[
https://issues.apache.org/jira/browse/HDFS-16949?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17703271#comment-17703271
]
ASF GitHub Bot commented on HDFS-16949:
---------------------------------------
mkuchenbecker commented on code in PR #5495:
URL: https://github.com/apache/hadoop/pull/5495#discussion_r1143648343
##########
hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/metrics2/lib/MutableInverseQuantiles.java:
##########
@@ -0,0 +1,102 @@
+/**
+ * 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.metrics2.lib;
+
+import org.apache.commons.lang3.StringUtils;
+import org.apache.hadoop.classification.InterfaceAudience;
+import org.apache.hadoop.classification.InterfaceStability;
+import org.apache.hadoop.classification.VisibleForTesting;
+import org.apache.hadoop.metrics2.MetricsInfo;
+import org.apache.hadoop.metrics2.util.Quantile;
+import org.apache.hadoop.metrics2.util.SampleQuantiles;
+import
org.apache.hadoop.thirdparty.com.google.common.util.concurrent.ThreadFactoryBuilder;
+import java.util.concurrent.Executors;
+import java.util.concurrent.ScheduledExecutorService;
+import java.util.concurrent.ScheduledFuture;
+import java.util.concurrent.TimeUnit;
+import static org.apache.hadoop.metrics2.lib.Interns.info;
+
+/**
+ * Watches a stream of long values, maintaining online estimates of specific
+ * quantiles with provably low error bounds. Inverse quantiles are meant for
+ * highly accurate low-percentile (e.g. 1st, 5th) latency metrics.
+ * InverseQuantiles are used for metrics where higher the value better it is.
+ * ( eg: data transfer rate ).
+ * The 1st percentile here corresponds to the 99th inverse percentile metric,
+ * 5th percentile to 95th and so on.
+ */
[email protected]
[email protected]
+public class MutableInverseQuantiles extends MutableQuantiles{
+
+ @VisibleForTesting
+ public static final Quantile[] INVERSE_QUANTILES = { new Quantile(0.50,
0.050),
Review Comment:
We should either invert the percentile, or report all percentiles. This only
changes the range without the list-order traversal change.
I would expect that this PR for "inverse quantiles" would report the P10 as
the P90, but not directly emit the P10. If we emit the P10 we should enhance
quantiles to emit all percents.
##########
hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/metrics2/lib/MutableInverseQuantiles.java:
##########
@@ -0,0 +1,102 @@
+/**
+ * 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.metrics2.lib;
+
+import org.apache.commons.lang3.StringUtils;
+import org.apache.hadoop.classification.InterfaceAudience;
+import org.apache.hadoop.classification.InterfaceStability;
+import org.apache.hadoop.classification.VisibleForTesting;
+import org.apache.hadoop.metrics2.MetricsInfo;
+import org.apache.hadoop.metrics2.util.Quantile;
+import org.apache.hadoop.metrics2.util.SampleQuantiles;
+import
org.apache.hadoop.thirdparty.com.google.common.util.concurrent.ThreadFactoryBuilder;
+import java.util.concurrent.Executors;
+import java.util.concurrent.ScheduledExecutorService;
+import java.util.concurrent.ScheduledFuture;
+import java.util.concurrent.TimeUnit;
+import static org.apache.hadoop.metrics2.lib.Interns.info;
+
+/**
+ * Watches a stream of long values, maintaining online estimates of specific
+ * quantiles with provably low error bounds. Inverse quantiles are meant for
+ * highly accurate low-percentile (e.g. 1st, 5th) latency metrics.
+ * InverseQuantiles are used for metrics where higher the value better it is.
+ * ( eg: data transfer rate ).
+ * The 1st percentile here corresponds to the 99th inverse percentile metric,
+ * 5th percentile to 95th and so on.
+ */
[email protected]
[email protected]
+public class MutableInverseQuantiles extends MutableQuantiles{
+
+ @VisibleForTesting
+ public static final Quantile[] INVERSE_QUANTILES = { new Quantile(0.50,
0.050),
+ new Quantile(0.25, 0.025), new Quantile(0.10, 0.010),
+ new Quantile(0.05, 0.005), new Quantile(0.01, 0.001) };
+
+ private ScheduledFuture<?> scheduledTask;
+
+ private static final ScheduledExecutorService SCHEDULAR = Executors
+ .newScheduledThreadPool(1, new ThreadFactoryBuilder().setDaemon(true)
+ .setNameFormat("MutableInverseQuantiles-%d").build());
+
+ /**
+ * Instantiates a new {@link MutableInverseQuantiles} for a metric that
rolls itself
+ * over on the specified time interval.
