Package net.sf.basedb.opengrid.engine

Class SlurmEngine

java.lang.Object

net.sf.basedb.opengrid.engine.SlurmEngine

All Implemented Interfaces:

ClusterEngine

public class SlurmEngine
extends Object
implements ClusterEngine

Cluster engine implementation for Slurm clusters.

Since:

1.4

Author:

nicklas

Nested Class Summary

Nested Classes

Modifier and Type	Class	Description
(package private) static class	SlurmEngine.PendingJob
static class	SlurmEngine.PendingJobsCmd	Issue an 'squeue' command that get a list of pending jobs sorted in priority order as of this moment.
static class	SlurmEngine.SacctCmd	Implements the 'sacct' command for getting information about a completed job.
static class	SlurmEngine.SlurmJobStatus	Job status information for Slurm jobs.
static class	SlurmEngine.SqueueCmd	Implements the 'squeue' command for getting information about a waiting or running job.
static class	SlurmEngine.StatusCmd	Implementation for getting information about a running or finished job.

Field Summary

Fields

Modifier and Type	Field	Description
private static final Set<String>	ignoredSbatchOptions
private static final ExtensionsLogger	logger
private SlurmEngine.PendingJobsCmd	pendingJobs

Constructor Summary

Constructors

Constructor	Description
SlurmEngine()

Method Summary

Modifier and Type	Method	Description
CmdResult<String>	cancelJob(OpenGridSession session, JobIdentifier jobId)	Tell the cluster that a running or waiting job should be cancelled.
String	createJobScript(JobDefinition job, String workFolder, String tmpFolder)	Generates a script that executes the job script.
JobSubmission	createJobSubmission(OpenGridSession session, JobDefinition job, String workFolder, String tmpFolder)	Create a job submission for executing a job on the cluster.
String	createSbatchScript(OpenGridSession session, JobDefinition job, String workFolder, String tmpFolder)	Generates a script that can be submitted with 'sbatch' to slurm.
private String	getJobStatusPath(OpenGridSession session, String jobId)	Get the path to where the status information for a job is saved.
private UploadSource	getScript(String name)
CmdResult<JobStatus>	getStatusIfFinished(OpenGridSession session, JobIdentifier jobId, int timeAdjustment)	Get information about a job that is expected to have finished.
CmdResult<JobStatus>	getStatusInQueue(OpenGridSession session, JobIdentifier jobId, int timeAdjustment)	Get information about a job that is expected to be waiting in the queue or running.
Collection<Signal>	getSupportedSignals(JobStatus newStatus, Job.Status currentStatus)	Get the signals that are supported for jobs with the given current and new job status.
private boolean	isSacctDisabled(OpenGridSession session)	Check the flag if 'sacct' is disabled on the cluster.
CmdResult<String>	modifyJob(OpenGridSession session, JobIdentifier jobId, Signal signal)	We support PAUSE and RESUME signals (assuming that the job has not started).
void	setDefaultConfig(ClusterConfig config)	Set default configuration parameters on the given configuration object.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Details

logger

private static final ExtensionsLogger logger

ignoredSbatchOptions

private static final Set<String> ignoredSbatchOptions

pendingJobs

private SlurmEngine.PendingJobsCmd pendingJobs

Constructor Details

SlurmEngine

public SlurmEngine()

Method Details

getSupportedSignals

public Collection<Signal> getSupportedSignals(JobStatus newStatus,
 Job.Status currentStatus)

Description copied from interface: ClusterEngine

Get the signals that are supported for jobs with the given current and new job status. This method is typically only called for status: WAITING, PAUSED and EXECUTING. It is safe to return null or an empty collection if no signals are supported.

Specified by:

getSupportedSignals in interface ClusterEngine

setDefaultConfig

public void setDefaultConfig(ClusterConfig config)

Description copied from interface: ClusterEngine

Set default configuration parameters on the given configuration object. It is expected that the ClusterConfig.getType() is a type that is compatible with the engine implementation.

Specified by:

setDefaultConfig in interface ClusterEngine

createJobSubmission

public JobSubmission createJobSubmission(OpenGridSession session,
 JobDefinition job,
 String workFolder,
 String tmpFolder)

Description copied from interface: ClusterEngine

Create a job submission for executing a job on the cluster.

Specified by:

createJobSubmission in interface ClusterEngine

Parameters:

session - A connected session that can be used to execute commands on the cluster

job - Information about the job

workFolder - The work folder where files needed for the job are stored

tmpFolder - A temporary folder where the job is allowed to store files

Returns:

A JobSubmission instance

createSbatchScript

public String createSbatchScript(OpenGridSession session,
 JobDefinition job,
 String workFolder,
 String tmpFolder)

Generates a script that can be submitted with 'sbatch' to slurm. This script will in turn call the actual job script with 'srun'.

createJobScript

public String createJobScript(JobDefinition job,
 String workFolder,
 String tmpFolder)

Generates a script that executes the job script.

getJobStatusPath

private String getJobStatusPath(OpenGridSession session,
 String jobId)

Get the path to where the status information for a job is saved.

Since:

1.7

isSacctDisabled

private boolean isSacctDisabled(OpenGridSession session)

Check the flag if 'sacct' is disabled on the cluster.

Since:

1.7

getScript

private UploadSource getScript(String name)

getStatusInQueue

public CmdResult<JobStatus> getStatusInQueue(OpenGridSession session,
 JobIdentifier jobId,
 int timeAdjustment)

Description copied from interface: ClusterEngine

Get information about a job that is expected to be waiting in the queue or running. If this is not the case, the CmdResult should return with exit status set to 1.

Specified by:

getStatusInQueue in interface ClusterEngine

Parameters:

session - A connected session that can be used to execute commands on the cluster

jobId - Job identifier

timeAdjustment - Adjustment in seconds that should be applied to all times returned by commands on the cluster (this will make times compatible with BASE server)

Returns:

A CmdResult instance with JobStatus information

getStatusIfFinished

public CmdResult<JobStatus> getStatusIfFinished(OpenGridSession session,
 JobIdentifier jobId,
 int timeAdjustment)

Description copied from interface: ClusterEngine

Get information about a job that is expected to have finished.

Specified by:

getStatusIfFinished in interface ClusterEngine

Parameters:

session - A connected session that can be used to execute commands on the cluster

jobId - Job identifier

timeAdjustment - Adjustment in seconds that should be applied to all times returned by commands on the cluster (this will make times compatible with BASE server)

Returns:

A CmdResult instance with JobStatus information

cancelJob

public CmdResult<String> cancelJob(OpenGridSession session,
 JobIdentifier jobId)

Description copied from interface: ClusterEngine

Tell the cluster that a running or waiting job should be cancelled.

Specified by:

cancelJob in interface ClusterEngine

Parameters:

session - A connected session that can be used to execute commands on the cluster

jobId - Job identifier

Returns:

The result of executing the command

modifyJob

public CmdResult<String> modifyJob(OpenGridSession session,
 JobIdentifier jobId,
 Signal signal)

We support PAUSE and RESUME signals (assuming that the job has not started). The commands we use are "scontrol uhold" and "scontrol release".

Specified by:

modifyJob in interface ClusterEngine