NormalizationTransform.java
/*
* Copyright 2014 Netflix, Inc.
*
* Licensed 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 com.netflix.servo.publish;
import com.netflix.servo.DefaultMonitorRegistry;
import com.netflix.servo.Metric;
import com.netflix.servo.annotations.DataSourceType;
import com.netflix.servo.monitor.Counter;
import com.netflix.servo.monitor.MonitorConfig;
import com.netflix.servo.monitor.Monitors;
import com.netflix.servo.tag.TagList;
import com.netflix.servo.util.Clock;
import com.netflix.servo.util.ClockWithOffset;
import com.netflix.servo.util.Preconditions;
import com.netflix.servo.util.VisibleForTesting;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import java.util.ArrayList;
import java.util.LinkedHashMap;
import java.util.List;
import java.util.Map;
import java.util.concurrent.TimeUnit;
/**
* Converts rate metrics into normalized values. See
* <a href="http://www.vandenbogaerdt.nl/rrdtool/process.php">Rates, normalizing and
* consolidating</a> for a
* discussion on normalization of rates as done by rrdtool.
*/
public final class NormalizationTransform implements MetricObserver {
private static final Logger LOGGER =
LoggerFactory.getLogger(NormalizationTransform.class);
private static final String DEFAULT_DSTYPE = DataSourceType.RATE.name();
static Counter newCounter(String name) {
Counter c = Monitors.newCounter(name);
DefaultMonitorRegistry.getInstance().register(c);
return c;
}
@VisibleForTesting
static final Counter HEARTBEAT_EXCEEDED = newCounter("servo.norm.heartbeatExceeded");
private final MetricObserver observer;
private final long heartbeatMillis;
private final long stepMillis;
private final Map<MonitorConfig, NormalizedValue> cache;
/**
* Creates a new instance with the specified sampling and heartbeat interval using the default
* clock implementation.
*
* @param observer downstream observer to forward values after rates have been normalized
* to step boundaries
* @param step sampling interval in milliseconds
* @param heartbeat how long to keep values before dropping them and treating new samples
* as first report
* (in milliseconds)
* @deprecated Please use a constructor that specifies the the timeunit explicitly.
*/
@Deprecated
public NormalizationTransform(MetricObserver observer, long step, long heartbeat) {
this(observer, step, heartbeat, TimeUnit.MILLISECONDS, ClockWithOffset.INSTANCE);
}
/**
* Creates a new instance with the specified sampling and heartbeat interval and the
* specified clock implementation.
*
* @param observer downstream observer to forward values after rates have been normalized
* to step boundaries
* @param step sampling interval in milliseconds
* @param heartbeat how long to keep values before dropping them and treating new samples as
* first report (in milliseconds)
* @param clock The {@link com.netflix.servo.util.Clock} to use for getting
* the current time.
* @deprecated Please use a constructor that specifies the the timeunit explicitly.
*/
@Deprecated
public NormalizationTransform(MetricObserver observer, long step, final long heartbeat,
final Clock clock) {
this(observer, step, heartbeat, TimeUnit.MILLISECONDS, ClockWithOffset.INSTANCE);
}
/**
* Creates a new instance with the specified sampling and heartbeat interval and the
* specified clock implementation.
*
* @param observer downstream observer to forward values after rates have been normalized
* to step boundaries
* @param step sampling interval
* @param heartbeat how long to keep values before dropping them and treating new samples
* as first report
* @param unit {@link java.util.concurrent.TimeUnit} in which step and heartbeat
* are specified.
*/
public NormalizationTransform(MetricObserver observer, long step, long heartbeat,
TimeUnit unit) {
this(observer, step, heartbeat, unit, ClockWithOffset.INSTANCE);
}
/**
* Creates a new instance with the specified sampling and heartbeat interval and the specified
* clock implementation.
*
* @param observer downstream observer to forward values after rates have been normalized
* to step boundaries
* @param step sampling interval
* @param heartbeat how long to keep values before dropping them and treating new samples
* as first report
* @param unit The {@link java.util.concurrent.TimeUnit} in which step
* and heartbeat are specified.
* @param clock The {@link com.netflix.servo.util.Clock}
* to use for getting the current time.