+ *
+ * @param name of the metric
+ * @param description long-form textual description of the metric
+ * @param sampleName type of items in the stream (e.g., "Ops")
+ * @param valueName type of the values
+ * @param interval rollover interval (in seconds) of the estimator
+ */
+ public MutableInverseQuantiles(String name, String description, String
sampleName,
+ String valueName, int interval) {
+ String ucName = StringUtils.capitalize(name);
+ String usName = StringUtils.capitalize(sampleName);
+ String uvName = StringUtils.capitalize(valueName);
+ String desc = StringUtils.uncapitalize(description);
+ String lsName = StringUtils.uncapitalize(sampleName);
+ String lvName = StringUtils.uncapitalize(valueName);
+
+ setNumInfo(info(ucName + "Num" + usName, String.format(
+ "Number of %s for %s with %ds interval", lsName, desc, interval)));
+ // Construct the MetricsInfos for the inverse quantiles, converting to
inverse percentiles
+ setQuantileInfos(INVERSE_QUANTILES.length);
+ String nameTemplate = ucName + "%dthInversePercentile" + uvName;
+ String descTemplate = "%d inverse percentile " + lvName + " with " +
interval
+ + " second interval for " + desc;
+ for (int i = 0; i < INVERSE_QUANTILES.length; i++) {
+ int inversePercentile = (int) (100 * (1 -
INVERSE_QUANTILES[i].quantile));
Review Comment:
I don't understand this. We are still inverting the quantile, but also
reporting P10?
##########
hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/metrics2/lib/MutableInverseQuantiles.java:
##########
@@ -0,0 +1,102 @@
+/**
+ * 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.metrics2.lib;
+
+import org.apache.commons.lang3.StringUtils;
+import org.apache.hadoop.classification.InterfaceAudience;
+import org.apache.hadoop.classification.InterfaceStability;
+import org.apache.hadoop.classification.VisibleForTesting;
+import org.apache.hadoop.metrics2.MetricsInfo;
+import org.apache.hadoop.metrics2.util.Quantile;
+import org.apache.hadoop.metrics2.util.SampleQuantiles;
+import
org.apache.hadoop.thirdparty.com.google.common.util.concurrent.ThreadFactoryBuilder;
+import java.util.concurrent.Executors;
+import java.util.concurrent.ScheduledExecutorService;
+import java.util.concurrent.ScheduledFuture;
+import java.util.concurrent.TimeUnit;
+import static org.apache.hadoop.metrics2.lib.Interns.info;
+
+/**
+ * Watches a stream of long values, maintaining online estimates of specific
+ * quantiles with provably low error bounds. Inverse quantiles are meant for
+ * highly accurate low-percentile (e.g. 1st, 5th) latency metrics.
+ * InverseQuantiles are used for metrics where higher the value better it is.
+ * ( eg: data transfer rate ).
+ * The 1st percentile here corresponds to the 99th inverse percentile metric,
+ * 5th percentile to 95th and so on.
+ */
[email protected]
[email protected]
+public class MutableInverseQuantiles extends MutableQuantiles{
+
+ @VisibleForTesting
+ public static final Quantile[] INVERSE_QUANTILES = { new Quantile(0.50,
0.050),
+ new Quantile(0.25, 0.025), new Quantile(0.10, 0.010),
+ new Quantile(0.05, 0.005), new Quantile(0.01, 0.001) };
+
+ private ScheduledFuture<?> scheduledTask;
+
+ private static final ScheduledExecutorService SCHEDULAR = Executors
+ .newScheduledThreadPool(1, new ThreadFactoryBuilder().setDaemon(true)
+ .setNameFormat("MutableInverseQuantiles-%d").build());
+
+ /**
+ * Instantiates a new {@link MutableInverseQuantiles} for a metric that
rolls itself
+ * over on the specified time interval.
+ *
+ * @param name of the metric
+ * @param description long-form textual description of the metric
+ * @param sampleName type of items in the stream (e.g., "Ops")
+ * @param valueName type of the values
+ * @param interval rollover interval (in seconds) of the estimator
+ */
+ public MutableInverseQuantiles(String name, String description, String
sampleName,
+ String valueName, int interval) {
+ String ucName = StringUtils.capitalize(name);
Review Comment:
I would recommend two things going this route:
- Encapsulate any logic you will not add a test for in a common function
shared between implementations.
