Bugzilla – Bug 5800
adjust globusrun-ws to changes in job termination in gram in GT4.2
Last modified: 2008-04-30 16:09:11
You need to log in before you can comment on or make changes to this bug.
Gram4's port type does no longer extend from the ImmediateResourceTermination port type. Instead Gram4 offers a method terminate as described in the attached document at http://www.globus.org/mail_archive/gram-dev/2008/01/msg00000.html Find the new terminate method in ws-gram/common/schema/gram/jdd/terminate_managed_job_provider_port_type.wsdl of ws-gram in branch ws-gram-bug5247 The terminate operation is implemented by a gram4 operation provider in 4.2. Both MEJS and MMJS inherit this functionality. A possible scenario for globusrun-ws could be: a) In batch mode: globusrun-ws -kill -j eprFile call terminate with destroyAfterCleanup set to false and destroyDelegatedCredentials set to true. Once the final notification came in (or after polling for job state and the job being in a final state) a second call to terminate() with destroyAfterCleanup set to true and destroyDelegatedCredentials set to true will destroy the resource and potentially delegated credentials. b) In interactive mode: 1. The job is fully processed: call terminate with destroyAfterCleanup set to true and destroyDelegatedCredentials set to true. This would destroy the job resource and potentially delegated credentials immediately. 2. The client interrupts using Ctrl-C: call terminate with destroyAfterCleanup set to false and destroyDelegatedCredentials set to true. Once the final notification came in (or after polling for job state and the job being in a final state) a second call to terminate() with destroyAfterCleanup set to true and destroyDelegatedCredentials set to true will destroy the resource. In case the client hits Ctrl-C again because termination takes too long a second call to terminate() with destroyAfterCleanup set to true and destroyDelegatedCredentials set to true will cause that the job resource will finally be destroyed and globusrun-ws can go away. The initial call to terminate with destroyAfterCleanup set to false gives the client a chance to definitely get informed about the success of the termination, either by the last notification (UserTerminateDone, UserTerminateFailed) or by polling for state in case the final notification didn't come through. The new code of ws-gram can be found in branch ws-gram-bug5247.
Committed code implementing this behavior to the ws-gram-bug5247 branch.
Joe can this all done? Is this code in the trunk? How did you test it? Are there any new globusrun-ws test for these changes? -Stu
This is all in trunk. Not sure how to automate tests for this.