kafka-storm-starter

Code examples that show how to integrate Apache Kafka 0.8+ with Apache Storm 0.9+ and Apache Spark 1.1+ while using Apache Avro as the data serialization format.

A great alternative to the examples in this repository, which require you to operate a Spark or Storm processing cluster: build elastic, distributed, fault-tolerant stream processing applications with Kafka's Streams API (read: no additional cluster required)

"Kafka Streams (source code), a component of open source Apache Kafka, is a powerful, easy-to-use library for building highly scalable, fault-tolerant, distributed stream processing applications on top of Apache Kafka. It builds upon important concepts for stream processing such as properly distinguishing between event-time and processing-time, handling of late-arriving data, and efficient management of application state."

Take a look at the Kafka Streams code examples at https://github.com/confluentinc/examples.

Table of Contents

Quick start
Features
Implementation details
Development
FAQ
- Kafka
- Storm
Known issues and limitations
- Upstream code
- kafka-storm-starter code
Change log
Contributing
License
References
- Wirbelsturm
- Kafka
- Storm
- Avro
- Kryo

Quick start

Show me!

$ ./sbt test

This command launches our test suite.

Notably it will run end-to-end tests of Kafka, Storm, and Kafka/Storm as well as Kafka/Spark Streaming integration. See this abridged version of the test output:

[...other tests removed...]

[info] KafkaSpec:
[info] Kafka
[info] - should synchronously send and receive a Tweet in Avro format
[info]   + Given a ZooKeeper instance
[info]   + And a Kafka broker instance
[info]   + And some tweets
[info]   + And a single-threaded Kafka consumer group
[info]   + When I start a synchronous Kafka producer that sends the tweets in Avro binary format
[info]   + Then the consumer app should receive the tweets
[info] - should asynchronously send and receive a Tweet in Avro format
[info]   + Given a ZooKeeper instance
[info]   + And a Kafka broker instance
[info]   + And some tweets
[info]   + And a single-threaded Kafka consumer group
[info]   + When I start an asynchronous Kafka producer that sends the tweets in Avro binary format
[info]   + Then the consumer app should receive the tweets
[info] StormSpec:
[info] Storm
[info] - should start a local cluster
[info]   + Given no cluster
[info]   + When I start a LocalCluster instance
[info]   + Then the local cluster should start properly
[info] - should run a basic topology
[info]   + Given a local cluster
[info]   + And a wordcount topology
[info]   + And the input words alice, bob, joe, alice
[info]   + When I submit the topology
[info]   + Then the topology should properly count the words
[info] KafkaStormSpec:
[info] As a user of Storm
[info] I want to read Avro-encoded data from Kafka
[info] so that I can quickly build Kafka<->Storm data flows
[info] Feature: AvroDecoderBolt[T]
[info]   Scenario: User creates a Storm topology that uses AvroDecoderBolt
[info]     Given a ZooKeeper instance
[info]     And a Kafka broker instance
[info]     And a Storm topology that uses AvroDecoderBolt and that reads tweets from topic testing-input} and writes them as-is to topic testing-output
[info]     And some tweets
[info]     And a synchronous Kafka producer app that writes to the topic testing-input
[info]     And a single-threaded Kafka consumer app that reads from topic testing-output and Avro-decodes the incoming data
[info]     And a Storm topology configuration that registers an Avro Kryo decorator for Tweet
[info]     When I run the Storm topology
[info]     And I Avro-encode the tweets and use the Kafka producer app to sent them to Kafka
[info]     Synchronously sending Tweet {"username": "ANY_USER_1", "text": "ANY_TEXT_1", "timestamp": 1411993272} to topic Some(testing-input)
[info]     Synchronously sending Tweet {"username": "ANY_USER_2", "text": "ANY_TEXT_2", "timestamp": 0} to topic Some(testing-input)
[info]     Synchronously sending Tweet {"username": "ANY_USER_3", "text": "ANY_TEXT_3", "timestamp": 1234} to topic Some(testing-input)
[info]     Then the Kafka consumer app should receive the original tweets from the Storm topology
[info] Feature: AvroScheme[T] for Kafka spout
[info]   Scenario: User creates a Storm topology that uses AvroScheme in Kafka spout
[info]     Given a ZooKeeper instance
[info]     And a Kafka broker instance
[info]     And a Storm topology that uses AvroScheme and that reads tweets from topic testing-input and writes them as-is to topic testing-output
[info]     And some tweets
[info]     And a synchronous Kafka producer app that writes to the topic testing-input
[info]     And a single-threaded Kafka consumer app that reads from topic testing-output and Avro-decodes the incoming data
[info]     And a Storm topology configuration that registers an Avro Kryo decorator for Tweet
[info]     When I run the Storm topology
[info]     And I Avro-encode the tweets and use the Kafka producer app to sent them to Kafka
[info]     Synchronously sending Tweet {"username": "ANY_USER_1", "text": "ANY_TEXT_1", "timestamp": 1411993272} to topic Some(testing-input)
[info]     Synchronously sending Tweet {"username": "ANY_USER_2", "text": "ANY_TEXT_2", "timestamp": 0} to topic Some(testing-input)
[info]     Synchronously sending Tweet {"username": "ANY_USER_3", "text": "ANY_TEXT_3", "timestamp": 1234} to topic Some(testing-input)
[info]     Then the Kafka consumer app should receive the original tweets from the Storm topology
[info] KafkaSparkStreamingSpec:
[info] As a user of Spark Streaming
[info] I want to read Avro-encoded data from Kafka
[info] so that I can quickly build Kafka<->Spark Streaming data flows
[info] Feature: Basic functionality
[info]   Scenario: User creates a Spark Streaming job that reads from and writes to Kafka
[info]     Given a ZooKeeper instance
[info]     And a Kafka broker instance
[info]     And some tweets
[info]     And a synchronous Kafka producer app that writes to the topic KafkaTopic(testing-input,1,1,{})
[info]     And a single-threaded Kafka consumer app that reads from topic KafkaTopic(testing-output,1,1,{}) and Avro-decodes the incoming data
[info]     When I Avro-encode the tweets and use the Kafka producer app to sent them to Kafka
[info]     And I run a streaming job that reads tweets from topic KafkaTopic(testing-input,1,1,{}) and writes them as-is to topic KafkaTopic(testing-output,1,1,{})
[info]     Then the Spark Streaming job should consume all tweets from Kafka
[info]     And the job should write back all tweets to Kafka
[info]     And the Kafka consumer app should receive the original tweets from the Spark Streaming job
[info] Run completed in 45 seconds, 787 milliseconds.
[info] Total number of tests run: 27
[info] Suites: completed 9, aborted 0
[info] Tests: succeeded 27, failed 0, canceled 0, ignored 0, pending 0
[info] All tests passed.

