Opened 11 years ago

Last modified 11 years ago

#1290 new enhancement

IZ3292: More than one $TMPDIR at different locations and persistent

Reported by: reuti Owned by:
Priority: normal Milestone:
Component: sge Version: 6.2u5
Severity: Keywords: kernel


[Imported from gridengine issuezilla]

        Issue #:      3292             Platform:     All           Reporter: reuti (reuti)
       Component:     gridengine          OS:        All
     Subcomponent:    kernel           Version:      6.2u5            CC:    None defined
        Status:       NEW              Priority:     P3
      Resolution:                     Issue type:    ENHANCEMENT
                                   Target milestone: ---
      Assigned to:    andreas (andreas)
      QA Contact:     andreas
       * Summary:     More than one $TMPDIR at different locations and persistent
   Status whiteboard:

     Issue 3292 blocks:
   Votes for issue 3292:

   Opened: Thu Nov 18 04:16:00 -0800 2010 

Some installations and/or applications need a central $TMPDIR instead of a local one on each node. Although different queues can get
different settings here, some applications need both: one (preferred) local one for the scratch files, and a second to share data between
nodes for the job. Creating an temporary scratch directory is also less invasive than having additional queues.

By having a list for "tmpdir" in the queue configuration this could be achieved. For now it can be implemented by a prolog/epilog, but the
user has to set any additional $TMPDIR1, $TMPDIR2 and so on on his own (unless set in a startup_method). For each created $TMPDIR an
additional variable could be set, with an upper limit of 9, i.e. max. one digit added.

   ------- Additional comments from reuti Thu Nov 18 11:53:17 -0800 2010 -------
This could even be combined with persistent directories on slave nodes. For now the local $TMPDIRs on slave nodes are removed again, when a
tightly integrated task leaves a node. But there exist applications, which need persistent directories on the nodes for their scratch data
along several `mpiexec` calls. It neither needs to be shared with other nodes, nor should it be on an NFS server for performance reasons.
While the creation of the $TMPDIR could be handled like now, all other created directories $TMPDIR1..9 could be created at job start, and
removed when the job ends. Whether they are central or on a node is up to the user, just specify something like "tmpdir /tmp /tmp/persistent
/nfs/tmp" to get all.

This would also cover:

Change History (0)

Note: See TracTickets for help on using tickets.