Madan Jampani | a14047d | 2015-02-25 12:23:02 -0800 | [diff] [blame] | 1 | /* |
Brian O'Connor | a09fe5b | 2017-08-03 21:12:30 -0700 | [diff] [blame] | 2 | * Copyright 2015-present Open Networking Foundation |
Madan Jampani | a14047d | 2015-02-25 12:23:02 -0800 | [diff] [blame] | 3 | * |
| 4 | * Licensed under the Apache License, Version 2.0 (the "License"); |
| 5 | * you may not use this file except in compliance with the License. |
| 6 | * You may obtain a copy of the License at |
| 7 | * |
| 8 | * http://www.apache.org/licenses/LICENSE-2.0 |
| 9 | * |
| 10 | * Unless required by applicable law or agreed to in writing, software |
| 11 | * distributed under the License is distributed on an "AS IS" BASIS, |
| 12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 13 | * See the License for the specific language governing permissions and |
| 14 | * limitations under the License. |
| 15 | */ |
Madan Jampani | bd6845d | 2015-02-25 11:43:48 -0800 | [diff] [blame] | 16 | package org.onosproject.store.cluster.impl; |
| 17 | |
| 18 | import static com.google.common.base.Preconditions.checkArgument; |
| 19 | import static com.google.common.base.Preconditions.checkNotNull; |
| 20 | |
| 21 | import java.util.Map; |
| 22 | |
| 23 | import org.apache.commons.math3.stat.descriptive.DescriptiveStatistics; |
| 24 | import org.onosproject.cluster.NodeId; |
| 25 | |
| 26 | import com.google.common.collect.Maps; |
| 27 | |
| 28 | /** |
| 29 | * Phi Accrual failure detector. |
| 30 | * <p> |
| 31 | * Based on a paper titled: "The φ Accrual Failure Detector" by Hayashibara, et al. |
| 32 | */ |
| 33 | public class PhiAccrualFailureDetector { |
| 34 | private final Map<NodeId, History> states = Maps.newConcurrentMap(); |
| 35 | |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 36 | // Default value |
| 37 | private static final int DEFAULT_WINDOW_SIZE = 250; |
| 38 | private static final int DEFAULT_MIN_SAMPLES = 25; |
Jordan Halterman | 4a082ec | 2018-05-14 15:00:42 -0700 | [diff] [blame^] | 39 | private static final long DEFAULT_MIN_STANDARD_DEVIATION_MILLIS = 50; |
Madan Jampani | bd6845d | 2015-02-25 11:43:48 -0800 | [diff] [blame] | 40 | |
| 41 | // If a node does not have any heartbeats, this is the phi |
| 42 | // value to report. Indicates the node is inactive (from the |
| 43 | // detectors perspective. |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 44 | private static final double DEFAULT_BOOTSTRAP_PHI_VALUE = 100.0; |
| 45 | |
Jordan Halterman | 4a082ec | 2018-05-14 15:00:42 -0700 | [diff] [blame^] | 46 | private final int minSamples; |
| 47 | private final long minStandardDeviationMillis; |
| 48 | private final double bootstrapPhiValue = DEFAULT_BOOTSTRAP_PHI_VALUE; |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 49 | |
Jordan Halterman | 4a082ec | 2018-05-14 15:00:42 -0700 | [diff] [blame^] | 50 | public PhiAccrualFailureDetector() { |
| 51 | this(DEFAULT_MIN_SAMPLES, DEFAULT_MIN_STANDARD_DEVIATION_MILLIS); |
| 52 | } |
| 53 | |
| 54 | public PhiAccrualFailureDetector(long minStandardDeviationMillis) { |
| 55 | this(DEFAULT_MIN_SAMPLES, minStandardDeviationMillis); |
| 56 | } |
| 57 | |
| 58 | public PhiAccrualFailureDetector(int minSamples, long minStandardDeviationMillis) { |
| 59 | checkArgument(minSamples > 0, "minSamples must be positive"); |
| 60 | checkArgument(minStandardDeviationMillis > 0, "minStandardDeviationMillis must be positive"); |
| 61 | this.minSamples = minSamples; |
| 62 | this.minStandardDeviationMillis = minStandardDeviationMillis; |
| 63 | } |
Madan Jampani | bd6845d | 2015-02-25 11:43:48 -0800 | [diff] [blame] | 64 | |
| 65 | /** |
Jordan Halterman | e706253 | 2018-01-22 11:22:46 -0800 | [diff] [blame] | 66 | * Returns the last heartbeat time for the given node. |
| 67 | * |
| 68 | * @param nodeId the node identifier |
| 69 | * @return the last heartbeat time for the given node |
| 70 | */ |
| 71 | public long getLastHeartbeatTime(NodeId nodeId) { |
| 72 | History nodeState = states.computeIfAbsent(nodeId, key -> new History()); |
| 73 | return nodeState.latestHeartbeatTime(); |
| 74 | } |
| 75 | |
| 76 | /** |
Madan Jampani | bd6845d | 2015-02-25 11:43:48 -0800 | [diff] [blame] | 77 | * Report a new heart beat for the specified node id. |
| 78 | * @param nodeId node id |
| 79 | */ |
| 80 | public void report(NodeId nodeId) { |
| 81 | report(nodeId, System.currentTimeMillis()); |
| 82 | } |
| 83 | |
| 84 | /** |
| 85 | * Report a new heart beat for the specified node id. |
