In addition to out-of-the-box scheduler for Aurora, Heron can also be deployed in a HPC cluster with the Slurm Scheduler. This allows a researcher to deploy Heron and execute streaming scientific work-flows.

How Slurm Deployment Works

Using the Slurm scheduler is similar to deploying Heron on other systems. The Heron (../../heron-cli) cli is used to deploy and manage topologies similar to other schedulers. The main difference is in the configuration.

A set of default configuration files are provided with Heron in the conf/slurm directory. The default configuration uses the local file system based state manager. It is possible that the local file system is mounted using NFS.

When a Heron topology is submitted, the Slurm scheduler allocates the nodes required to run the job and starts the Heron processes in those nodes. It uses a script found in conf/slum directory to submit the topoloy as a batch job to the slurm scheduler.

Slurm Scheduler Configuration

To configure Heron to use slurm scheduler, specify the following in scheduler.yaml config file:

  • heron.class.scheduler — Indicates the class to be loaded for slurm scheduler. Set this to com.twitter.heron.scheduler.slurm.SlurmScheduler

  • heron.class.launcher — Specifies the class to be loaded for launching topologies. Set this to com.twitter.heron.scheduler.slurm.SlurmLauncher

  • — The shared directory to be used as Heron sandbox directory.

  • heron.package.core.uri — Indicates the location of the heron core binary package. The local scheduler uses this URI to download the core package to the working directory.

  • — This is used to specify the java home to be used when running topologies in the containers. Set to ${JAVA_HOME} to use the value set in the bash environment variable $JAVA_HOME.

  • — Indicate whether the scheduler is a service. In the case of Slurm, it should be set to False.

Example Slurm Scheduler Configuration

# scheduler class for distributing the topology for execution
heron.class.scheduler: com.twitter.heron.scheduler.slurm.SlurmScheduler

# launcher class for submitting and launching the topology
heron.class.launcher: com.twitter.heron.scheduler.slurm.SlurmLauncher

# working directory for the topologies ${HOME}/.herondata/topologies/${CLUSTER}/${TOPOLOGY}

# location of java - pick it up from shell environment ${JAVA_HOME}

# Invoke the IScheduler as a library directly False

Slurm Script

The script is used by the scheduler to submit the Heron job to the Slurm scheduler. Edit this file to set specific slurm settings like time, account. The script and scheduler.yaml must be included with other cluster configuration files.