*/
public NormalizationTransform(MetricObserver observer, long step, final long heartbeat,
TimeUnit unit, final Clock clock) {
this.observer = Preconditions.checkNotNull(observer, "observer");
Preconditions.checkArgument(step > 0, "step must be positive");
this.stepMillis = unit.toMillis(step);
Preconditions.checkArgument(heartbeat > 0, "heartbeat must be positive");
this.heartbeatMillis = unit.toMillis(heartbeat);
this.cache = new LinkedHashMap<MonitorConfig, NormalizedValue>(16, 0.75f, true) {
protected boolean removeEldestEntry(Map.Entry<MonitorConfig, NormalizedValue> eldest) {
final long lastMod = eldest.getValue().lastUpdateTime;
if (lastMod < 0) {
return false;
}
final long now = clock.now();
final boolean expired = (now - lastMod) > heartbeatMillis;
if (expired) {
HEARTBEAT_EXCEEDED.increment();
LOGGER.debug("heartbeat interval exceeded, expiring {}", eldest.getKey());
}
return expired;
}
};
}
private static String getDataSourceType(Metric m) {
final TagList tags = m.getConfig().getTags();
final String value = tags.getValue(DataSourceType.KEY);
if (value != null) {
return value;
} else {
return DEFAULT_DSTYPE;
}
}
private static boolean isGauge(String dsType) {
return dsType.equals(DataSourceType.GAUGE.name());
}
private static boolean isRate(String dsType) {
return dsType.equals(DataSourceType.RATE.name());
}
private static boolean isNormalized(String dsType) {
return dsType.equals(DataSourceType.NORMALIZED.name());
}
private static boolean isInformational(String dsType) {
return dsType.equals(DataSourceType.INFORMATIONAL.name());
}
private Metric normalize(Metric m, long stepBoundary) {
NormalizedValue normalizedValue = cache.get(m.getConfig());
if (normalizedValue == null) {
normalizedValue = new NormalizedValue();
cache.put(m.getConfig(), normalizedValue);
}
double value = normalizedValue.updateAndGet(m.getTimestamp(),
m.getNumberValue().doubleValue());
return new Metric(m.getConfig(), stepBoundary, value);
}
/**
* {@inheritDoc}
*/
@Override
public void update(List<Metric> metrics) {
Preconditions.checkNotNull(metrics, "metrics");
final List<Metric> newMetrics = new ArrayList<>(metrics.size());
for (Metric m : metrics) {
long offset = m.getTimestamp() % stepMillis;
long stepBoundary = m.getTimestamp() - offset;
String dsType = getDataSourceType(m);
if (isGauge(dsType) || isNormalized(dsType)) {
Metric atStepBoundary = new Metric(m.getConfig(), stepBoundary, m.getValue());
newMetrics.add(atStepBoundary); // gauges are not normalized
} else if (isRate(dsType)) {
Metric normalized = normalize(m, stepBoundary);
if (normalized != null) {
newMetrics.add(normalized);
}
} else if (!isInformational(dsType)) {
// unknown type - use a safe fallback
newMetrics.add(m); // we cannot normalize this
}
}
observer.update(newMetrics);
}
/**
* {@inheritDoc}
*/
@Override
public String getName() {
return observer.getName();
}
private static final long NO_PREVIOUS_UPDATE = -1L;
private class NormalizedValue {
private long lastUpdateTime = NO_PREVIOUS_UPDATE;
private double lastValue = 0.0;
private double weightedValue(long offset, double value) {
double weight = (double) offset / stepMillis;
return value * weight;
}
double updateAndGet(long timestamp, double value) {
double result = Double.NaN;
if (timestamp > lastUpdateTime) {
if (lastUpdateTime > 0 && timestamp - lastUpdateTime > heartbeatMillis) {
lastUpdateTime = NO_PREVIOUS_UPDATE;
lastValue = 0.0;
}
long offset = timestamp % stepMillis;
long stepBoundary = timestamp - offset;
if (lastUpdateTime < stepBoundary) {
if (lastUpdateTime != NO_PREVIOUS_UPDATE) {
long intervalOffset = lastUpdateTime % stepMillis;
lastValue += weightedValue(stepMillis - intervalOffset, value);
result = lastValue;
} else if (offset == 0) {
result = value;
} else {
result = weightedValue(stepMillis - offset, value);
}
lastValue = weightedValue(offset, value);
} else {
// Didn't cross step boundary, so update is more frequent than step
// and we just need to
// add in the weighted value
long intervalOffset = timestamp - lastUpdateTime;
lastValue += weightedValue(intervalOffset, value);
result = lastValue;
}
}
lastUpdateTime = timestamp;
return result;
}
}
}