Show me one more time!

$ ./sbt run

This command launches KafkaStormDemo. This demo starts in-memory instances of ZooKeeper, Kafka, and Storm. It then runs a demo Storm topology that connects to and reads from the Kafka instance.

You will see output similar to the following (some parts removed to improve readability):

7031 [Thread-19] INFO  backtype.storm.daemon.worker - Worker 3f7f1a51-5c9e-43a5-b431-e39a7272215e for storm kafka-storm-starter-1-1400839826 on daa60807-d440-4b45-94fc-8dd7798453d2:1027 has finished loading
7033 [Thread-29-kafka-spout] INFO  storm.kafka.DynamicBrokersReader - Read partition info from zookeeper: GlobalPartitionInformation{partitionMap={0=127.0.0.1:9092}}
7050 [Thread-29-kafka-spout] INFO  backtype.storm.daemon.executor - Opened spout kafka-spout:(1)
7051 [Thread-29-kafka-spout] INFO  backtype.storm.daemon.executor - Activating spout kafka-spout:(1)
7051 [Thread-29-kafka-spout] INFO  storm.kafka.ZkCoordinator - Refreshing partition manager connections
7065 [Thread-29-kafka-spout] INFO  storm.kafka.DynamicBrokersReader - Read partition info from zookeeper: GlobalPartitionInformation{partitionMap={0=127.0.0.1:9092}}
7066 [Thread-29-kafka-spout] INFO  storm.kafka.ZkCoordinator - Deleted partition managers: []
7066 [Thread-29-kafka-spout] INFO  storm.kafka.ZkCoordinator - New partition managers: [Partition{host=127.0.0.1:9092, partition=0}]
7083 [Thread-29-kafka-spout] INFO  storm.kafka.PartitionManager - Read partition information from: /kafka-spout/kafka-storm-starter/partition_0  --> null
7100 [Thread-29-kafka-spout] INFO  storm.kafka.PartitionManager - No partition information found, using configuration to determine offset
7105 [Thread-29-kafka-spout] INFO  storm.kafka.PartitionManager - Starting Kafka 127.0.0.1:0 from offset 18
7106 [Thread-29-kafka-spout] INFO  storm.kafka.ZkCoordinator - Finished refreshing

