++
-      draft2LanguageFactoryDependencies ++
+      mockServerDependencies ++
       drsLocalizerDependencies ++
       engineDependencies ++
       gcsFileSystemDependencies ++

services/src/main/scala/cromwell/services/metrics/bard/BardConfig.scala

@@ -0,0 +1,13 @@
+package cromwell.services.metrics.bard
+
+import com.typesafe.config.Config
+import net.ceedubs.ficus.Ficus._
+
+final case class BardConfig(enabled: Boolean, baseUrl: String, connectionPoolSize: Int)
+
+object BardConfig {
+  def apply(config: Config): BardConfig = BardConfig(config.as[Boolean]("enabled"),
+                                                     config.as[String]("bard.base-url"),
+                                                     config.as[Int]("bard.connection-pool-size")
+  )
+}

services/src/main/scala/cromwell/services/metrics/bard/BardEventing.scala

@@ -0,0 +1,13 @@
+package cromwell.services.metrics.bard
+
+import cromwell.services.ServiceRegistryActor.ServiceRegistryMessage
+import cromwell.services.metrics.bard.model.BardEvent
+
+object BardEventing {
+  sealed trait BardEventingMessage extends ServiceRegistryMessage {
+    override def serviceName: String = "BardEventing"
+  }
+
+  case class BardEventRequest(event: BardEvent) extends BardEventingMessage
+
+}

services/src/main/scala/cromwell/services/metrics/bard/BardService.scala

@@ -0,0 +1,57 @@
+package cromwell.services.metrics.bard
+
+import akka.actor.ActorRef
+import bio.terra.bard.api.DefaultApi
+import bio.terra.bard.client.ApiClient
+import bio.terra.bard.model.EventsEventLogRequest
+import cats.data.NonEmptyList
+import com.typesafe.scalalogging.LazyLogging
+import cromwell.services.instrumentation.CromwellInstrumentation
+import cromwell.services.metrics.bard.model.BardEvent
+import org.apache.http.impl.client.HttpClients
+import org.apache.http.impl.conn.PoolingHttpClientConnectionManager
+import org.springframework.http.client.HttpComponentsClientHttpRequestFactory
+import org.springframework.web.client.RestTemplate
+
+class BardService(bardUrl: String, connectionPoolSize: Int, serviceRegistry: ActorRef)
+    extends LazyLogging
+    with CromwellInstrumentation {
+
+  private val restTemplate = makeRestTemplateWithPooling
+  private val client = getEventApi(restTemplate)
+  private val appId = "cromwell"
+
+  override lazy val serviceRegistryActor: ActorRef = serviceRegistry
+
+  def sendEvent(event: BardEvent): Unit = {
+    try {
+      val eventLogRequest = new EventsEventLogRequest().properties(event.getProperties)
+      client.eventsEventLog(event.eventName, appId, eventLogRequest)
+      increment(NonEmptyList.of("send_event", "success"), Some("bard"))
+    } catch {
+      case e: Exception =>
+        logger.error(s"Failed to send event to Bard: ${e.getMessage}", e)
+        increment(NonEmptyList.of("send_event", "failure"), Some("bard"))
+    }
+    ()
+  }
+
+  private def getEventApi(restTemplate: RestTemplate): DefaultApi = {
+    val bardClient = new ApiClient(restTemplate)
+    bardClient.setBasePath(bardUrl)
+    new DefaultApi(bardClient)
+  }
+
+  /**
+    * @return a new RestTemplate backed by a pooling connection manager
+    */
+  private def makeRestTemplateWithPooling: RestTemplate = {
+    val poolingConnManager = new PoolingHttpClientConnectionManager()
+    poolingConnManager.setMaxTotal(connectionPoolSize)
+    poolingConnManager.setDefaultMaxPerRoute(connectionPoolSize)
+    val httpClient = HttpClients.custom.setConnectionManager(poolingConnManager).build
+    val factory = new HttpComponentsClientHttpRequestFactory(httpClient)
+    new RestTemplate(factory)
+  }
+
+}