Version: 1.0.2

Flux

A framework for creating and deploying Apache Storm streaming computations with less friction.

Definition

flux |fləks| noun

  1. The action or process of flowing or flowing out
  2. Continuous change
  3. In physics, the rate of flow of a fluid, radiant energy, or particles across a given area
  4. A substance mixed with a solid to lower its melting point

Rationale

Bad things happen when configuration is hard-coded. No one should have to recompile or repackage an application in order to change configuration.

About

Flux is a framework and set of utilities that make defining and deploying Apache Storm topologies less painful and deveoper-intensive.

Have you ever found yourself repeating this pattern?:


public static void main(String[] args) throws Exception {
    // logic to determine if we're running locally or not...
    // create necessary config options...
    boolean runLocal = shouldRunLocal();
    if(runLocal){
        LocalCluster cluster = new LocalCluster();
        cluster.submitTopology(name, conf, topology);
    } else {
        StormSubmitter.submitTopology(name, conf, topology);
    }
}

Wouldn't something like this be easier:

storm jar mytopology.jar org.apache.storm.flux.Flux --local config.yaml

or:

storm jar mytopology.jar org.apache.storm.flux.Flux --remote config.yaml

Another pain point often mentioned is the fact that the wiring for a Topology graph is often tied up in Java code, and that any changes require recompilation and repackaging of the topology jar file. Flux aims to alleviate that pain by allowing you to package all your Storm components in a single jar, and use an external text file to define the layout and configuration of your topologies.

Features

  • Easily configure and deploy Storm topologies (Both Storm core and Microbatch API) without embedding configuration in your topology code
  • Support for existing topology code (see below)
  • Define Storm Core API (Spouts/Bolts) using a flexible YAML DSL
  • YAML DSL support for most Storm components (storm-kafka, storm-hdfs, storm-hbase, etc.)
  • Convenient support for multi-lang components
  • External property substitution/filtering for easily switching between configurations/environments (similar to Maven-style ${variable.name} substitution)

Usage

To use Flux, add it as a dependency and package all your Storm components in a fat jar, then create a YAML document to define your topology (see below for YAML configuration options).

Building from Source

The easiest way to use Flux, is to add it as a Maven dependency in you project as described below.

If you would like to build Flux from source and run the unit/integration tests, you will need the following installed on your system:

  • Python 2.6.x or later
  • Node.js 0.10.x or later

Building with unit tests enabled:

mvn clean install

Building with unit tests disabled:

If you would like to build Flux without installing Python or Node.js you can simply skip the unit tests:

mvn clean install -DskipTests=true

Note that if you plan on using Flux to deploy topologies to a remote cluster, you will still need to have Python installed since it is required by Apache Storm.

Building with integration tests enabled:

mvn clean install -DskipIntegration=false

Packaging with Maven

To enable Flux for your Storm components, you need to add it as a dependency such that it's included in the Storm topology jar. This can be accomplished with the Maven shade plugin (preferred) or the Maven assembly plugin (not recommended).

Flux Maven Dependency

The current version of Flux is available in Maven Central at the following coordinates: xml <dependency> <groupId>org.apache.storm</groupId> <artifactId>flux-core</artifactId> <version>${storm.version}</version> </dependency>

Using shell spouts and bolts requires additional Flux Wrappers library: xml <dependency> <groupId>org.apache.storm</groupId> <artifactId>flux-wrappers</artifactId> <version>${storm.version}</version> </dependency>

Creating a Flux-Enabled Topology JAR

The example below illustrates Flux usage with the Maven shade plugin:

<!-- include Flux and user dependencies in the shaded jar -->
<dependencies>
    <!-- Flux include -->
    <dependency>
        <groupId>org.apache.storm</groupId>
        <artifactId>flux-core</artifactId>
        <version>${storm.version}</version>
    </dependency>
    <!-- Flux Wrappers include -->
    <dependency>
        <groupId>org.apache.storm</groupId>
        <artifactId>flux-wrappers</artifactId>
        <version>${storm.version}</version>
    </dependency>

