[drmaa-wg] DRMAA-WG May 3, 2005 meeting minutes

[Peter Troeger]

>A. Meeting secretary for this meeting?
>  
>
Peter.

>B. Acceptance of the Apr 19, 2005 meeting minutes
>  
>
Accepted.

>C. Admin
>   	- status of the existing trackers
>	- remaining trackers - see Peter's mail
>  
>
    * Placeholder issue (#805) : Hrabri will ask Bill for PBS semantics
    * Discussion about error code proposals in 1177 (closely related to
      1178, 1180, 1183, 1184) :


    * Proposal to add DRMAA_ERRNO_DRM_COMMUNICATION_FAILURE to all functions
    * Use case: SGE DRMAA implementation uses sockets for communication
      with DRMS, so comm failure may occur in any case
    * -> Acceptance of the idea suggested : an implementation is not
      enforced to make us of it, but there are cases ('remote DRMAA')
      where this is needed

    * Proposal to add DRMAA_ERRNO_NO_ACTIVE_SESSION to all manipulation
      functions for job templates
    * Leads to discussion about job template reuse between sessions
    * Actual SGE Java-DRMAA implementation deletes all job templates on
      session closing (otherwise memory leak with underlying C-DRMAA)
    * problem of different attribute semantics in different sessions ->
      agreement that this can mostly only be checked on job submission
      time -> would open possibility for job template reuse
    * Counter-argument: Session is responsible for memory management,
      job template reuse may be difficult to implement
    * Discussion result: Realizability of job template reuse depends on
      environmental conditions of the specific language, no decision
      possible for language-independent spec
    * -> Acceptance of the idea suggested : an implementation might have
      a need to express this error condition, in case it does not allow
      job template reuse

    * Proposal to add DRMAA_ERRNO_AUTH_FAILURE to all functions
    * original DRMAA idea was that authentication happens on
      initialization time
    * -> Rejection of the idea suggested (e.g. for 1177, 1183, 1148, 1180)

    * Proposal to start one mail thread per error code, in order to
      reach final decisions