At this point Storm is connected to Kafka (more precisely: to the testing topic in Kafka). Not much will happen afterwards because a) we are not sending any data to the Kafka topic and b) this demo Storm topology only reads from the Kafka topic but it does nothing to the data that was read.

Note that this example will actually run two in-memory instances of ZooKeeper: the first (listening at 127.0.0.1:2181/tcp) is used by the Kafka instance, the second (listening at 127.0.0.1:2000/tcp) is automatically started and used by the in-memory Storm cluster. This is because, when running in local aka in-memory mode, Storm versions < 0.9.3 do not allow you to reconfigure or disable its own ZooKeeper instance (see the Storm FAQ below for further information).

To stop the demo application you must kill or Ctrl-C the process in the terminal.

You can use KafkaStormDemo as a starting point to create your own, "real" Storm topologies that read from a "real" Kafka, Storm, and ZooKeeper infrastructure. An easy way to get started with such an infrastructure is by deploying Kafka, Storm, and ZooKeeper via a tool such as Wirbelsturm.

Features

What features do we showcase in kafka-storm-starter? Note that we focus on showcasing, and not necessarily on "production ready".

How to integrate Kafka and Storm as well as Kafka and Spark Streaming
How to use Avro with Kafka, Storm, and Spark Streaming.
Kafka standalone code examples
- KafkaProducerApp: A simple Kafka producer app for writing Avro-encoded data into Kafka. KafkaSpec puts this producer to use and shows how to use Twitter Bijection to Avro-encode the messages being sent to Kafka.
- KafkaConsumerApp: A simple Kafka consumer app for reading Avro-encoded data from Kafka. KafkaSpec puts this consumer to use and shows how to use Twitter Bijection to Avro-decode the messages being read from Kafka.
Storm standalone code examples
- AvroDecoderBolt[T]: An AvroDecoderBolt[T <: org.apache.avro.specific.SpecificRecordBase] that can be parameterized with the type of the Avro record T it will deserialize its data to (i.e. no need to write another decoder bolt just because the bolt needs to handle a different Avro schema).
- AvroScheme[T]: An AvroScheme[T <: org.apache.avro.specific.SpecificRecordBase] scheme, i.e. a custom backtype.storm.spout.Scheme to auto-deserialize a spout's incoming data. The scheme can be parameterized with the type of the Avro record T it will deserializes its data to (i.e. no need to write another scheme just because the scheme needs to handle a different Avro schema).
  - You can opt to configure a spout (such as the Kafka spout) with AvroScheme if you want to perform the Avro decoding step directly in the spout instead of placing an AvroDecoderBolt after the Kafka spout. You may want to profile your topology which of the two approaches works best for your use case.
- TweetAvroKryoDecorator: A custom backtype.storm.serialization.IKryoDecorator, i.e. a custom Kryo serializer for Storm.
  - Unfortunately we have not figured out a way to implement a parameterized AvroKryoDecorator[T] variant yet. (A "straight-forward" approach we tried -- similar to the other parameterized components -- compiled fine but failed at runtime when running the tests). Code contributions are welcome!
Kafka and Storm integration
- AvroKafkaSinkBolt[T]: An AvroKafkaSinkBolt[T <: org.apache.avro.specific.SpecificRecordBase] that can be parameterized with the type of the Avro record T it will serialize its data to before sending the encoded data to Kafka (i.e. no need to write another Kafka sink bolt just because the bolt needs to handle a different Avro schema).
- Storm topologies that read Avro-encoded data from Kafka: KafkaStormDemo and KafkaStormSpec
- A Storm topology that writes Avro-encoded data to Kafka: KafkaStormSpec
Kafka and Spark Streaming integration
- KafkaSparkStreamingSpec a streaming job that reads input data from Kafka and writes output data to Kafka. It demonstrates how to read from all partitions of a topic in parallel, how to decouple the downstream parallelism from the number of parttions (think: use 20 "threads" for processing the Kafka data even though the Kafka topic has only 5 partitions), and how to write the output of the streaming job back into Kafka. The input and output data is in Avro format, and we use Twitter Bijection for the serialization work. See my blog post on Integrating Kafka and Spark Streaming for further details.
Unit testing
- AvroDecoderBoltSpec
- AvroSchemeSpec
- And more under src/test/scala.
Integration testing
- KafkaSpec: Tests for Kafka, which launch and run against in-memory instances of Kafka and ZooKeeper. See EmbeddedKafkaZooKeeperCluster and its constituents KafkaEmbedded and ZooKeeperEmbedded.
- StormSpec: Tests for Storm, which launch and run against in-memory instances of Storm and ZooKeeper.
- KafkaStormSpec: Tests for integrating Storm and Kafka, which launch and run against in-memory instances of Kafka, Storm, and ZooKeeper.
- KafkaSparkStreamingSpec: Tests for integrating Spark Streaming and Kafka, which launch and run against in-memory instances of Kafka, Spark Streaming, and ZooKeeper.