| 86 | * @param nodeId node id |
| 87 | * @param arrivalTime arrival time |
| 88 | */ |
| 89 | public void report(NodeId nodeId, long arrivalTime) { |
| 90 | checkNotNull(nodeId, "NodeId must not be null"); |
| 91 | checkArgument(arrivalTime >= 0, "arrivalTime must not be negative"); |
Jordan Halterman | 4a082ec | 2018-05-14 15:00:42 -0700 | [diff] [blame^] | 92 | History nodeState = states.computeIfAbsent(nodeId, key -> new History()); |
Madan Jampani | bd6845d | 2015-02-25 11:43:48 -0800 | [diff] [blame] | 93 | synchronized (nodeState) { |
| 94 | long latestHeartbeat = nodeState.latestHeartbeatTime(); |
| 95 | if (latestHeartbeat != -1) { |
| 96 | nodeState.samples().addValue(arrivalTime - latestHeartbeat); |
| 97 | } |
| 98 | nodeState.setLatestHeartbeatTime(arrivalTime); |
| 99 | } |
| 100 | } |
| 101 | |
Jordan Halterman | 1315d3e | 2018-01-16 19:41:31 -0800 | [diff] [blame] | 102 | /** |
| 103 | * Resets the failure detector for the given node. |
| 104 | * |
| 105 | * @param nodeId node identifier for the node for which to reset the failure detector |
| 106 | */ |
| 107 | public void reset(NodeId nodeId) { |
Jordan Halterman | e706253 | 2018-01-22 11:22:46 -0800 | [diff] [blame] | 108 | states.remove(nodeId); |
Jordan Halterman | 1315d3e | 2018-01-16 19:41:31 -0800 | [diff] [blame] | 109 | } |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 110 | |
Madan Jampani | bd6845d | 2015-02-25 11:43:48 -0800 | [diff] [blame] | 111 | /** |
| 112 | * Compute phi for the specified node id. |
| 113 | * @param nodeId node id |
| 114 | * @return phi value |
| 115 | */ |
Madan Jampani | a14047d | 2015-02-25 12:23:02 -0800 | [diff] [blame] | 116 | public double phi(NodeId nodeId) { |
| 117 | checkNotNull(nodeId, "NodeId must not be null"); |
Madan Jampani | bd6845d | 2015-02-25 11:43:48 -0800 | [diff] [blame] | 118 | if (!states.containsKey(nodeId)) { |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 119 | return bootstrapPhiValue; |
Madan Jampani | bd6845d | 2015-02-25 11:43:48 -0800 | [diff] [blame] | 120 | } |
Madan Jampani | bd6845d | 2015-02-25 11:43:48 -0800 | [diff] [blame] | 121 | History nodeState = states.get(nodeId); |
| 122 | synchronized (nodeState) { |
| 123 | long latestHeartbeat = nodeState.latestHeartbeatTime(); |
| 124 | DescriptiveStatistics samples = nodeState.samples(); |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 125 | if (latestHeartbeat == -1 || samples.getN() < minSamples) { |
Madan Jampani | bd6845d | 2015-02-25 11:43:48 -0800 | [diff] [blame] | 126 | return 0.0; |
| 127 | } |
| 128 | return computePhi(samples, latestHeartbeat, System.currentTimeMillis()); |
| 129 | } |
| 130 | } |
| 131 | |
| 132 | private double computePhi(DescriptiveStatistics samples, long tLast, long tNow) { |
Jordan Halterman | 4a082ec | 2018-05-14 15:00:42 -0700 | [diff] [blame^] | 133 | long elapsedTime = tNow - tLast; |
| 134 | double meanMillis = samples.getMean(); |
| 135 | double y = (elapsedTime - meanMillis) / Math.max(samples.getStandardDeviation(), minStandardDeviationMillis); |
| 136 | double e = Math.exp(-y * (1.5976 + 0.070566 * y * y)); |
| 137 | if (elapsedTime > meanMillis) { |
| 138 | return -Math.log10(e / (1.0 + e)); |
| 139 | } else { |
| 140 | return -Math.log10(1.0 - 1.0 / (1.0 + e)); |
| 141 | } |
Madan Jampani | bd6845d | 2015-02-25 11:43:48 -0800 | [diff] [blame] | 142 | } |
| 143 | |
| 144 | private static class History { |
Jordan Halterman | 4a082ec | 2018-05-14 15:00:42 -0700 | [diff] [blame^] | 145 | DescriptiveStatistics samples = new DescriptiveStatistics(DEFAULT_WINDOW_SIZE); |
Madan Jampani | bd6845d | 2015-02-25 11:43:48 -0800 | [diff] [blame] | 146 | long lastHeartbeatTime = -1; |
| 147 | |
Jordan Halterman | 4a082ec | 2018-05-14 15:00:42 -0700 | [diff] [blame^] | 148 | DescriptiveStatistics samples() { |
Madan Jampani | bd6845d | 2015-02-25 11:43:48 -0800 | [diff] [blame] | 149 | return samples; |
| 150 | } |
| 151 | |
Jordan Halterman | 4a082ec | 2018-05-14 15:00:42 -0700 | [diff] [blame^] | 152 | long latestHeartbeatTime() { |
Madan Jampani | bd6845d | 2015-02-25 11:43:48 -0800 | [diff] [blame] | 153 | return lastHeartbeatTime; |
| 154 | } |
| 155 | |
Jordan Halterman | 4a082ec | 2018-05-14 15:00:42 -0700 | [diff] [blame^] | 156 | void setLatestHeartbeatTime(long value) { |
Madan Jampani | bd6845d | 2015-02-25 11:43:48 -0800 | [diff] [blame] | 157 | lastHeartbeatTime = value; |
| 158 | } |
| 159 | } |
| 160 | } |