    <!-- add user dependencies here... -->

</dependencies>
<!-- create a fat jar that includes all dependencies -->
<build>
    <plugins>
        <plugin>
            <groupId>org.apache.maven.plugins</groupId>
            <artifactId>maven-shade-plugin</artifactId>
            <version>1.4</version>
            <configuration>
                <createDependencyReducedPom>true</createDependencyReducedPom>
            </configuration>
            <executions>
                <execution>
                    <phase>package</phase>
                    <goals>
                        <goal>shade</goal>
                    </goals>
                    <configuration>
                        <transformers>
                            <transformer
                                    implementation="org.apache.maven.plugins.shade.resource.ServicesResourceTransformer"/>
                            <transformer
                                    implementation="org.apache.maven.plugins.shade.resource.ManifestResourceTransformer">
                                <mainClass>org.apache.storm.flux.Flux</mainClass>
                            </transformer>
                        </transformers>
                    </configuration>
                </execution>
            </executions>
        </plugin>
    </plugins>
</build>

Deploying and Running a Flux Topology

Once your topology components are packaged with the Flux dependency, you can run different topologies either locally or remotely using the storm jar command. For example, if your fat jar is named myTopology-0.1.0-SNAPSHOT.jar you could run it locally with the command:

storm jar myTopology-0.1.0-SNAPSHOT.jar org.apache.storm.flux.Flux --local my_config.yaml

Command line options

usage: storm jar <my_topology_uber_jar.jar> org.apache.storm.flux.Flux
             [options] <topology-config.yaml>
 -d,--dry-run                 Do not run or deploy the topology. Just
                              build, validate, and print information about
                              the topology.
 -e,--env-filter              Perform environment variable substitution.
                              Replace keys identified with `${ENV-[NAME]}`
                              will be replaced with the corresponding
                              `NAME` environment value
 -f,--filter <file>           Perform property substitution. Use the
                              specified file as a source of properties,
                              and replace keys identified with {$[property
                              name]} with the value defined in the
                              properties file.
 -i,--inactive                Deploy the topology, but do not activate it.
 -l,--local                   Run the topology in local mode.
 -n,--no-splash               Suppress the printing of the splash screen.
 -q,--no-detail               Suppress the printing of topology details.
 -r,--remote                  Deploy the topology to a remote cluster.
 -R,--resource                Treat the supplied path as a classpath
                              resource instead of a file.
 -s,--sleep <ms>              When running locally, the amount of time to
                              sleep (in ms.) before killing the topology
                              and shutting down the local cluster.
 -z,--zookeeper <host:port>   When running in local mode, use the
                              ZooKeeper at the specified <host>:<port>
                              instead of the in-process ZooKeeper.
                              (requires Storm 0.9.3 or later)

NOTE: Flux tries to avoid command line switch collision with the storm command, and allows any other command line switches to pass through to the storm command.

For example, you can use the storm command switch -c to override a topology configuration property. The following example command will run Flux and override the nimbus.seeds configuration:

storm jar myTopology-0.1.0-SNAPSHOT.jar org.apache.storm.flux.Flux --remote my_config.yaml -c 'nimbus.seeds=["localhost"]'

Sample output

███████╗██╗     ██╗   ██╗██╗  ██╗
██╔════╝██║     ██║   ██║╚██╗██╔╝
█████╗  ██║     ██║   ██║ ╚███╔╝
██╔══╝  ██║     ██║   ██║ ██╔██╗
██║     ███████╗╚██████╔╝██╔╝ ██╗
╚═╝     ╚══════╝ ╚═════╝ ╚═╝  ╚═╝
+-         Apache Storm        -+
+-  data FLow User eXperience  -+
Version: 0.3.0
Parsing file: /Users/hsimpson/Projects/donut_domination/storm/shell_test.yaml
---------- TOPOLOGY DETAILS ----------
Name: shell-topology
--------------- SPOUTS ---------------
sentence-spout[1](org.apache.storm.flux.wrappers.spouts.FluxShellSpout)
---------------- BOLTS ---------------
splitsentence[1](org.apache.storm.flux.wrappers.bolts.FluxShellBolt)
log[1](org.apache.storm.flux.wrappers.bolts.LogInfoBolt)
count[1](org.apache.storm.testing.TestWordCounter)
--------------- STREAMS ---------------
sentence-spout --SHUFFLE--> splitsentence
splitsentence --FIELDS--> count
count --SHUFFLE--> log
--------------------------------------
Submitting topology: 'shell-topology' to remote cluster...

