Github user koeninger commented on a diff in the pull request: https://github.com/apache/spark/pull/15307#discussion_r81642196 --- Diff: sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/StreamMetrics.scala --- @@ -0,0 +1,252 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.sql.execution.streaming + +import scala.collection.mutable + +import com.codahale.metrics.{Gauge, MetricRegistry} + +import org.apache.spark.internal.Logging +import org.apache.spark.metrics.source.{Source => CodahaleSource} +import org.apache.spark.util.Clock + +class StreamMetrics(sources: Set[Source], triggerClock: Clock, codahaleSourceName: String) + extends CodahaleSource with Logging { + + import StreamMetrics._ + + // Trigger infos + private val triggerInfo = new mutable.HashMap[String, String] + private val sourceTriggerInfo = new mutable.HashMap[Source, mutable.HashMap[String, String]] + + // Rate estimators for sources and sinks + private val inputRates = new mutable.HashMap[Source, RateCalculator] + private val processingRates = new mutable.HashMap[Source, RateCalculator] + private val outputRate = new RateCalculator + + // Number of input rows in the current trigger + private val numInputRows = new mutable.HashMap[Source, Long] + private var numOutputRows: Option[Long] = None + private var currentTriggerStartTimestamp: Long = -1 + private var previousTriggerStartTimestamp: Long = -1 + private var latency: Option[Double] = None + + override val sourceName: String = codahaleSourceName + override val metricRegistry: MetricRegistry = new MetricRegistry + + // =========== Initialization =========== + + // Metric names should not have . in them, so that all the metrics of a query are identified + // together in Ganglia as a single metric group + registerGauge("inputRate-total", currentInputRate) + registerGauge("processingRate-total", () => currentProcessingRate) + registerGauge("outputRate", () => currentOutputRate) + registerGauge("latency", () => currentLatency().getOrElse(-1.0)) + + sources.foreach { s => + inputRates.put(s, new RateCalculator) + processingRates.put(s, new RateCalculator) + sourceTriggerInfo.put(s, new mutable.HashMap[String, String]) + + registerGauge(s"inputRate-${s.toString}", () => currentSourceInputRate(s)) + registerGauge(s"processingRate-${s.toString}", () => currentSourceProcessingRate(s)) + } + + // =========== Setter methods =========== + + def reportTriggerStarted(triggerId: Long): Unit = synchronized { + numInputRows.clear() + numOutputRows = None + triggerInfo.clear() + sourceTriggerInfo.values.foreach(_.clear()) + + reportTriggerInfo(TRIGGER_ID, triggerId) + sources.foreach(s => reportSourceTriggerInfo(s, TRIGGER_ID, triggerId)) + reportTriggerInfo(ACTIVE, true) + currentTriggerStartTimestamp = triggerClock.getTimeMillis() + reportTriggerInfo(START_TIMESTAMP, currentTriggerStartTimestamp) + } + + def reportTimestamp(key: String): Unit = synchronized { + triggerInfo.put(key, triggerClock.getTimeMillis().toString) + } + + def reportLatency(key: String, latencyMs: Long): Unit = synchronized { + triggerInfo.put(key, latencyMs.toString) + } + + def reportLatency(source: Source, key: String, latencyMs: Long): Unit = synchronized { + sourceTriggerInfo(source).put(key, latencyMs.toString) + } + + def reportTriggerInfo[T](key: String, value: T): Unit = synchronized { + triggerInfo.put(key, value.toString) + } + + def reportSourceTriggerInfo[T](source: Source, key: String, value: T): Unit = synchronized { + sourceTriggerInfo(source).put(key, value.toString) + } + + def reportNumRows(inputRows: Map[Source, Long], outputRows: Option[Long]): Unit = synchronized { + numInputRows ++= inputRows + numOutputRows = outputRows + } + + def reportTriggerFinished(): Unit = synchronized { + require(currentTriggerStartTimestamp >= 0) + val currentTriggerFinishTimestamp = triggerClock.getTimeMillis() + reportTriggerInfo(FINISH_TIMESTAMP, currentTriggerFinishTimestamp) + reportTriggerInfo(STATUS_MESSAGE, "") + reportTriggerInfo(ACTIVE, false) + + // Report number of rows + val totalNumInputRows = numInputRows.values.sum + reportTriggerInfo(NUM_INPUT_ROWS, totalNumInputRows) + reportTriggerInfo(NUM_OUTPUT_ROWS, numOutputRows.getOrElse(0)) + numInputRows.foreach { case (s, r) => + reportSourceTriggerInfo(s, NUM_SOURCE_INPUT_ROWS, r) + } + + val currentTriggerDuration = currentTriggerFinishTimestamp - currentTriggerStartTimestamp + val previousInputIntervalOption = if (previousTriggerStartTimestamp >= 0) { + Some(currentTriggerStartTimestamp - previousTriggerStartTimestamp) + } else None + + // Update input rate = num rows received by each source during the previous trigger interval + // Interval is measures as interval between start times of previous and current trigger. + // + // TODO: Instead of trigger start, we should use time when getOffset was called on each source + // as this may be different for each source if there are many sources in the query plan + // and getOffset is called serially on them. + if (previousInputIntervalOption.nonEmpty) { + sources.foreach { s => + inputRates(s).update(numInputRows.getOrElse(s, 0), previousInputIntervalOption.get) + } + } + + // Update processing rate = num rows processed for each source in current trigger duration + sources.foreach { s => + processingRates(s).update(numInputRows.getOrElse(s, 0), currentTriggerDuration) + } + + // Update output rate = num rows output to the sink in current trigger duration + outputRate.update(numOutputRows.getOrElse(0), currentTriggerDuration) + logDebug("Output rate updated to " + outputRate.currentRate) + + // Update latency = if data present, 0.5 * previous trigger interval + current trigger duration + if (previousInputIntervalOption.nonEmpty && totalNumInputRows > 0) { + latency = Some((previousInputIntervalOption.get.toDouble / 2) + currentTriggerDuration) + } else { + latency = None + } + + previousTriggerStartTimestamp = currentTriggerStartTimestamp + currentTriggerStartTimestamp = -1 + } + + // =========== Getter methods =========== + + def currentInputRate(): Double = synchronized { + // Since we are calculating source input rates using the same time interval for all sources + // it is fine to calculate total input rate as the sum of per source input rate. + inputRates.map(_._2.currentRate).sum --- End diff -- If RateCalculator has a zero and an addition method, then you can do inputRates.map(_._2).foldLeft(RateCalculator.zero)(_+_) and not have to rely on the comment about using currentRate being safe because time intervals are the same. If you don't want to mess with greatest common divisor or whatever until you have an actual different interval, you can just make the addition method throw unless the intervals in the two rate objects are the same.
--- If your project is set up for it, you can reply to this email and have your reply appear on GitHub as well. If your project does not have this feature enabled and wishes so, or if the feature is enabled but not working, please contact infrastructure at infrastruct...@apache.org or file a JIRA ticket with INFRA. --- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org