[
https://issues.apache.org/jira/browse/HADOOP-19729?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18033628#comment-18033628
]
ASF GitHub Bot commented on HADOOP-19729:
-----------------------------------------
anujmodi2021 commented on code in PR #8043:
URL: https://github.com/apache/hadoop/pull/8043#discussion_r2470505964
##########
hadoop-tools/hadoop-azure/src/main/java/org/apache/hadoop/fs/azurebfs/services/SlidingWindowHdrHistogram.java:
##########
@@ -0,0 +1,249 @@
+/**
+ * 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.hadoop.fs.azurebfs.services;
+import org.HdrHistogram.Histogram;
+import org.HdrHistogram.Recorder;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+
+import java.util.concurrent.atomic.AtomicLong;
+import java.util.concurrent.locks.ReentrantLock;
+import java.util.concurrent.atomic.AtomicInteger;
+
+import org.apache.hadoop.classification.VisibleForTesting;
+
+public class SlidingWindowHdrHistogram {
+ private static final Logger LOG =
LoggerFactory.getLogger(SlidingWindowHdrHistogram.class);
+
+ // Configuration
+ private final long windowSizeMillis; // Total analysis window
+ private final long timeSegmentDurationMillis; // Subdivision on
analysis window
+ private final int numSegments;
+ private final long highestTrackableValue;
+ private final int significantFigures;
+
+ // Ring buffer of immutable snapshots for completed time segments
+ private final Histogram[] completedSegments;
+ private final AtomicInteger currentIndex = new AtomicInteger(0);
+
+ // Active Time Segment
+ private volatile Recorder activeSegmentRecorder;
+ private Histogram currentSegmentAccumulation;
+ private volatile long currentSegmentStartMillis;
+ private final AtomicLong currentTotalCount = new AtomicLong(0L);
+
+ // Synchronization
+ // Writers never take locks. Readers (queries) and rotation use this lock
+ // to mutate currentAccumulation and ring-buffer pointers safely.
+ private final ReentrantLock rotateLock = new ReentrantLock();
+
+ // Reusable temp histograms to minimize allocations
+ private Histogram tmpForDelta;
+ private Histogram tmpForMerge;
+
+ private final AbfsRestOperationType operationType;
+
+ private boolean isAnalysisWindowFilled = false;
+ private int minSampleSize;
+ private int tailLatencyPercentile;
+ private int tailLatencyMinDeviation;
+
+ private double p50 = 0.0;
+ private double p90 = 0.0;
+ private double p99 = 0.0;
+ private double tailLatency = 0.0;
+ private int deviation = 0;
+
+ public SlidingWindowHdrHistogram(long windowSizeMillis,
+ int numberOfSegments,
+ int minSampleSize,
+ int tailLatencyPercentile,
+ int tailLatencyMinDeviation,
+ long highestTrackableValue,
+ int significantFigures,
+ final AbfsRestOperationType operationType) {
+ if (windowSizeMillis <= 0) throw new
IllegalArgumentException("windowSizeMillis > 0");
+ if (numberOfSegments <= 0) throw new
IllegalArgumentException("bucketDurationMillis > 0");
+ if (highestTrackableValue <= 0) throw new
IllegalArgumentException("highestTrackableValue > 0");
+ if (significantFigures < 1 || significantFigures > 5) throw new
IllegalArgumentException("significantFigures in [1,5]");
+
+ this.windowSizeMillis = windowSizeMillis;
+ this.numSegments = numberOfSegments;
+ this.timeSegmentDurationMillis = windowSizeMillis/numberOfSegments;
+ this.highestTrackableValue = highestTrackableValue;
+ this.significantFigures = significantFigures;
+ this.operationType = operationType;
+ this.minSampleSize = minSampleSize;
+ this.tailLatencyPercentile = tailLatencyPercentile;
+ this.tailLatencyMinDeviation = tailLatencyMinDeviation; // 5ms
+
+ this.completedSegments = new Histogram[numSegments];
+ long now = System.currentTimeMillis();
+ this.currentSegmentStartMillis = alignToSegmentDuration(now);
+ currentIndex.set(0);
+ this.activeSegmentRecorder = new Recorder(highestTrackableValue,
significantFigures);
+ this.currentSegmentAccumulation = new Histogram(highestTrackableValue,
significantFigures);