YAML Configuration

Flux topologies are defined in a YAML file that describes a topology. A Flux topology definition consists of the following:

  1. A topology name
  2. A list of topology "components" (named Java objects that will be made available in the environment)
  3. EITHER (A DSL topology definition):
    • A list of spouts, each identified by a unique ID
    • A list of bolts, each identified by a unique ID
    • A list of "stream" objects representing a flow of tuples between spouts and bolts
  4. OR (A JVM class that can produce a org.apache.storm.generated.StormTopology instance:
    • A topologySource definition.

For example, here is a simple definition of a wordcount topology using the YAML DSL:

name: "yaml-topology"
config:
  topology.workers: 1

# spout definitions
spouts:
  - id: "spout-1"
    className: "org.apache.storm.testing.TestWordSpout"
    parallelism: 1

# bolt definitions
bolts:
  - id: "bolt-1"
    className: "org.apache.storm.testing.TestWordCounter"
    parallelism: 1
  - id: "bolt-2"
    className: "org.apache.storm.flux.wrappers.bolts.LogInfoBolt"
    parallelism: 1

#stream definitions
streams:
  - name: "spout-1 --> bolt-1" # name isn't used (placeholder for logging, UI, etc.)
    from: "spout-1"
    to: "bolt-1"
    grouping:
      type: FIELDS
      args: ["word"]

  - name: "bolt-1 --> bolt2"
    from: "bolt-1"
    to: "bolt-2"
    grouping:
      type: SHUFFLE


Property Substitution/Filtering

It's common for developers to want to easily switch between configurations, for example switching deployment between a development environment and a production environment. This can be accomplished by using separate YAML configuration files, but that approach would lead to unnecessary duplication, especially in situations where the Storm topology does not change, but configuration settings such as host names, ports, and parallelism paramters do.

For this case, Flux offers properties filtering to allow you two externalize values to a .properties file and have them substituted before the .yaml file is parsed.

To enable property filtering, use the --filter command line option and specify a .properties file. For example, if you invoked flux like so:

storm jar myTopology-0.1.0-SNAPSHOT.jar org.apache.storm.flux.Flux --local my_config.yaml --filter dev.properties

With the following dev.properties file:

kafka.zookeeper.hosts: localhost:2181

You would then be able to reference those properties by key in your .yaml file using ${} syntax:

  - id: "zkHosts"
    className: "org.apache.storm.kafka.ZkHosts"
    constructorArgs:
      - "${kafka.zookeeper.hosts}"

In this case, Flux would replace ${kafka.zookeeper.hosts} with localhost:2181 before parsing the YAML contents.

Environment Variable Substitution/Filtering

Flux also allows environment variable substitution. For example, if an environment variable named ZK_HOSTS if defined, you can reference it in a Flux YAML file with the following syntax:

${ENV-ZK_HOSTS}

Components

Components are essentially named object instances that are made available as configuration options for spouts and bolts. If you are familiar with the Spring framework, components are roughly analagous to Spring beans.

Every component is identified, at a minimum, by a unique identifier (String) and a class name (String). For example, the following will make an instance of the org.apache.storm.kafka.StringScheme class available as a reference under the key "stringScheme" . This assumes the org.apache.storm.kafka.StringScheme has a default constructor.

components:
  - id: "stringScheme"
    className: "org.apache.storm.kafka.StringScheme"

Contructor Arguments, References, Properties and Configuration Methods

Constructor Arguments

Arguments to a class constructor can be configured by adding a contructorArgs element to a components. constructorArgs is a list of objects that will be passed to the class' constructor. The following example creates an object by calling the constructor that takes a single string as an argument:

  - id: "zkHosts"
    className: "org.apache.storm.kafka.ZkHosts"
    constructorArgs:
      - "localhost:2181"

References

Each component instance is identified by a unique id that allows it to be used/reused by other components. To reference an existing component, you specify the id of the component with the ref tag.