- Any logic that could be parametarized or overridden.
- Test all new logic added.
This function repeats much of the constructor from the superclass to supply
the inverse percentile. I would advocate for a DRY subclass (assuming we
subclass).
##########
hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/metrics2/lib/MutableInverseQuantiles.java:
##########
@@ -0,0 +1,102 @@
+/**
+ * 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.metrics2.lib;
+
+import org.apache.commons.lang3.StringUtils;
+import org.apache.hadoop.classification.InterfaceAudience;
+import org.apache.hadoop.classification.InterfaceStability;
+import org.apache.hadoop.classification.VisibleForTesting;
+import org.apache.hadoop.metrics2.MetricsInfo;
+import org.apache.hadoop.metrics2.util.Quantile;
+import org.apache.hadoop.metrics2.util.SampleQuantiles;
+import
org.apache.hadoop.thirdparty.com.google.common.util.concurrent.ThreadFactoryBuilder;
+import java.util.concurrent.Executors;
+import java.util.concurrent.ScheduledExecutorService;
+import java.util.concurrent.ScheduledFuture;
+import java.util.concurrent.TimeUnit;
+import static org.apache.hadoop.metrics2.lib.Interns.info;
+
+/**
+ * Watches a stream of long values, maintaining online estimates of specific
+ * quantiles with provably low error bounds. Inverse quantiles are meant for
+ * highly accurate low-percentile (e.g. 1st, 5th) latency metrics.
+ * InverseQuantiles are used for metrics where higher the value better it is.
+ * ( eg: data transfer rate ).
+ * The 1st percentile here corresponds to the 99th inverse percentile metric,
+ * 5th percentile to 95th and so on.
+ */
[email protected]
[email protected]
+public class MutableInverseQuantiles extends MutableQuantiles{
+
+ @VisibleForTesting
+ public static final Quantile[] INVERSE_QUANTILES = { new Quantile(0.50,
0.050),
+ new Quantile(0.25, 0.025), new Quantile(0.10, 0.010),
+ new Quantile(0.05, 0.005), new Quantile(0.01, 0.001) };
+
+ private ScheduledFuture<?> scheduledTask;
+
+ private static final ScheduledExecutorService SCHEDULAR = Executors
+ .newScheduledThreadPool(1, new ThreadFactoryBuilder().setDaemon(true)
+ .setNameFormat("MutableInverseQuantiles-%d").build());
+
+ /**
+ * Instantiates a new {@link MutableInverseQuantiles} for a metric that
rolls itself
+ * over on the specified time interval.
+ *
+ * @param name of the metric
+ * @param description long-form textual description of the metric
+ * @param sampleName type of items in the stream (e.g., "Ops")
+ * @param valueName type of the values
+ * @param interval rollover interval (in seconds) of the estimator
+ */
+ public MutableInverseQuantiles(String name, String description, String
sampleName,
Review Comment:
Its unclear to me why the constructor would need to change as compared to
changing how we iterate during the calculation.
Why is not calling `super` here sufficient?
> Update ReadTransferRate to ReadLatencyPerGB for effective percentile metrics
> ----------------------------------------------------------------------------
>
> Key: HDFS-16949
> URL: https://issues.apache.org/jira/browse/HDFS-16949
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: datanode
> Reporter: Ravindra Dingankar
> Assignee: Ravindra Dingankar
> Priority: Minor
> Labels: pull-request-available
> Fix For: 3.3.0, 3.4.0
>
>
> HDFS-16917 added ReadTransferRate quantiles to calculate the rate which data
> is read per unit of time.
> With percentiles the values are sorted in ascending order and hence for the
> transfer rate p90 gives us the value where 90 percent rates are lower
> (worse), p99 gives us the value where 99 percent values are lower (worse).
> Note that value(p90) < p(99) thus p99 is a better transfer rate as compared
> to p90.
> However as the percentile increases the value should become worse in order to
> know how good our system is.
> Hence instead of calculating the data read transfer rate, we should calculate
> it's inverse. We will instead calculate the time taken for a GB of data to be
> read. ( seconds / GB )
> After this the p90 value will give us 90 percentage of total values where the
> time taken is less than value(p90), similarly for p99 and others.
> Also p(90) < p(99) and here p(99) will become a worse value (taking more time
> each byte) as compared to p(90)
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