Implementation details

We use Twitter Bijection for Avro encoding and decoding.
We use Twitter Chill (which in turn uses Bijection) to implement a custom Kryo serializer for Storm that handles our Avro-derived Java class Tweet from twitter.avsc.
Unit and integration tests are implemented with ScalaTest.
We use ZooKeeper 3.4.5.
We use the official Kafka spout of the Storm project, which is compatible with Kafka 0.8.

Development

Git setup: git-flow

This project follows the git-flow approach. This means, for instance, that:

The branch develop is used for integration of the "next release".
The branch master is used for bringing forth production releases.

Follow the git-flow installation instructions for your development machine.

See git-flow and the introduction article Why aren't you using git-flow? for details.

Build requirements

Your development machine requires:

Oracle JDK or OpenJDK for Java 7 (Oracle JDK preferred).

This project also needs Scala 2.10.4 and sbt 0.13.2, but these will be automatically downloaded and made available (locally/sandboxed) to the project as part of the build setup.

Building the code

$ ./sbt clean compile

If you want to only (re)generate Java classes from Avro schemas:

$ ./sbt avro:generate

Generated Java sources are stored under target/scala-*/src_managed/main/compiled_avro/.

Running the tests

$ ./sbt clean test

Here are some examples that demonstrate how you can run only a certain subset of tests:

# Use `-l` to exclude tests by tag:
# Run all tests WITH THE EXCEPTION of those tagged as integration tests
$ ./sbt "test-only * -- -l com.miguno.kafkastorm.integration.IntegrationTest"

# Use `-n` to include tests by tag (and skip all tests that lack the tag):
# Run ONLY tests tagged as integration tests
$ ./sbt "test-only * -- -n com.miguno.kafkastorm.integration.IntegrationTest"

# Run only the tests in suite AvroSchemeSpec:
$ ./sbt "test-only com.miguno.kafkastorm.storm.serialization.AvroSchemeSpec"

# You can also combine the examples above, of course.

Test reports in JUnit XML format are written to target/test-reports/junitxml/*.xml. Make sure that your actual build steps run the ./sbt test task, otherwise the JUnit XML reports will not be generate (note that ./sbt scoverage:test will not generate the JUnit XML reports unfortunately).

Integration with CI servers:

Jenkins integration:
- Configure the build job.
- Go to Post-build Actions.
- Add a post-build action for Publish JUnit test result report.
- In the Test report XMLs field add the pattern **/target/test-reports/junitxml/*.xml.
- Now each build of your job will have a Test Result link.
TeamCity integration:
- Edit the build configuration.
- Select configuration step 3, Build steps.
- Under Additional Build Features add a new build feature.
- Use the following build feature configuration:
  - Report type: Ant JUnit
  - Monitoring rules: target/test-reports/junitxml/*.xml
- Now each build of your job will have a Tests tab.

Further details are available at:

How to tag tests in ScalaTest: Tagging your tests
How to selectively run tests: Using ScalaTest with sbt and How to Run Tagged Scala Tests with SBT and ScalaTest

Creating code coverage reports

We are using sbt-scoverage to create code coverage reports for unit tests.