In the following example, a component with the id "stringScheme" is created, and later referenced, as a an argument to another component's constructor:

components:
  - id: "stringScheme"
    className: "org.apache.storm.kafka.StringScheme"

  - id: "stringMultiScheme"
    className: "org.apache.storm.spout.SchemeAsMultiScheme"
    constructorArgs:
      - ref: "stringScheme" # component with id "stringScheme" must be declared above.

N.B.: References can only be used after (below) the object they point to has been declared.

Properties

In addition to calling constructors with different arguments, Flux also allows you to configure components using JavaBean-like setter methods and fields declared as public:

  - id: "spoutConfig"
    className: "org.apache.storm.kafka.SpoutConfig"
    constructorArgs:
      # brokerHosts
      - ref: "zkHosts"
      # topic
      - "myKafkaTopic"
      # zkRoot
      - "/kafkaSpout"
      # id
      - "myId"
    properties:
      - name: "ignoreZkOffsets"
        value: true
      - name: "scheme"
        ref: "stringMultiScheme"

In the example above, the properties declaration will cause Flux to look for a public method in the SpoutConfig with the signature setForceFromStart(boolean b) and attempt to invoke it. If a setter method is not found, Flux will then look for a public instance variable with the name ignoreZkOffsets and attempt to set its value.

References may also be used as property values.

Configuration Methods

Conceptually, configuration methods are similar to Properties and Constructor Args -- they allow you to invoke an arbitrary method on an object after it is constructed. Configuration methods are useful for working with classes that don't expose JavaBean methods or have constructors that can fully configure the object. Common examples include classes that use the builder pattern for configuration/composition.

The following YAML example creates a bolt and configures it by calling several methods:

bolts:
  - id: "bolt-1"
    className: "org.apache.storm.flux.test.TestBolt"
    parallelism: 1
    configMethods:
      - name: "withFoo"
        args:
          - "foo"
      - name: "withBar"
        args:
          - "bar"
      - name: "withFooBar"
        args:
          - "foo"
          - "bar"

The signatures of the corresponding methods are as follows:

    public void withFoo(String foo);
    public void withBar(String bar);
    public void withFooBar(String foo, String bar);

Arguments passed to configuration methods work much the same way as constructor arguments, and support references as well.

Using Java enums in Contructor Arguments, References, Properties and Configuration Methods

You can easily use Java enum values as arguments in a Flux YAML file, simply by referencing the name of the enum.

For example, Storm's HDFS module includes the following enum definition (simplified for brevity):

public static enum Units {
    KB, MB, GB, TB
}

And the org.apache.storm.hdfs.bolt.rotation.FileSizeRotationPolicy class has the following constructor:

public FileSizeRotationPolicy(float count, Units units)

The following Flux component definition could be used to call the constructor:

  - id: "rotationPolicy"
    className: "org.apache.storm.hdfs.bolt.rotation.FileSizeRotationPolicy"
    constructorArgs:
      - 5.0
      - MB

The above definition is functionally equivalent to the following Java code:

// rotate files when they reach 5MB
FileRotationPolicy rotationPolicy = new FileSizeRotationPolicy(5.0f, Units.MB);

Topology Config

The config section is simply a map of Storm topology configuration parameters that will be passed to the org.apache.storm.StormSubmitter as an instance of the org.apache.storm.Config class:

config:
  topology.workers: 4
  topology.max.spout.pending: 1000
  topology.message.timeout.secs: 30

Existing Topologies

If you have existing Storm topologies, you can still use Flux to deploy/run/test them. This feature allows you to leverage Flux Constructor Arguments, References, Properties, and Topology Config declarations for existing topology classes.

