2014 09 18

• Agenda:

  • open TODOs
    • TODO SJ: doc is missing 10.000 feet overview (SJ?)
    • DONE SJ: contact Helmut about SuperMuc allocationtheradicalgroup
    • DONE AM: reduce testing frequency further after release
    • WIP AM: follow up with Anjani on EC2
    • WIP AM: RP slide decks
    • DONE AM: test supermuc, check out gsi contexts
    • DONE AM: suspend testing on FG
    • DONE MS: verify state model wrt. data staging
    • TODO AM: better name for STATE_X
    • TODO SJ: obtain credentials for theradicalgroup@gmail.com
  • open agenda items
    • WIP AM: AGENDA: cleanup
    • WIP AM: AGENDA: configuration files
    • TODO AM: AGENDA: test suite granularity
    • TODO AM: AGENDA: performance PTY / SHELL / SAGA
    • TODO AM: AGENDA: discuss how to ensure test coverage
    • TODO AM: AGENDA: discuss #307, async call semantics
    • TODO AM: AGENDA: student project: plotting
  • MS.9
    • What are goals for the next couple of weeks?
    • (check on open tickets)
  • eval tutorial with Indiana: online, Scott and Abhinav
    • TODO AM: provide application code
    • TODO RADICAL: CCM
    • TODO AM: follow up on progress
  • configuration files
    • AM: we could re-use what we had in OWMS (code exists in utils)
      • RP resource configs remain as is
      • user configs can be used to overwrite those default settings, like:

        # $HOME/radical/pilot.cfg
        {
            "resources" : {
        
                # add a custom host
                "boskop" : {
                    "defaults"           : "localhost"
                    "pilot_agent"        : "rp-agent-testing.py",
                    "lrms"               : "TORQUE",
                    "task_launch_method" : "SSH",
                    "mpi_launch_method"  : "MPIRUN",
                    "global_virtenv"     : "$HOME/ve/"
                },
        
                # change some user specific variable in 
                # existing RP config entries
                "*.futuregrid.org" : {
                    "username" : "merzky"
                },
                "sierra.futuregrid.org" : {
                    "default_queue"    : "batch"
                },
            }
        }
        

  • cleanup modes
    • 1: cleanup database entries: session.close (cleanup=TRUE)
    • 2: terminate pilots: session.close (terminate=TRUE)
    • 3: clean pilot sandbox: pilot_description.cleanup = TRUE
    • 4: clean unit sandbox: unit_description.cleanup = TRUE
    • 1, 2 are enacted by RP/Application on clean application shutdown
    • 3, 4 are enacted by agent on clean pilot shutdown
    • 1, 2 can be performed after application finishes, via radicalpilot-cleanup
    • 3, 4 cannot be performed after application finishes (yet)
  • STATE_X:
    • AM: should be SCHEDULING: the CU has reached the scheduler but has not yet been assigned to a pilot (e.g., if none is free to run the CU).

      # scheduler
      for task in wait_q :
          task.state = SCHEDULING
      
          while True :
              pilot = find_free_pilot (task)
              if  pilot :
                  task.pilot = pilot
                  break
              
      task.state = PENDING_EXECUTION
      submit_task_to_pilot (task)
      

    • but: SCHEDULING already used within the agent:

      # agent
      for task in mongodb.find (pid  : my_pid, 
                                state: PENDING_EXECUTION)
          task.state = SCHEDULING
      
          while True :
              cores = find_free_cores (task)
              if  cores :
                  task.cores = cores
                  break
              
      task.state = EXECUTING
      submit_task_to_cores (task)
      

• Notes:

  • MS.9:
    • MS: CU startup timing
    • MS: (understanding of) perf
    • MS: data: pushing caps into pilot level, data intell.
    • MS: docs, tickets
    • OW: documentation, tutorial
    • OW: incremental steps, stability
    • OW: don't pack feature tickets
    • SJ: bigger, better, faster, self-maintaining ;)
    • SJ: careful what we want to achieve
    • SJ: easy of use, feature complete documentation
    • SJ: mission: replace BJ, serve our own needs
    • SJ: testing
  • configuration:
    • agree on adding resources
    • no urgent need for changing individual settings
    • TODO AM: reply on #327 with app level config
  • cleanup:
    • complete
    • unit sandbox: pilot cleanup is maybe too late to enforce unit cleanup
    • Antons ticket, Matteo ticket
    • MS: address issues incompletely
  • STATE_X
    • both are SCHEDULING
    • MS: SCHEDULING / ALLOCATION
  • TODO SJ: cleanup new member information