Configuring a Cluster

To setup a Heron cluster, you need to configure a few files. Each file configures a component of the Heron streaming framework.

  • scheduler.yaml — This file specifies the required classes for launcher, scheduler, and for managing the topology at runtime. Any other specific parameters for the scheduler go into this file.

  • statemgr.yaml — This file contains the classes and the configuration for state manager. The state manager maintains the running state of the topology as logical plan, physical plan, scheduler state, and execution state.

  • uploader.yaml — This file specifies the classes and configuration for the uploader, which uploads the topology jars to storage. Once the containers are scheduled, they will download these jars from the storage for running.

  • heron_internals.yaml — This file contains parameters that control how heron behaves. Tuning these parameters requires advanced knowledge of heron architecture and its components. For starters, the best option is just to copy the file provided with sample configuration. Once you are familiar with the system you can tune these parameters to achieve high throughput or low latency topologies.

  • metrics_sinks.yaml — This file specifies where the run-time system and topology metrics will be routed. By default, the file sink and tmaster sink need to be present. In addition, scribe sink and graphite sink are also supported.

  • packing.yaml — This file specifies the classes for packing algorithm, which defaults to Round Robin, if not specified.

  • client.yaml — This file controls the behavior of the heron client. This is optional.

Assembling the Configuration

All configuration files are assembled together to form the cluster configuration. For example, a cluster named devcluster that uses the Aurora for scheduler, ZooKeeper for state manager and HDFS for uploader will have the following set of configurations.

scheduler.yaml (for Aurora)

# scheduler class for distributing the topology for execution
heron.class.scheduler: com.twitter.heron.scheduler.aurora.AuroraScheduler

# launcher class for submitting and launching the topology
heron.class.launcher: com.twitter.heron.scheduler.aurora.AuroraLauncher

# location of java /usr/lib/jvm/java-1.8.0-openjdk-amd64/

# Invoke the IScheduler as a library directly False

statemgr.yaml (for ZooKeeper)

# zookeeper state manager class for managing state in a persistent fashion
heron.class.state.manager: com.twitter.heron.statemgr.zookeeper.curator.CuratorStateManager

# zookeeper state manager connection string
heron.statemgr.connection.string:  ""

# path of the root address to store the state in zookeeper  
heron.statemgr.root.path: "/heron"

# create the zookeeper nodes, if they do not exist True

uploader.yaml (for HDFS)

# Directory of config files for hadoop client to read from              "/home/hadoop/hadoop/conf/"

# The URI of the directory for uploading topologies in the hdfs      "hdfs:///heron/topology/"

packing.yaml (for Round Robin)

# packing algorithm for packing instances into containers
heron.class.packing.algorithm:    com.twitter.heron.packing.roundrobin.RoundRobinPacking

client.yaml (for heron cli)

# should the role parameter be required
heron.config.role.required: false

# should the environ parameter be required
heron.config.env.required: false