The easiest way to use an existing topology class is to define a getTopology() instance method with one of the following signatures:

public StormTopology getTopology(Map<String, Object> config)

or:

public StormTopology getTopology(Config config)

You could then use the following YAML to configure your topology:

name: "existing-topology"
topologySource:
  className: "org.apache.storm.flux.test.SimpleTopology"

If the class you would like to use as a topology source has a different method name (i.e. not getTopology), you can override it:

name: "existing-topology"
topologySource:
  className: "org.apache.storm.flux.test.SimpleTopology"
  methodName: "getTopologyWithDifferentMethodName"

N.B.: The specified method must accept a single argument of type java.util.Map<String, Object> or org.apache.storm.Config, and return a org.apache.storm.generated.StormTopology object.

YAML DSL

Spouts and Bolts

Spout and Bolts are configured in their own respective section of the YAML configuration. Spout and Bolt definitions are extensions to the component definition that add a parallelism parameter that sets the parallelism for a component when the topology is deployed.

Because spout and bolt definitions extend component they support constructor arguments, references, and properties as well.

Shell spout example:

spouts:
  - id: "sentence-spout"
    className: "org.apache.storm.flux.wrappers.spouts.FluxShellSpout"
    # shell spout constructor takes 2 arguments: String[], String[]
    constructorArgs:
      # command line
      - ["node", "randomsentence.js"]
      # output fields
      - ["word"]
    parallelism: 1

Kafka spout example:

components:
  - id: "stringScheme"
    className: "org.apache.storm.kafka.StringScheme"

  - id: "stringMultiScheme"
    className: "org.apache.storm.spout.SchemeAsMultiScheme"
    constructorArgs:
      - ref: "stringScheme"

  - id: "zkHosts"
    className: "org.apache.storm.kafka.ZkHosts"
    constructorArgs:
      - "localhost:2181"

# Alternative kafka config
#  - id: "kafkaConfig"
#    className: "org.apache.storm.kafka.KafkaConfig"
#    constructorArgs:
#      # brokerHosts
#      - ref: "zkHosts"
#      # topic
#      - "myKafkaTopic"
#      # clientId (optional)
#      - "myKafkaClientId"

  - id: "spoutConfig"
    className: "org.apache.storm.kafka.SpoutConfig"
    constructorArgs:
      # brokerHosts
      - ref: "zkHosts"
      # topic
      - "myKafkaTopic"
      # zkRoot
      - "/kafkaSpout"
      # id
      - "myId"
    properties:
      - name: "ignoreZkOffsets"
        value: true
      - name: "scheme"
        ref: "stringMultiScheme"

config:
  topology.workers: 1

# spout definitions
spouts:
  - id: "kafka-spout"
    className: "org.apache.storm.kafka.KafkaSpout"
    constructorArgs:
      - ref: "spoutConfig"

Bolt Examples:

# bolt definitions
bolts:
  - id: "splitsentence"
    className: "org.apache.storm.flux.wrappers.bolts.FluxShellBolt"
    constructorArgs:
      # command line
      - ["python", "splitsentence.py"]
      # output fields
      - ["word"]
    parallelism: 1
    # ...

  - id: "log"
    className: "org.apache.storm.flux.wrappers.bolts.LogInfoBolt"
    parallelism: 1
    # ...

  - id: "count"
    className: "org.apache.storm.testing.TestWordCounter"
    parallelism: 1
    # ...

Streams and Stream Groupings

Streams in Flux are represented as a list of connections (Graph edges, data flow, etc.) between the Spouts and Bolts in a topology, with an associated Grouping definition.

A Stream definition has the following properties:

name: A name for the connection (optional, currently unused)

from: The id of a Spout or Bolt that is the source (publisher)

to: The id of a Spout or Bolt that is the destination (subscriber)

grouping: The stream grouping definition for the Stream

A Grouping definition has the following properties:

type: The type of grouping. One of ALL,CUSTOM,DIRECT,SHUFFLE,LOCAL_OR_SHUFFLE,FIELDS,GLOBAL, or NONE.

streamId: The Storm stream ID (Optional. If unspecified will use the default stream)

args: For the FIELDS grouping, a list of field names.

customClass For the CUSTOM grouping, a definition of custom grouping class instance

