org.apache.flink.runtime.healthmanager.plugins.resolvers.DryRunParallelismResolver Maven / Gradle / Ivy

Show more of this group Show more artifacts with this name
Show all versions of flink-runtime_2.11 Show documentation
There is a newer version: 1.5.1
/*
 * 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.flink.runtime.healthmanager.plugins.resolvers;

import org.apache.flink.runtime.healthmanager.HealthMonitor;
import org.apache.flink.runtime.healthmanager.plugins.Resolver;
import org.apache.flink.runtime.healthmanager.plugins.Symptom;
import org.apache.flink.runtime.healthmanager.plugins.actions.RescaleJobParallelism;
import org.apache.flink.runtime.healthmanager.plugins.executors.DryRunActionExecutor;
import org.apache.flink.runtime.healthmanager.plugins.symptoms.DryRunCheckSignal;
import org.apache.flink.runtime.healthmanager.plugins.utils.HealthMonitorOptions;
import org.apache.flink.runtime.healthmanager.plugins.utils.TaskMetrics;
import org.apache.flink.runtime.jobgraph.JobVertexID;

import org.slf4j.Logger;
import org.slf4j.LoggerFactory;

import java.util.HashMap;
import java.util.HashSet;
import java.util.List;
import java.util.Map;
import java.util.Set;

/**
 * Resolver for dry run.
 */
public class DryRunParallelismResolver implements Resolver {

	private static final Logger LOGGER = LoggerFactory.getLogger(DryRunParallelismResolver.class);

	private HealthMonitor monitor;

	private ParallelismScaler parallelismScaler = new ParallelismScaler();

	private ParallelismResolverUtils resolverUtils;

	private double scaleTpsRatio;

	private double maxRatio;

	private double minDiffParallelismRatio;

	@Override
	public void open(HealthMonitor monitor) {
		this.monitor = monitor;
		this.parallelismScaler.open(monitor);
		this.resolverUtils = new ParallelismResolverUtils(monitor.getJobTopologyAnalyzer(), monitor.getConfig());
		this.scaleTpsRatio = monitor.getConfig().getDouble(HealthMonitorOptions.PARALLELISM_MIN_RATIO);
		this.maxRatio = monitor.getConfig().getDouble(HealthMonitorOptions.PARALLELISM_MAX_RATIO);
		this.minDiffParallelismRatio = monitor.getConfig().getDouble(HealthMonitorOptions.PARALLELISM_SCALE_MIN_DIFF_RATIO);
	}

	@Override
	public void close() {
		parallelismScaler.close();
	}

	@Override
	public RescaleJobParallelism resolve(List symptomList) {

		boolean triggered = false;
		for (Symptom symptom : symptomList) {
			if (symptom == DryRunCheckSignal.INSTANCE) {
				triggered = true;
				break;
			}
		}

		if (!triggered) {
			return null;
		}

		parallelismScaler.diagnose(symptomList);

		Map taskMetrics = parallelismScaler.getTaskMetrics();
		if (taskMetrics == null || !parallelismScaler.canRescale(taskMetrics)) {
			// only work with task metrics.
			LOGGER.debug("Can not rescale now, task metrics {}", taskMetrics);
			return null;
		}

		for (JobVertexID vertexID : taskMetrics.keySet()) {
			LOGGER.debug("Task Metrics {}", taskMetrics.get(vertexID));
			if (taskMetrics.get(vertexID).getWorkload() < 0) {
				LOGGER.debug("Can not rescale for vertex {} workload is invalid.", vertexID);
				return null;
			}
		}

		Map targetWorkload =
				resolverUtils.estimateVertexWorkload(taskMetrics);

		LOGGER.debug("Estimated target workload: {}", targetWorkload);

		Map minParallelisms = parallelismScaler.getMinParallelism(taskMetrics);

		LOGGER.debug("min parallelism: {}", minParallelisms);

		DryRunActionExecutor dryRunActionExecutor = (DryRunActionExecutor) monitor.getActionExecutor();
		Map currentParallelism = dryRunActionExecutor.getCurrentParallelism();

		LOGGER.debug("current parallelism: {}", currentParallelism);

		Set vertexToRescale = new HashSet<>();
		for (JobVertexID vertexID : targetWorkload.keySet()) {

			// vertex need to scale up.
			if (currentParallelism.get(vertexID) < targetWorkload.get(vertexID) * (1 + minDiffParallelismRatio)) {
				vertexToRescale.add(vertexID);
			}

			// vertex need to scale down.
			if (currentParallelism.get(vertexID) > Math.ceil(taskMetrics.get(vertexID).getWorkload() * maxRatio) &&
					currentParallelism.get(vertexID) - minParallelisms.get(vertexID) > Math.ceil(currentParallelism.get(vertexID) * minDiffParallelismRatio)) {
				vertexToRescale.add(vertexID);
			}
		}

		LOGGER.debug("rescaling {}", vertexToRescale);

		Map targetParallelisms =
				getVertexTargetParallelisms(targetWorkload, vertexToRescale);

		LOGGER.debug("target parallelism before apply constraint {}", targetParallelisms);

		parallelismScaler.updateTargetParallelismsSubjectToConstraints(
				targetParallelisms,
				minParallelisms,
				monitor.getJobConfig());

		LOGGER.debug("target parallelism after apply constraint {}", targetParallelisms);

		return parallelismScaler.generateRescaleParallelismAction(
				targetParallelisms, minParallelisms, monitor.getJobConfig());

	}

	private Map getVertexTargetParallelisms(Map targetStatus, Set vertexToRescale) {
		Map result = new HashMap<>();
		for (JobVertexID vertexID : vertexToRescale) {
			result.put(vertexID, (int) Math.ceil(targetStatus.get(vertexID) * scaleTpsRatio));
		}
		return result;
	}

}