Github user jolynch commented on a diff in the pull request:
https://github.com/apache/cassandra/pull/283#discussion_r237958792
--- Diff: src/java/org/apache/cassandra/locator/dynamicsnitch/DynamicEndpointSnitchEMA.java
---
@@ -0,0 +1,133 @@
+/*
+ * 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.cassandra.locator.dynamicsnitch;
+
+import java.util.Collections;
+import java.util.HashMap;
+import java.util.Map;
+import java.util.Optional;
+
+import com.google.common.annotations.VisibleForTesting;
+
+import org.apache.cassandra.locator.DynamicEndpointSnitch;
+import org.apache.cassandra.locator.IEndpointSnitch;
+import org.apache.cassandra.locator.InetAddressAndPort;
+import org.apache.cassandra.metrics.ExponentialMovingAverage;
+
+
+/**
+ * A dynamic snitching implementation that uses Exponentially Moving Averages as a low
pass filter to prefer
+ * or de-prefer hosts
+ *
+ * This implementation generates a few orders of magnitude less garbage than histograms
and is close to 10x faster,
+ * but as it is not a Median LPF (it is an Average LPF), it is more vulnerable to noise.
This may be acceptable but
+ * given the significant change in behavior this is not the default in 4.0
+ */
+public class DynamicEndpointSnitchEMA extends DynamicEndpointSnitch
+{
+ // A ~10 sample EMA heavily weighted to the past values to minimize noise
+ private static final double EMA_ALPHA = 0.10;
+
+ protected static class EMASnitchMeasurement implements ISnitchMeasurement
+ {
+ public final ExponentialMovingAverage avg;
+
+ EMASnitchMeasurement(double initial)
+ {
+ avg = new ExponentialMovingAverage(EMA_ALPHA, initial);
+ }
+
+ @Override
+ public void sample(long value)
+ {
+ avg.update(value);
+ }
+
+ @Override
+ public double measure()
+ {
+ return avg.getAvg();
+ }
+
+ @Override
+ public Iterable<Double> measurements()
+ {
+ return Collections.singletonList(avg.getAvg());
+ }
+ }
+
+ // Called via reflection
+ public DynamicEndpointSnitchEMA(IEndpointSnitch snitch)
+ {
+ this(snitch, "ema");
+ }
+
+ public DynamicEndpointSnitchEMA(IEndpointSnitch snitch, String instance)
+ {
+ super(snitch, instance);
+ }
+
+ @Override
+ protected ISnitchMeasurement measurementImpl(long initialValue)
+ {
+ return new EMASnitchMeasurement(initialValue);
+ }
+
+ /**
+ * Unlike the Histogram implementation, calling this measure method is reasonably
cheap (doesn't require a
+ * Snapshot or anything) so we can skip a round of iterations and just normalize
the scores slightly
+ * differently
+ */
+ @Override
+ public Map<InetAddressAndPort, Double> calculateScores()
+ {
+ // We're going to weight the latency for each host against the worst one we see,
to
+ // arrive at sort of a 'badness percentage' for them. First, find the worst for
each:
+ HashMap<InetAddressAndPort, Double> newScores = new HashMap<>(samples.size());
+ Optional<Double> maxObservedAvgLatency = samples.values().stream()
--- End diff --
I went ahead and removed the EMA entirely, let's follow up on that in another change.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: pr-unsubscribe@cassandra.apache.org
For additional commands, e-mail: pr-help@cassandra.apache.org
|