org.apache.hadoop.hdfs.server.namenode.top.window.RollingWindowManager Maven / Gradle / Ivy
/**
* 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.hdfs.server.namenode.top.window;
import java.util.Iterator;
import java.util.List;
import java.util.Map;
import java.util.Set;
import java.util.Stack;
import java.util.concurrent.ConcurrentHashMap;
import com.google.common.base.Preconditions;
import com.google.common.collect.Lists;
import org.apache.hadoop.classification.InterfaceAudience;
import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.hdfs.DFSConfigKeys;
import org.apache.hadoop.metrics2.util.Metrics2Util.NameValuePair;
import org.apache.hadoop.metrics2.util.Metrics2Util.TopN;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
/**
* A class to manage the set of {@link RollingWindow}s. This class is the
* interface of metrics system to the {@link RollingWindow}s to retrieve the
* current top metrics.
*
* Thread-safety is provided by each {@link RollingWindow} being thread-safe as
* well as {@link ConcurrentHashMap} for the collection of them.
*/
@InterfaceAudience.Private
public class RollingWindowManager {
public static final Logger LOG = LoggerFactory.getLogger(
RollingWindowManager.class);
private final int windowLenMs;
private final int bucketsPerWindow; // e.g., 10 buckets per minute
private final int topUsersCnt; // e.g., report top 10 metrics
static private class RollingWindowMap extends
ConcurrentHashMap {
private static final long serialVersionUID = -6785807073237052051L;
}
/**
* Represents a snapshot of the rolling window. It contains one Op per
* operation in the window, with ranked users for each Op.
*/
public static class TopWindow {
private final int windowMillis;
private final List top;
public TopWindow(int windowMillis) {
this.windowMillis = windowMillis;
this.top = Lists.newArrayList();
}
public void addOp(Op op) {
top.add(op);
}
public int getWindowLenMs() {
return windowMillis;
}
public List getOps() {
return top;
}
}
/**
* Represents an operation within a TopWindow. It contains a ranked
* set of the top users for the operation.
*/
public static class Op {
private final String opType;
private final List topUsers;
private final long totalCount;
public Op(String opType, long totalCount) {
this.opType = opType;
this.topUsers = Lists.newArrayList();
this.totalCount = totalCount;
}
public void addUser(User u) {
topUsers.add(u);
}
public String getOpType() {
return opType;
}
public List getTopUsers() {
return topUsers;
}
public long getTotalCount() {
return totalCount;
}
}
/**
* Represents a user who called an Op within a TopWindow. Specifies the
* user and the number of times the user called the operation.
*/
public static class User {
private final String user;
private final long count;
public User(String user, long count) {
this.user = user;
this.count = count;
}
public String getUser() {
return user;
}
public long getCount() {
return count;
}
}
/**
* A mapping from each reported metric to its {@link RollingWindowMap} that
* maintains the set of {@link RollingWindow}s for the users that have
* operated on that metric.
*/
public ConcurrentHashMap metricMap =
new ConcurrentHashMap();
public RollingWindowManager(Configuration conf, int reportingPeriodMs) {
windowLenMs = reportingPeriodMs;
bucketsPerWindow =
conf.getInt(DFSConfigKeys.NNTOP_BUCKETS_PER_WINDOW_KEY,
DFSConfigKeys.NNTOP_BUCKETS_PER_WINDOW_DEFAULT);
Preconditions.checkArgument(bucketsPerWindow > 0,
"a window should have at least one bucket");
Preconditions.checkArgument(bucketsPerWindow <= windowLenMs,
"the minimum size of a bucket is 1 ms");
//same-size buckets
Preconditions.checkArgument(windowLenMs % bucketsPerWindow == 0,
"window size must be a multiplication of number of buckets");
topUsersCnt =
conf.getInt(DFSConfigKeys.NNTOP_NUM_USERS_KEY,
DFSConfigKeys.NNTOP_NUM_USERS_DEFAULT);
Preconditions.checkArgument(topUsersCnt > 0,
"the number of requested top users must be at least 1");
}
/**
* Called when the metric command is changed by "delta" units at time "time"
* via user "user"
*
* @param time the time of the event
* @param command the metric that is updated, e.g., the operation name
* @param user the user that updated the metric
* @param delta the amount of change in the metric, e.g., +1
*/
public void recordMetric(long time, String command,
String user, long delta) {
RollingWindow window = getRollingWindow(command, user);
window.incAt(time, delta);
}
/**
* Take a snapshot of current top users in the past period.
*
* @param time the current time
* @return a TopWindow describing the top users for each metric in the
* window.
*/
public TopWindow snapshot(long time) {
TopWindow window = new TopWindow(windowLenMs);
Set metricNames = metricMap.keySet();
LOG.debug("iterating in reported metrics, size={} values={}",
metricNames.size(), metricNames);
for (Map.Entry entry : metricMap.entrySet()) {
String metricName = entry.getKey();
RollingWindowMap rollingWindows = entry.getValue();
TopN topN = getTopUsersForMetric(time, metricName, rollingWindows);
final int size = topN.size();
if (size == 0) {
continue;
}
Op op = new Op(metricName, topN.getTotal());
window.addOp(op);
// Reverse the users from the TopUsers using a stack,
// since we'd like them sorted in descending rather than ascending order
Stack reverse = new Stack();
for (int i = 0; i < size; i++) {
reverse.push(topN.poll());
}
for (int i = 0; i < size; i++) {
NameValuePair userEntry = reverse.pop();
User user = new User(userEntry.getName(), userEntry.getValue());
op.addUser(user);
}
}
return window;
}
/**
* Calculates the top N users over a time interval.
*
* @param time the current time
* @param metricName Name of metric
* @return
*/
private TopN getTopUsersForMetric(long time, String metricName,
RollingWindowMap rollingWindows) {
TopN topN = new TopN(topUsersCnt);
Iterator> iterator =
rollingWindows.entrySet().iterator();
while (iterator.hasNext()) {
Map.Entry entry = iterator.next();
String userName = entry.getKey();
RollingWindow aWindow = entry.getValue();
long windowSum = aWindow.getSum(time);
// do the gc here
if (windowSum == 0) {
LOG.debug("gc window of metric: {} userName: {}",
metricName, userName);
iterator.remove();
continue;
}
LOG.debug("offer window of metric: {} userName: {} sum: {}",
metricName, userName, windowSum);
topN.offer(new NameValuePair(userName, windowSum));
}
LOG.debug("topN users size for command {} is: {}",
metricName, topN.size());
return topN;
}
/**
* Get the rolling window specified by metric and user.
*
* @param metric the updated metric
* @param user the user that updated the metric
* @return the rolling window
*/
private RollingWindow getRollingWindow(String metric, String user) {
RollingWindowMap rwMap = metricMap.get(metric);
if (rwMap == null) {
rwMap = new RollingWindowMap();
RollingWindowMap prevRwMap = metricMap.putIfAbsent(metric, rwMap);
if (prevRwMap != null) {
rwMap = prevRwMap;
}
}
RollingWindow window = rwMap.get(user);
if (window != null) {
return window;
}
window = new RollingWindow(windowLenMs, bucketsPerWindow);
RollingWindow prevWindow = rwMap.putIfAbsent(user, window);
if (prevWindow != null) {
window = prevWindow;
}
return window;
}
}