[DRMAA-WG] Meeting Minutes August 18th

[Peter Tröger]

Participants: Peter, Dan, Hrabri

- Continuation of discussion about synchronization approach in DRMAA2
- Dan: Use cases from previous mail come down to two major cases
    - Wait for state transitions to happen, in best case without  
loosing any of them
    - Wait for a job to have some state
- Lossless transition monitoring is the major problem
    - Hard to model in a pull API
    - Too many feasible use cases to just ignore it
- Hrabri:Focus on different application layout, instead of covering  
everything in the API
- Conclusion: Transition monitoring is too hard with a pull model
    - Let transition monitoring be based on an optional callback  
mechanism in DRMAA2
    - Burden of book-keeping / timing issues on the the DRMAA2 library  
implementation
    - DRMAAv2 should pre-define as much as possible (callback  
registration method, callback method signature)
    - Specific technical realization by the language binding
    - Might be not implementable for some systems (e.g. DRMAA Grid  
frontend), therefore optional feature
- Conclusion for state waiting
    - Careful extension of DRMAA1 wait() semantics
    - Wait for permanent status -  "started" or "terminated" - each of  
them includes several DRMAA job states
    - Might be realized by two functions (e.g. "waitAnyStarted" /  
"WaitAnyTerminated")
- Reaping as decided before
    - wait functions on session level take a list of jobs as argument,  
in order to allow iterative job list processing by the application
    - Job state information is explicitly thrown away by deletion  
methods
    - no implicit reaping based on returned state information
- Dan raised new open issue - consideration of "re-scheduled" as  
monitorable job status
- Schedule for working group
    - DRMAAv2 presentation at Sun HPC workshop in Regensburg, Sept. 8th
    - Presentation of all major DRMAA2 concepts at OGF27 by Dan -  
"feature complete" until October
    - Final document submission at the end of the year
    - Hopefully proposed recommendation status at OGF28, big meeting  
(Dan, Peter, Daniel, ...)