The streams definition example below sets up a topology with the following wiring:

    kafka-spout --> splitsentence --> count --> log
#stream definitions
# stream definitions define connections between spouts and bolts.
# note that such connections can be cyclical
# custom stream groupings are also supported

streams:
  - name: "kafka --> split" # name isn't used (placeholder for logging, UI, etc.)
    from: "kafka-spout"
    to: "splitsentence"
    grouping:
      type: SHUFFLE

  - name: "split --> count"
    from: "splitsentence"
    to: "count"
    grouping:
      type: FIELDS
      args: ["word"]

  - name: "count --> log"
    from: "count"
    to: "log"
    grouping:
      type: SHUFFLE

Custom Stream Groupings

Custom stream groupings are defined by setting the grouping type to CUSTOM and defining a customClass parameter that tells Flux how to instantiate the custom class. The customClass definition extends component, so it supports constructor arguments, references, and properties as well.

The example below creates a Stream with an instance of the org.apache.storm.testing.NGrouping custom stream grouping class.

  - name: "bolt-1 --> bolt2"
    from: "bolt-1"
    to: "bolt-2"
    grouping:
      type: CUSTOM
      customClass:
        className: "org.apache.storm.testing.NGrouping"
        constructorArgs:
          - 1

Includes and Overrides

Flux allows you to include the contents of other YAML files, and have them treated as though they were defined in the same file. Includes may be either files, or classpath resources.

Includes are specified as a list of maps:

includes:
  - resource: false
    file: "src/test/resources/configs/shell_test.yaml"
    override: false

If the resource property is set to true, the include will be loaded as a classpath resource from the value of the file attribute, otherwise it will be treated as a regular file.

The override property controls how includes affect the values defined in the current file. If override is set to true, values in the included file will replace values in the current file being parsed. If override is set to false, values in the current file being parsed will take precedence, and the parser will refuse to replace them.

N.B.: Includes are not yet recursive. Includes from included files will be ignored.

Basic Word Count Example

This example uses a spout implemented in JavaScript, a bolt implemented in Python, and a bolt implemented in Java

Topology YAML config:

---
name: "shell-topology"
config:
  topology.workers: 1

# spout definitions
spouts:
  - id: "sentence-spout"
    className: "org.apache.storm.flux.wrappers.spouts.FluxShellSpout"
    # shell spout constructor takes 2 arguments: String[], String[]
    constructorArgs:
      # command line
      - ["node", "randomsentence.js"]
      # output fields
      - ["word"]
    parallelism: 1

# bolt definitions
bolts:
  - id: "splitsentence"
    className: "org.apache.storm.flux.wrappers.bolts.FluxShellBolt"
    constructorArgs:
      # command line
      - ["python", "splitsentence.py"]
      # output fields
      - ["word"]
    parallelism: 1

  - id: "log"
    className: "org.apache.storm.flux.wrappers.bolts.LogInfoBolt"
    parallelism: 1

  - id: "count"
    className: "org.apache.storm.testing.TestWordCounter"
    parallelism: 1

#stream definitions
# stream definitions define connections between spouts and bolts.
# note that such connections can be cyclical
# custom stream groupings are also supported

streams:
  - name: "spout --> split" # name isn't used (placeholder for logging, UI, etc.)
    from: "sentence-spout"
    to: "splitsentence"
    grouping:
      type: SHUFFLE

  - name: "split --> count"
    from: "splitsentence"
    to: "count"
    grouping:
      type: FIELDS
      args: ["word"]

  - name: "count --> log"
    from: "count"
    to: "log"
    grouping:
      type: SHUFFLE

Micro-Batching (Trident) API Support

Currenty, the Flux YAML DSL only supports the Core Storm API, but support for Storm's micro-batching API is planned.

To use Flux with a Trident topology, define a topology getter method and reference it in your YAML config:

name: "my-trident-topology"

config:
  topology.workers: 1

topologySource:
  className: "org.apache.storm.flux.test.TridentTopologySource"
  # Flux will look for "getTopology", this will override that.
  methodName: "getTopologyWithDifferentMethodName"