Run the unit tests via:

$ ./sbt clean scoverage:test

An HTML report will be created at target/scala-2.10/scoverage-report/index.html.
XML reports will be created at:
- ./target/scala-2.10/coverage-report/cobertura.xml
- ./target/scala-2.10/scoverage-report/scoverage.xml

Integration with CI servers:

Jenkins integration:
- Configure the build.
- Go to Post-build Actions.
- Add a post-build action for Publish Cobertura Coverage Report.
- In the Cobertura xml report pattern field add the pattern **/target/scala-2.10/coverage-report/cobertura.xml.
- Now each build of your job will have a Coverage Report link.
TeamCity integration:
- Edit the build configuration.
- Select configuration step 1, General settings.
- In the Artifact Paths field add the path target/scala-2.10/scoverage-report/** => coberturareport/.
- Now each build of your job will have a Cobertura Coverage Report tab.

Packaging the code

To create a normal ("slim") jar:

$ ./sbt clean package

>>> Generates `target/scala-2.10/kafka-storm-starter_2.10-0.2.0-SNAPSHOT.jar`

To create a fat jar, which includes any dependencies of kafka-storm-starter:

$ ./sbt assembly

>>> Generates `target/scala-2.10/kafka-storm-starter-assembly-0.2.0-SNAPSHOT.jar`

Note: By default, assembly by itself will NOT run any tests. If you want to run tests before assembly, chain sbt commands in sequence, e.g. ./sbt test assembly. See assembly.sbt` for details why we do this.

To create a scaladoc/javadoc jar:

$ ./sbt packageDoc

>>> Generates `target/scala-2.10/kafka-storm-starter_2.10-0.2.0-SNAPSHOT-javadoc.jar`

To create a sources jar:

$ ./sbt packageSrc

>>> Generates `target/scala-2.10/kafka-storm-starter_2.10-0.2.0-SNAPSHOT-sources.jar`

To create API docs:

$ ./sbt doc

>>> Generates `target/scala-2.10/api/*` (HTML files)

IDE support

IntelliJ IDEA

kafka-storm-starter integrates the sbt-idea plugin. Use the following command to build IDEA project files:

$ ./sbt gen-idea

You can then open kafka-storm-starter as a project in IDEA via File > Open... and selecting the top-level directory of kafka-storm-starter.

Important note: There is a bug when using the sbt plugins for Avro and for IntelliJ IDEA in combination. The sbt plugin for Avro reads the Avro *.avsc schemas stored under src/main/avro and generates the corresponding Java classes, which it stores under target/scala-2.10/src_managed/main/compiled_avro (in the case of kafka-storm-starter, a Tweet.java class will be generated from the Avro schema twitter.avsc). The latter path must be added to IDEA's Source Folders setting, which will happen automatically for you. However the aforementioned bug will add a second, incorrect path to Source Folders, too, which will cause IDEA to complain about not being able to find the Avro-generated Java classes (here: the Tweet class).

Until this bug is fixed upstream you can use the following workaround, which you must perform everytime you run ./sbt gen-idea:

In IntelliJ IDEA open the project structure for kafka-storm-starter via File > Project Structure....
Under Project settings on the left-hand side select Modules.
Select the Sources tab on the right-hand side.
Remove the problematic target/scala-2.10/src_managed/main/compiled_avro/com entry from the Source Folders listing (the source folders are colored in light-blue). Note the trailing .../com, which comes from com.miguno.avro.Tweet in the twitter.avsc Avro schema.
Click Ok.

See also this screenshot (click to enlarge):

Eclipse

kafka-storm-starter integrates the sbt-eclipse plugin. Use the following command to build Eclipse project files:

$ ./sbt eclipse

Then use the Import Wizard in Eclipse to import Existing Projects into Workspace.

FAQ

Kafka

ZooKeeper exceptions "KeeperException: NoNode for /[ZK path]" logged at INFO level

In short you can normally safely ignore those errors -- it's for a reason they are logged at INFO level and not at ERROR level.

As described in the mailing list thread Zookeeper exceptions:

"The reason you see those NoNode error code is the following. Every time we want to create a new [ZK] path, say /brokers/ids/1, we try to create it directly. If this fails because the parent path doesn't exist, we try to create the parent path first. This will happen recursively. However, the NoNode error should show up only once, not every time a broker is started (assuming ZK data hasn't been cleaned up)."

