rxGetJobs: Get Distributed Computing Jobs

Artigo
07/15/2019

Description

Returns a list of job objects associated with the given compute context and matching the specified parameters.

Usage

  rxGetJobs(computeContext, exactMatch = FALSE, startTime = NULL, endTime = NULL, 
            states = NULL, verbose = TRUE)

Arguments

`computeContext`

A compute context object.

`exactMatch`

Determines if jobs are matched using the full compute context, or a simpler subset. If TRUE, only jobs which use the same context object are returned. If FALSE, all jobs which have the same headNode (if available) and ShareDir are returned.

`startTime`

A time, specified as a POSIXct object. If specified, only jobs created at or after startTime are returned. For non-RxHadoopMR contexts, this time should be specified in the user's local time; for RxHadoopMR contexts, the time should specified in GMT. See below for more details.

`endTime`

A time, specified as a POSIXct object. If specified, only jobs created at or before endTime are returned. For non-RxHadoopMR contexts, this time should be specified in the user's local time; for RxHadoopMR contexts, the time should specified in GMT. See below for more details.

`states`

If specified (as a character vector of states that can include "none", "finished", "failed", "canceled", "undetermined" "queued"or "running"), only jobs in those states are returned. Otherwise, no filtering is performed on job state.

`verbose`

If TRUE (the default), a brief summary of each job is printed as it is found. This includes the current job status as returned by rxGetJobStatus, the modification time of the job, and the current job ID (this is used as the component name in the returned list of job information objects). If no job status is returned, the job status shows none.

Details

One common use of rxGetJobs is as input to the rxCleanupJobs function, which is used to clean up completed non-waiting jobs when autoCleanup is not specified.

If exactMatch=FALSE, only the shared directory shareDir and the cluster head name headNode (if available) are compared. Otherwise, all slots are compared. However, if the nodes slot in either compute context is NULL, that slot is also omitted from the comparison.

On LSF clusters, job information by default is held for only one hour (although this is configurable using the LSF parameter CLEAN_PERIOD); jobs older than the CLEAN_PERIOD setting will have status "undetermined".

For non-RxHadoopMR cluster types, all time values are specified and displayed in the user's computer's local time settings, regardless of the time zone settings and differences between the user's computer and the cluster. Thus, start and end times for job filtering should be provided in local time, with the expectation that cluster time values will also be converted for the user into system local time. For RxHadoopMR, the job time and comparison times are stored and performed based on a GMT time.

Note also that when there are a large number of jobs on the cluster, you can improve performance by using the startTime and endTime parameters to narrow your search.

Value

Returns a rxJobInfoList, list of job information objects based on the compute context.

Author(s)

Microsoft Corporation Microsoft Technical Support

Examples


 ## Not run:

myCluster <- RxComputeContext("RxSpark",
   # Location of Revo64 on each node
   revoPath = file.path(defaultRNodePath, "bin", "x64"),                                           
   nameNode = "cluster-head2", 
   # User directory for read/write                                         
   shareDir = "\\AllShare\\myName"                            
   )

rxSetComputeContext(computeContext = myCluster )

# Get all jobs older than a week and newer than two weeks that are finished or canceled.
rxGetJobs(myCluster, startTime = Sys.time() - 3600 * 24 * 14, endTime = Sys.time() - 3600 * 24 * 7, 
         exactMatch = FALSE, states = c( "finished", "canceled") )

# Get all jobs associated with myCluster compute context and then get job output and results
myJobs <- rxGetJobs(myCluster)
print(myJobs)
# returns
# rxJob_1461  
myJobs$rxJob_1461
rxGetJobOutput(myJobs$rxJob_1461)
rxGetJobResults(myJobs$rxJob_1461)
## End(Not run)

Partilhar via