xintongsong commented on a change in pull request #9760: [FLINK-13982][runtime] Implement memory calculation logics URL: https://github.com/apache/flink/pull/9760#discussion_r333809794
########## File path: flink-runtime/src/main/java/org/apache/flink/runtime/clusterframework/TaskExecutorResourceUtils.java ########## @@ -0,0 +1,659 @@ +/* + * 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.clusterframework; + +import org.apache.flink.configuration.Configuration; +import org.apache.flink.configuration.IllegalConfigurationException; +import org.apache.flink.configuration.MemorySize; +import org.apache.flink.configuration.NettyShuffleEnvironmentOptions; +import org.apache.flink.configuration.TaskManagerOptions; +import org.apache.flink.runtime.util.ConfigurationParserUtils; + +import java.util.HashMap; +import java.util.Map; + +import static org.apache.flink.util.Preconditions.checkArgument; +import static org.apache.flink.util.Preconditions.checkNotNull; + +/** + * Utility class for TaskExecutor memory configurations. + * + * <p>A TaskExecutor's memory consists of the following components. + * <ul> + * <li>Framework Heap Memory</li> + * <li>Task Heap Memory</li> + * <li>Task Off-Heap Memory</li> + * <li>Shuffle Memory</li> + * <li>Managed Memory</li> + * <ul> + * <li>On-Heap Managed Memory</li> + * <li>Off-Heap Managed Memory</li> + * </ul> + * <li>JVM Metaspace</li> + * <li>JVM Overhead</li> + * </ul> + * Among all the components, Framework Heap Memory, Task Heap Memory and On-Heap Managed Memory use on heap memory, + * while the rest use off heap memory. We use Total Process Memory to refer to all the memory components, while Total + * Flink Memory refering to all the components except JVM Metaspace and JVM Overhead. + * + * <p>The relationships of TaskExecutor memory components are shown below. + * <pre> + * ┌ ─ ─ Total Process Memory ─ ─ ┐ + * ┌ ─ ─ Total Flink Memory ─ ─ ┐ + * │ ┌───────────────────────────┐ │ + * ││ Framework Heap Memory ││ ─┐ + * │ └───────────────────────────┘ │ │ + * │┌───────────────────────────┐│ │ + * │ │ Task Heap Memory │ │ ─┤ + * │└───────────────────────────┘│ │ + * │ ┌───────────────────────────┐ │ │ + * ┌─ ││ Task Off-Heap Memory ││ │ + * │ │ └───────────────────────────┘ │ ├─ On-Heap + * │ │┌───────────────────────────┐│ │ + * ├─ │ │ Shuffle Memory │ │ │ + * │ │└───────────────────────────┘│ │ + * │ │ ┌───── Managed Memory ──────┐ │ │ + * │ ││┌─────────────────────────┐││ │ + * │ │ ││ On-Heap Managed Memory ││ │ ─┘ + * │ ││├─────────────────────────┤││ + * Off-Heap ─┼─ │ ││ Off-Heap Managed Memory ││ │ + * │ ││└─────────────────────────┘││ + * │ │ └───────────────────────────┘ │ + * │ └ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ┘ + * │ │┌─────────────────────────────┐│ + * ├─ │ JVM Metaspace │ + * │ │└─────────────────────────────┘│ + * │ ┌─────────────────────────────┐ + * └─ ││ JVM Overhead ││ + * └─────────────────────────────┘ + * └ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ┘ + * </pre> + */ +public class TaskExecutorResourceUtils { + + private TaskExecutorResourceUtils() {} + + // ------------------------------------------------------------------------ + // Generating JVM Parameters + // ------------------------------------------------------------------------ + + public static String generateJvmParametersStr(final TaskExecutorResourceSpec taskExecutorResourceSpec) { + final MemorySize jvmHeapSize = taskExecutorResourceSpec.getFrameworkHeapSize() + .add(taskExecutorResourceSpec.getTaskHeapSize()) + .add(taskExecutorResourceSpec.getOnHeapManagedMemorySize()); + final MemorySize jvmDirectSize = taskExecutorResourceSpec.getTaskOffHeapSize() + .add(taskExecutorResourceSpec.getShuffleMemSize()); + final MemorySize jvmMetaspaceSize = taskExecutorResourceSpec.getJvmMetaspaceSize(); + + return "-Xmx" + jvmHeapSize.getBytes() + + " -Xms" + jvmHeapSize.getBytes() + + " -XX:MaxDirectMemorySize=" + jvmDirectSize.getBytes() + + " -XX:MetaspaceSize=" + jvmMetaspaceSize.getBytes(); + } + + // ------------------------------------------------------------------------ + // Generating Dynamic Config Options + // ------------------------------------------------------------------------ + + public static String generateDynamicConfigsStr(final TaskExecutorResourceSpec taskExecutorResourceSpec) { + final Map<String, String> configs = new HashMap<>(); + configs.put(TaskManagerOptions.FRAMEWORK_HEAP_MEMORY.key(), taskExecutorResourceSpec.getFrameworkHeapSize().getBytes() + "b"); + configs.put(TaskManagerOptions.TASK_HEAP_MEMORY.key(), taskExecutorResourceSpec.getTaskHeapSize().getBytes() + "b"); + configs.put(TaskManagerOptions.TASK_OFF_HEAP_MEMORY.key(), taskExecutorResourceSpec.getTaskOffHeapSize().getBytes() + "b"); + configs.put(TaskManagerOptions.SHUFFLE_MEMORY_MIN.key(), taskExecutorResourceSpec.getShuffleMemSize().getBytes() + "b"); + configs.put(TaskManagerOptions.SHUFFLE_MEMORY_MAX.key(), taskExecutorResourceSpec.getShuffleMemSize().getBytes() + "b"); + configs.put(TaskManagerOptions.MANAGED_MEMORY_SIZE.key(), taskExecutorResourceSpec.getManagedMemorySize().getBytes() + "b"); + configs.put(TaskManagerOptions.MANAGED_MEMORY_OFFHEAP_SIZE.key(), taskExecutorResourceSpec.getOffHeapManagedMemorySize().getBytes() + "b"); + return assembleDynamicConfigsStr(configs); + } + + private static String assembleDynamicConfigsStr(final Map<String, String> configs) { + final StringBuilder sb = new StringBuilder(); + for (Map.Entry<String, String> entry : configs.entrySet()) { + sb.append("-D ").append(entry.getKey()).append("=").append(entry.getValue()).append(" "); + } + return sb.toString(); + } + + // ------------------------------------------------------------------------ + // Memory Configuration Calculations + // ------------------------------------------------------------------------ + + public static TaskExecutorResourceSpec resourceSpecFromConfig(final Configuration config) { + if (isTaskHeapMemorySizeExplicitlyConfigured(config) && isManagedMemorySizeExplicitlyConfigured(config)) { Review comment: We can derive managed memory size from its fraction in either of the following two ways: - `managed_size = total_flink_size * managed_fraction` - `managed_size = remaining_flink_internal_size / (1 - managed_fraction) * managed_fraction` If total flink memory is set or can be derived from total process memory, we use the first equation to derive managed memory size in later 'else-if' branches. For this 'if' branch, if managed memory size is not explicitly configured and we want to derive it from its fraction, we will have to use the second equation. In that case, we need `remaining_flink_internal_size` to be determinate, which may not always be the case given that shuffle memory size can also not be explicitly configured. There are some alternatives that I can think of, but I'm not sure whether it's necessary to go with either of them, given the complexity for user to understand and for the maintainability. - We can derive managed and shuffle memory size together, with the following equations - `managed_and_shuffle_size = remaining_flink_internal_size / (1 - managed_fraction - shuffle_fraction) * (managed_fraction + shuffle_fraction)` - `managed_size = managed_and_shuffle_size / (managed_fraction + shuffle_fraction) * managed_fraction` - `shuffle_size = managed_and_shuffle_size / (managed_fraction + shuffle_fraction) * shuffle_fraction` But given that shuffle memory size also need to respect to the min/max, this approach can be even more complicated. - We can require either managed memory size or shuffle memory size to be explicitly configured, and derive the other with its fraction. This also complicates the calculation logic by introducing more 'if-else' branches. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services
