Subsections

Job submission and visualisation of job outcomes


The Job editor

The UNICORE Rich Client offers graphical editors for setting up job descriptions. Instead of having to edit text-based job descriptions, the user is provided high level interfaces which are taylored to the applications he wants to execute on remote systems. The client is easily extensible with new application specific user interfaces as new applications are introduced to the Grid environment. Setting up a job description only requires a few simple steps and can be performed within a couple of seconds. The first step is the creation of a job project.

Figure 6: Creating a job or workflow project
Image create_project

Creating a job project

There are different ways to create a new job project:
  1. Select File $\rightarrow$ New $\rightarrow$ Job Project from the menu bar (see 1 in Figure 6).
  2. Open the context menu of the Navigator view and select New $\rightarrow$ Job Project.
  3. Use the create job item from the context menu of a target system node.
  4. Choose the restore Job description item from a job's context menu in the Grid Browser.

Figure 7: Wizard for creating a job project
Image job_creation_wizard_1 Image job_creation_wizard_2
Image job_creation_wizard_3  

The first three of these options will pop up a series of wizard dialogs which will guide the user through the creation of the job project (see Figure 7). The first step of the wizard is used to choose an application to be run on the target system. In our example, we would like to execute a simple shell script. Therefore, we have selected the Script application 1. By pressing the Finish button the new job project is created. Click Next which will take you to the next wizard step. Here, a different name for the project 2 and the job file 3 can be set. The third wizard page allows for selecting a different target system for job submission 4. The selected target system can also be modified after the project has been created. When the job is created with the last option, both the target system selection and application to be run are restored from the server. Therefore, the job creation wizard shows the second wizard page, only (where you can set names for the project and job file).

Editing mode

Figure 8: Job editing and submission
Image job_submission

The most convenient way to create a job project is using the context menu of a target system node (see 1 in Figure 8), as the corresponding target system will be pre-selected and the job creation wizard can be completed on the first page. Once the job project and job file have been created, a new job editor will be opened in editing mode, displaying a graphical user interface (GUI) for the application 2. It allows for defining the input parameters of the job to be run. The GUI for the Script application provides an embedded text editor for typing in the shell script 3. New application GUIs can be installed by selecting Help $\rightarrow$ Software Updates $\rightarrow$ Download Grid Applications from the workbench's menu bar. This option requires an application GUI server to be available on the Grid (if no server has been found, the option is not available). The job editor holds several tabs. First the application specific tabs are shown for setting parameters in a user friendly way. In addition, the editor holds three generic panels:

When all parameters are set, click the green submit button (see 7 in Figure 8) to submit the job to the selected target system.

An additional action in the tool bar of the job editor is used to set the job's lifetime 8. When the job has reached the end of its lifetime, the job resource representing the submitted job is destroyed and its working directory is cleaned up automatically. This implies that the job's outcomes cannot be accessed hereafter. The default lifetime for jobs is set to 720 hours (30 days).


Figure 9: Job monitoring and fetching job outcomes
Image job_outcomes


Monitoring Mode

As soon as a job is being submitted, the job file is copied into a newly created subfolder of the `submitted' folder in the job project. The subfolder's name consists of the String `submitted at ', followed by a timestamp, e.g. `2010-03-29 16-00-34' that indicates when the job was submitted (1 in Figure 9). This way, a history of all submitted versions of the job is kept and the user can later look up old job descriptions and compare the results of the associated job executions. The copied version of the job file is then opened in a new job editor.

In order to inform the user about the execution state of the job, the editor is put into monitoring mode. This means that the job description cannot be edited anymore and the title of the job editor indicates the current execution status 2. The status may be one of the four values submitting, running, finished, and failed. If the job editor is closed in state submitting the job submission cannot be performed successfully and the subfolder with the copy of the job file is deleted automatically. If the editor is closed in state running, execution of the job will continue normally on the server side. By double-clicking the job file copy in the Navigator view, the job editor will be re-opened in monitoring mode and continue to watch the job execution. Jobs can be aborted by selecting the `abort' item in their context menu. Aborting a job will interrupt the execution of the associated application as soon as possible (this depends on the target system's ability to abort application runs), but leave the job node (and its working directory node) accessible in the Grid Browser. In contrast, destroying a job will first abort the job and then clean up all used resources including the job's working directory.


Fetching Job Outcomes

Once the job has finished successfully, the fetch output files action becomes available in the tool bar of the job editor in monitoring mode 3. After clicking it, a dialog will appear that shows all produced output files and allows you to deselect files you do not want to download. After clicking OK the selected files are downloaded to the `output files' directory in the subfolder that contains the copy of the submitted job 4. Finally, a new application specific Job Outcome view will appear showing the contents of the job's output files 5. In our example a simple text editor shows the output of the script, but more advanced visualisation software is used for displaying the results of scientific applications (e.g. 3D molecule visualisations for chemical applications). Alternatively, job outcomes can be fetched by selecting fetch output files from the context menu of job nodes in the Grid Browser view 6.

Bastian Demuth
2010-05-25