A similar answer was given in the thread Clean up kafka environment:

"These info messages show up when Kafka tries to create new consumer groups. While trying to create the children of /consumers/[group], if the parent path doesn't exist, the zookeeper server logs these messages. Kafka internally handles these cases correctly by first creating the parent node."

Storm

Storm `LocalCluster` and ZooKeeper

LocalCluster starts an embedded ZooKeeper instance listening at localhost:2000/tcp. If a different process is already bound to 2000/tcp, then Storm will increment the embedded ZooKeeper's port until it finds a free port (2000 -> 2001 -> 2002, and so on). LocalCluster then reads the Storm defaults and overrides some of Storm's configuration (see the mk-local-storm-cluster function in testing.clj and the mk-inprocess-zookeeper function in zookeeper.clj for details):

STORM-CLUSTER-MODE "local"
STORM-ZOOKEEPER-PORT zk-port
STORM-ZOOKEEPER-SERVERS ["localhost"]}

where zk-port is the final port chosen.

In Storm versions <= 0.9.2 it is not possible to launch a local Storm cluster via LocalCluster without its own embedded ZooKeeper. Likewise it is not possible to control on which port the embedded ZooKeeper process will listen -- it will always follow the 2000/tcp based algorithm above to set the port.

In Storm 0.9.3 and later you can configure LocalCluster to use a custom ZooKeeper instance, thanks to STORM-213.

Known issues and limitations

This section lists known issues and limitations a) for the upstream projects such as Storm and Kafka, and b) for our own code.

Upstream code

ZooKeeper throws InstanceAlreadyExistsException during tests

Note: We squelch this message during test runs. See log4j.properties.

You may see the following exception when running the integration tests, which you can safely ignore:

[2014-03-07 11:56:59,250] WARN Failed to register with JMX (org.apache.zookeeper.server.ZooKeeperServer)
javax.management.InstanceAlreadyExistsException: org.apache.ZooKeeperService:name0=StandaloneServer_port-1

The root cause is that in-memory ZooKeeper instances have a hardcoded JMX setup. And because we cannot prevent Storm's LocalCluster to start its own ZooKeeper instance alongside "ours" (see FAQ section above), there will be two ZK instances trying to use the same JMX setup. Since the JMX setup is not relevant for our testing the exception can be safely ignored, albeit we'd prefer to come up with a proper fix, of course.

See also ZOOKEEPER-1350: Make JMX registration optional in LearnerZooKeeperServer, which will make it possible to disable JMX registration when using Curator's TestServer to run an in-memory ZooKeeper instance (this patch will be included in ZooKeeper 3.5.0, see JIRA ticket above).

ZooKeeper version 3.3.4 recommended for use with Kafka 0.8

At the time of writing Kafka 0.8 is not officially compatible with ZooKeeper 3.4.x, which is the latest stable version of ZooKeeper. Instead the Kafka project recommends ZooKeeper 3.3.4.

So which version of ZooKeeper should you do pick, particularly if you are already running a ZooKeeper cluster for other parts of your infrastructure (such as an Hadoop cluster)?

The TL;DR version is: Try using ZooKeeper 3.4.5 for both Kafka and Storm, but see the caveats and workarounds below. In the worst case use separate ZooKeeper clusters/versions for Storm (3.4.5) and Kafka (3.3.4). Generally speaking though, the best 3.3.x version of ZooKeeper is 3.3.6, which is the latest stable 3.3.x version. This is because 3.3.6 fixed a number of serious bugs that could lead to data corruption.

Tip: You can verify the exact ZK version used in kafka-storm-starter by running ./sbt dependency-graph.

Notes:

There is an open Kafka JIRA ticket that covers upgrading Kafka to ZK 3.4.5, see KAFKA-854: Upgrade dependencies for 0.8.
If in a production environment you run into problems when using ZooKeeper 3.4.5 with Storm <= 0.9.1, you can try a workaround using Google jarjar in order to deploy ZooKeeper 3.4.5 alongside Storm's/Curator's hard dependency on ZooKeeper 3.3.3. Another user reported that he uses ZK 3.4.5 in production and ZK 3.3.3 for local testing by not including ZooKeeper in the uber jar and putting the correct ZK version in the CLASSPATH at runtime. STORM-70: Use ZooKeeper 3.4.5.