Merge pull request #743 from pwendell/app-metrics

Add application metrics to standalone master

conf/metrics.properties.template

@@ -1,48 +1,45 @@
-# syntax: [instance].[sink|source].[name].[options]
+#  syntax: [instance].sink|source.[name].[options]=[value]
 
-#  "instance" specify "who" (the role) use metrics system. In spark there are
+#  This file configures Spark's internal metrics system. The metrics system is
-#  several roles like master, worker, executor, driver, these roles will
+#  divided into instances which correspond to internal components.
-#  create metrics system for monitoring. So instance represents these roles.
+#  Each instance can be configured to report its metrics to one or more sinks.
-#  Currently in Spark, several instances have already implemented: master,
+#  Accepted values for [instance] are "master", "worker", "executor", "driver", 
-#  worker, executor, driver.
+#  and "applications". A wild card "*" can be used as an instance name, in 
+#  which case all instances will inherit the supplied property.
 #
-#  [instance] field can be "master", "worker", "executor", "driver", which means
+#  Within an instance, a "source" specifies a particular set of grouped metrics.
-#  only the specified instance has this property.
+#  there are two kinds of sources:
-#  a wild card "*" can be used to represent instance name, which means all the
+#    1. Spark internal sources, like MasterSource, WorkerSource, etc, which will
-#  instances will have this property.
+#    collect a Spark component's internal state. Each instance is paired with a
+#    Spark source that is added automatically.
+#    2. Common sources, like JvmSource, which will collect low level state.
+#    These can be added through configuration options and are then loaded
+#    using reflection.
 #
-#  "source" specify "where" (source) to collect metrics data. In metrics system,
+#  A "sink" specifies where metrics are delivered to. Each instance can be
-#  there exists two kinds of source:
+#  assigned one or more sinks.
-#    1. Spark internal source, like MasterSource, WorkerSource, etc, which will
-#    collect Spark component's internal state, these sources are related to
-#    instance and will be added after specific metrics system is created.
-#    2. Common source, like JvmSource, which will collect low level state, is
-#    configured by configuration and loaded through reflection.
 #
-#  "sink" specify "where" (destination) to output metrics data to. Several sinks
+#  The sink|source field specifies whether the property relates to a sink or 
-#  can be coexisted and flush metrics to all these sinks.
+#  source.
 #
-#  [sink|source] field specify this property is source related or sink, this
+#  The [name] field specifies the name of source or sink.
-#  field can only be source or sink.
 #
-#  [name] field specify the name of source or sink, this is custom defined.
+#  The [options] field is the specific property of this source or sink. The
-#
+#  source or sink is responsible for parsing this property.
-#  [options] field is the specific property of this source or sink, this source
-#  or sink is responsible for parsing this property.
 #
 #  Notes:
-#    1. Sinks should be added through configuration, like console sink, class
+#    1. To add a new sink, set the "class" option to a fully qualified class 
-#    full name should be specified by class property.
+#    name (see examples below).
-#    2. Some sinks can specify polling period, like console sink, which is 10 seconds,
+#    2. Some sinks involve a polling period. The minimum allowed polling period
-#    it should be attention minimal polling period is 1 seconds, any period
+#    is  1 second.
-#    below than 1s is illegal.
+#    3. Wild card properties can be overridden by more specific properties. 
-#    3. Wild card property can be overlapped by specific instance property, for
+#    For example, master.sink.console.period takes precedence over 
-#    example, *.sink.console.period can be overlapped by master.sink.console.period.
+#    *.sink.console.period.
 #    4. A metrics specific configuration
 #    "spark.metrics.conf=${SPARK_HOME}/conf/metrics.properties" should be
-#    added to Java property using -Dspark.metrics.conf=xxx if you want to
+#    added to Java properties using -Dspark.metrics.conf=xxx if you want to
-#    customize metrics system, or you can put it in ${SPARK_HOME}/conf,
+#    customize metrics system. You can also put the file in ${SPARK_HOME}/conf
-#    metrics system will search and load it automatically.
+#    and it will be loaded automatically.
 
