arcsub is the key command when submitting jobs to Grid enabled computing resources with the ARC client.
As default arcsub is able to submit jobs to A-REX, CREAM and EMI ES enabled computing elements (CEs), and
as always for successful submission you need to be authenticated at the targeted computing services.
Since arcsub is build on a modular library, modules can be installed which enables submission to other
targets, e.g. the classic ARC CE Grid-Manager.
Job submission can be accomplished by specifying a job description file to submit as an argument. arcsub
will then by default perform resource discovery on the Grid and then the discovered resources will be
matched to the job description and ranked according to the chosen broker (--broker option). If no Grid
environment has been configured, please contact your system administrator, or setup one yourself in the
client configuration file (see files section). Another option is to explicitly specify a registry service
(or multiple) to arcsub using the --index option, which accepts an URL, alias or group. Alternatively a
specific CE (or multiple) can be targeted by using the --cluster option. If such a scenario is the most
common, it is worthwhile to specify those CEs in the client configuration as default services, which
makes it superfluous to specify them as argument. In the same manner aliases and groups, defined in the
configuration file, can be utilized, and can be used as argument to the --cluster or --index options. In
all of the above scenarios arcsub obtains resource information from the services which is then used for
matchmaking against the job description, however that step can be avoided by specifying the --direct
option, in which case the job description is submitted directly to first specified endpoint.
The format of a classic GRIDFTP-based cluster URLs:
[ldap://]<hostname>[:2135/nordugrid-cluster-name=<hostname>,Mds-Vo-name=local,o=grid]
Only the hostname part has to be specified, the rest of the URL is automatically generated.
The format of an A-REX URL is:
[https://]<hostname>[:<port>][/<path>]
Here the port is 443 by default, but the path cannot be guessed, so if it is not specified, then the
service is assumed to live on the root path.
Job descriptions can also be specified using the --jobdescrfile option which expect the file name of the
description as argument, or the --jobdescrstring option which expect as argument the job description as a
string, and both options can be specified multiple times and one does not exclude the other. The default
supported job description languages are xRSL and EMIES ADL.
If the job description is successfully submitted a job-ID is returned and printed. This job-ID uniquely
identifies the job while it is being executed. On the other hand it is also possible that no CEs matches
the constraints defined in the description in which case no submission will be done. Upon successful
submission, the job-ID along with more technical job information is stored in the job-list file
(described below). The stored information enables the job management commands of the ARC client to manage
jobs easily, and thus the job-ID need not to be saved manually. By default the job-list file is stored in
the .arc directory in the home directory of the user, however another location can be specified using the
--joblist option taking the location of this file as argument. If the --joblist option was used during
submission, it should also be specified in the consecutive commands when managing the job. If a Computing
Element has multiple job submission interfaces (e.g. gridftp, EMI-ES, BES), then the brokering algorithm
will choose one of them. With the --submissioninterface option the requested interface can be specified,
and in that case only those Computing Elements will be considered which has that specific interface, and
only that interface will be used to submit the jobs.
As mentioned above registry or index services can be specified with the --index option. Specifying one or
multiple index servers instructs the arcsub command to query the servers for registered CEs, the returned
CEs will then be matched against the job description and those matching will be ranked by the chosen
broker (see below) and submission will be tried in order until successful or reaching the end. From the
returned list of CEs it might happen that a troublesome or undesirable CE is selected for submission, in
that case it possible to reject that cluster using the --rejectdiscovery option and providing the URL (or
just the hostname) of the CE, which will disregard that CE as a target for submission.
When multiple CEs are targeted for submission, the resource broker will be used to filter out CEs which
do not match the job description requirements and then rank the remaining CEs. The broker used by default
will rank the CEs randomly, however a different broker can be chosen by using the --broker option, which
takes the name of the broker as argument. The broker type can also be specified in client.conf. The
brokers available can be seen using arcsub-P. By default the following brokers are available:
Random (default)
Chooses a random CE matching the job requirements.
FastestQueue
Ranks matching CEs according to the length of the job queue at the CEs, ranking those with
shortest queue first/highest.
Benchmark
Ranks matching CEs according to a specified benchmark, which should be specified by appending the
broker name with ':' and then the name of the benchmark. If no option is given to the Benchmark
broker then CEs will be ranked according to the 'specint2000' benchmark.
Data Ranks matching CEs according to the amount of input data cached by each CE, by querying the CE.
Only CEs with the A-REX BES interface support this operation.
Null Choose a random CE with no filtering at all of CEs.
PythonBroker
User-defined custom brokers can be created in Python. See the example broker SampleBroker.py or
ACIXBroker.py (like Data broker but uses the ARC Cache Index) that come installed with ARC for
more details of how to write your own broker. A PythonBroker is specified by --brokerPythonBroker:Filename.Class:args, where Filename is the file containing the class Class which
implements the broker interface. The directory containing this file must be in the PYTHONPATH.
args is optional and allows specifying arguments to the broker.
Before submission, arcsub performs an intelligent modification of the job description (adding or
modifying attributes, even converting the description language to fit the needs of the CE) ensuring that
it is valid. The modified job description can be printed by specifying the --dumpdescription option. The
format, i.e. job description language, of the printed job description cannot be specified, and will be
that which will be sent to and accepted by the chosen target. Further information from arcsub can be
obtained by increasing the verbosity, which is done with the --debug option where the default verbosity
level is WARNING. Setting the level to DEBUG will show all messages, while setting it to FATAL will only
show fatal log messages.
To validate your job description without actually submitting a job, use the --dryrun option: it will
capture possible syntax or other errors, but will instruct the site not to submit the job for execution.
Only the grid-manager (ARC0) and A-REX (ARC1) CEs support this feature.
Install Now
PNG Icons
SVG Icons