+ this.tmpForDelta = new Histogram(highestTrackableValue,
significantFigures);
+ this.tmpForMerge = new Histogram(highestTrackableValue,
significantFigures);
+
+ LOG.debug("[{}] Initialized SlidingWindowHdrHistogram with WindowSize {},
TimeSegmentDur: {}, NumOfSegments: {}", operationType, windowSizeMillis,
timeSegmentDurationMillis, numSegments);
+ }
+
+ /** Record a single latency value (in your chosen time unit). Thread-safe
and lock-free. */
+ public void recordValue(long value) {
+ if (value < 0 || value > highestTrackableValue) {
+ LOG.warn("[{}] Value {} outside of range [0, {}]. Ignoring",
+ operationType, value, highestTrackableValue);
+ return;
+ }
+ activeSegmentRecorder.recordValue(value);
+ currentTotalCount.incrementAndGet();
+ LOG.debug("[{}] Recorded latency value: {}. Current total count: {}",
+ operationType, value, currentTotalCount.get());
+ }
+
+ /** Get any percentile over the current sliding window. */
+ public void computeLatency() {
+ if (getCurrentTotalCount() < minSampleSize) {
+ LOG.debug("[{}] Not enough data to report percentiles. Current total
count: {}",
+ operationType, getCurrentTotalCount());
+ } else {
+ rotateLock.lock();
+ try {
+ tmpForMerge.reset();
+ for (int i = 0; i < numSegments; i++) {
+ Histogram h = completedSegments[i];
+ if (h != null && h.getTotalCount() > 0) {
+ tmpForMerge.add(h);
+ }
+ }
+
+ if (tmpForMerge.getTotalCount() == 0) return;
+
+ tailLatency = tmpForMerge.getValueAtPercentile(tailLatencyPercentile);
+ p50 = tmpForMerge.getValueAtPercentile(50);
+ p90 = tmpForMerge.getValueAtPercentile(90);
+ p99 = tmpForMerge.getValueAtPercentile(99);
+ deviation = (int) ((tailLatency - p50)/p50 * 100);
+ } finally {
+ rotateLock.unlock();
+ }
+ }
+ LOG.debug("[{}] Computed Latencies. p50: {}, p90: {}, p99: {},
tailLatency: {}, deviation with p50: {} Current total count: {}",
+ operationType, p50, p90, p99, tailLatency, deviation,
getCurrentTotalCount());
+ }
+
+ private long alignToSegmentDuration(long timeMs) {
+ return timeMs - (timeMs % timeSegmentDurationMillis);
+ }
+
+ /** Ensure active bucket is aligned to current time; rotate if we've crossed
a boundary. */
+ public void rotateIfNeeded() {
+ LOG.debug("[{}] Triggering Histogram Rotation", operationType);
+ long now = System.currentTimeMillis();
Review Comment:
Taken
> ABFS: [Perf] Network Profiling of Tailing Requests and Killing Bad
> Connections Proactively
> ------------------------------------------------------------------------------------------
>
> Key: HADOOP-19729
> URL: https://issues.apache.org/jira/browse/HADOOP-19729
> Project: Hadoop Common
> Issue Type: Sub-task
> Components: fs/azure
> Affects Versions: 3.4.2
> Reporter: Anuj Modi
> Assignee: Anuj Modi
> Priority: Major
> Labels: pull-request-available
>
> It has been observed that certain requests taking more time than expected to
> complete hinders the performance of whole workload. Such requests are known
> as tailing requests. They can be taking more time due to a number of reasons
> and the prominent among them is a bad network connection. In Abfs driver we
> cache network connections and keeping such bad connections in cache and
> reusing them can be bad for perf.
> In this effort we try to identify such connections and close them so that new
> good connetions can be established and perf can be improved. There are two
> parts of this effort.
> # Identifying Tailing Requests: This involves profiling all the network
> calls and getting percentiles value optimally. By default we consider p99 as
> the tail latency and all the future requests taking more than tail latency
> will be considere as Tailing requests.
> # Proactively Killing Socket Connections: With Apache client, we can now
> kill the socket connection and fail the tailing request. Such failures will
> not be thrown back to user and retried immediately without any sleep but from
> another socket connection.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
