SteNicholas commented on code in PR #3858: URL: https://github.com/apache/celeborn/pull/3858#discussion_r4444775417
########## worker/src/main/scala/org/apache/celeborn/service/deploy/worker/metrics/ScaleMetricCollector.scala: ########## @@ -0,0 +1,29 @@ +/* + * 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.celeborn.service.deploy.worker.metrics + +import org.apache.celeborn.common.metrics.WorkerStats +import org.apache.celeborn.service.deploy.worker.Worker + +trait ScaleMetricCollector { + def init(worker: Worker): Unit Review Comment: [P2] Make the collector initialization API accessible to external Scala implementations `Worker` is declared `private[celeborn]`, so a Scala collector in a user package cannot implement this method. I reproduced this against 20120aa0: an implementation in `com.example.scaling` fails with `class Worker in package worker cannot be accessed`, while the equivalent implementation under `org.apache.celeborn` compiles. This prevents users from implementing the configurable collector in their own Scala namespace, and the current integration test misses it because its collector is inside the Celeborn package. Could we expose a public collector context with the required observations, or otherwise make the parameter type accessible, and add a compilation test for an implementation outside `org.apache.celeborn`? ########## worker/src/main/scala/org/apache/celeborn/service/deploy/worker/metrics/DefaultScaleMetricCollector.scala: ########## @@ -0,0 +1,111 @@ +/* + * 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.celeborn.service.deploy.worker.metrics + +import java.lang.management.{ManagementFactory, OperatingSystemMXBean} +import java.util.concurrent.{ScheduledFuture, TimeUnit} + +import scala.util.control.NonFatal + +import org.apache.celeborn.common.CelebornConf +import org.apache.celeborn.common.internal.Logging +import org.apache.celeborn.common.metrics.WorkerStats +import org.apache.celeborn.common.util.{DiskUtils, ThreadUtils} +import org.apache.celeborn.service.deploy.worker.Worker + +class DefaultScaleMetricCollector private[metrics] ( + conf: CelebornConf, + operatingSystem: OperatingSystemMXBean) + extends ScaleMetricCollector with Logging { + + def this(conf: CelebornConf) = + this(conf, ManagementFactory.getOperatingSystemMXBean) + + private val memoryWindow = new MetricSlidingWindow(conf.scaleMetricMemoryWindowSize) + private val executor = + ThreadUtils.newDaemonSingleThreadScheduledExecutor("scale-metric-collector") + private var scheduledTask: ScheduledFuture[_] = _ + + @volatile private var stopped = false + @volatile private var currentStats: Option[WorkerStats] = None + + override def init(worker: Worker): Unit = synchronized { + require(!stopped && scheduledTask == null, "Metric collector can only be initialized once") + scheduledTask = executor.scheduleWithFixedDelay( + new Runnable { + override def run(): Unit = collectOnce(worker) + }, + conf.scaleMetricInterval, + conf.scaleMetricInterval, + TimeUnit.MILLISECONDS) + } + + override def stop(): Unit = synchronized { + stopped = true + currentStats = None + if (scheduledTask != null) { + scheduledTask.cancel(true) + } + executor.shutdownNow() + } + + override def currentWorkerStats(): Option[WorkerStats] = { + if (stopped) None + else currentStats + } + + private[metrics] def collectOnce(worker: Worker): Unit = { + try { + val metrics = memoryMetrics(worker) ++ systemLoadMetrics() ++ diskMetrics(worker) + currentStats = Some(WorkerStats(metrics)) + } catch { + case NonFatal(e) => + logWarning("Failed to collect worker scaling metrics", e) + } + } + + private def memoryMetrics(worker: Worker): Map[String, Double] = { + val capacity = worker.memoryManager.maxDirectMemory + val used = worker.memoryManager.getMemoryUsage + memoryWindow.update(used) + Map(WorkerStats.NettyMemoryUsedRatio -> (memoryWindow.average / capacity)) Review Comment: [P3] Align the memory metric name with its numerator `MemoryManager.getMemoryUsage()` returns `getNettyUsedDirectMemory() + sortMemoryCounter.get()`, so this reports combined memory pressure under the name `NettyMemoryUsedRatio`. When sorting reserves memory, the reported value can rise without Netty usage increasing, making the metric misleading for consumers and comparisons with the existing `NettyMemory` gauge. If the combined accounting is intended, could we rename the metric to something such as `DirectMemoryUsedRatio` and update the configuration documentation and tests? If it is intended to measure only Netty usage, the numerator should use `getNettyUsedDirectMemory()` instead. -- 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. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