 # Enable JmxSink for all instances by class name
 #*.sink.jmx.class=spark.metrics.sink.JmxSink

core/src/main/scala/spark/deploy/master/ApplicationInfo.scala

@@ -34,6 +34,7 @@ private[spark] class ApplicationInfo(
   var executors = new mutable.HashMap[Int, ExecutorInfo]
   var coresGranted = 0
   var endTime = -1L
+  val appSource = new ApplicationSource(this)
 
   private var nextExecutorId = 0
 

core/src/main/scala/spark/deploy/master/ApplicationSource.scala

@@ -0,0 +1,24 @@
+package spark.deploy.master
+
+import com.codahale.metrics.{Gauge, MetricRegistry}
+
+import spark.metrics.source.Source
+
+class ApplicationSource(val application: ApplicationInfo) extends Source {
+  val metricRegistry = new MetricRegistry()
+  val sourceName = "%s.%s.%s".format("application", application.desc.name,
+    System.currentTimeMillis())
+
+  metricRegistry.register(MetricRegistry.name("status"), new Gauge[String] {
+    override def getValue: String = application.state.toString
+  })
+
+  metricRegistry.register(MetricRegistry.name("runtime_ms"), new Gauge[Long] {
+    override def getValue: Long = application.duration
+  })
+
+  metricRegistry.register(MetricRegistry.name("cores", "number"), new Gauge[Int] {
+    override def getValue: Int = application.coresGranted
+  })
+
+}

core/src/main/scala/spark/deploy/master/Master.scala

@@ -38,6 +38,7 @@ import spark.util.AkkaUtils
 private[spark] class Master(host: String, port: Int, webUiPort: Int) extends Actor with Logging {
   val DATE_FORMAT = new SimpleDateFormat("yyyyMMddHHmmss")  // For application IDs
   val WORKER_TIMEOUT = System.getProperty("spark.worker.timeout", "60").toLong * 1000
+  val RETAINED_APPLICATIONS = System.getProperty("spark.deploy.retainedApplications", "200").toInt
 
   var nextAppNumber = 0
   val workers = new HashSet[WorkerInfo]
@@ -59,7 +60,8 @@ private[spark] class Master(host: String, port: Int, webUiPort: Int) extends Act
 
   Utils.checkHost(host, "Expected hostname")
 
-  val metricsSystem = MetricsSystem.createMetricsSystem("master")
+  val masterMetricsSystem = MetricsSystem.createMetricsSystem("master")
+  val applicationMetricsSystem = MetricsSystem.createMetricsSystem("applications")
   val masterSource = new MasterSource(this)
 
   val masterPublicAddress = {
@@ -79,13 +81,15 @@ private[spark] class Master(host: String, port: Int, webUiPort: Int) extends Act
     webUi.start()
     context.system.scheduler.schedule(0 millis, WORKER_TIMEOUT millis)(timeOutDeadWorkers())
 
-    metricsSystem.registerSource(masterSource)
+    masterMetricsSystem.registerSource(masterSource)
-    metricsSystem.start()
+    masterMetricsSystem.start()
+    applicationMetricsSystem.start()
   }
 
   override def postStop() {
     webUi.stop()
-    metricsSystem.stop()
+    masterMetricsSystem.stop()
+    applicationMetricsSystem.stop()
   }
 
   override def receive = {
@@ -275,6 +279,7 @@ private[spark] class Master(host: String, port: Int, webUiPort: Int) extends Act
     val now = System.currentTimeMillis()
     val date = new Date(now)
     val app = new ApplicationInfo(now, newApplicationId(date), desc, date, driver, desc.appUiUrl)
+    applicationMetricsSystem.registerSource(app.appSource)
     apps += app
     idToApp(app.id) = app
     actorToApp(driver) = app
@@ -300,7 +305,14 @@ private[spark] class Master(host: String, port: Int, webUiPort: Int) extends Act
       idToApp -= app.id
       actorToApp -= app.driver
       addressToApp -= app.driver.path.address
-      completedApps += app   // Remember it in our history
+      if (completedApps.size >= RETAINED_APPLICATIONS) {
+        val toRemove = math.max(RETAINED_APPLICATIONS / 10, 1)
+        completedApps.take(toRemove).foreach( a => {
+          applicationMetricsSystem.removeSource(a.appSource)
+        })
+        completedApps.trimStart(toRemove)
+      }
+      completedApps += app // Remember it in our history
       waitingApps -= app
       for (exec <- app.executors.values) {
         exec.worker.removeExecutor(exec)

core/src/main/scala/spark/metrics/MetricsSystem.scala

@@ -17,7 +17,7 @@
 
 package spark.metrics
 
-import com.codahale.metrics.{JmxReporter, MetricSet, MetricRegistry}
+import com.codahale.metrics.{Metric, MetricFilter, MetricRegistry}
 
 import java.util.Properties
 import java.util.concurrent.TimeUnit
@@ -93,6 +93,13 @@ private[spark] class MetricsSystem private (val instance: String) extends Loggin
     }
   }
 
+  def removeSource(source: Source) {
+    sources -= source
+    registry.removeMatching(new MetricFilter {
+      def matches(name: String, metric: Metric): Boolean = name.startsWith(source.sourceName)
+    })
+  }
+
   def registerSources() {
     val instConfig = metricsConfig.getInstance(instance)
     val sourceConfigs = metricsConfig.subProperties(instConfig, MetricsSystem.SOURCE_REGEX)