Currently, the job SLA maxTaskResubmits attribute is only used in conjunction with Task.setResubmit(), that is, when a task explicitely resubmits itself. This does not allow to prevent task resubmission when an error occurs on the node to which the tasks were submitted. In this scenario, for instance, it is not possible to avoid the side effects of persistent operations performed by the tasks, like yupdating a database or creating/modifying/deleting files on a file system.

We propose to enable maxTaskResubmits to be applied when a node error occurs, with an additional explicit flag. This will be done differently based on the branch in which it is implemented:

• in branch 4.1, we will have the server use a value in the job metadata, for instance: job.getMetadata().getParameter("jppf.job.applyMaxResubmitOnNodeError", false);
• in the trunk (for upcoming v4.2) we will add an SLA attribute for this, which will override the metadata parameter if it is present and when this SLA attribute is true