Custom Query (431 matches)

Filters
 
Or
 
  
 
Columns

Show under each result:


Results (7 - 9 of 431)

1 2 3 4 5 6 7 8 9 10 11 12 13
Ticket Resolution Summary Owner Reporter
#33 fixed IZ243: Memory leak in sge_schedd ernst
Description

[Imported from gridengine issuezilla http://gridengine.sunsource.net/issues/show_bug.cgi?id=243]

        Issue #:      243              Platform:     All              Reporter: ernst (ernst)
       Component:     gridengine          OS:        All
     Subcomponent:    kernel           Version:      5.3                 CC:    None defined
        Status:       VERIFIED         Priority:     P3
      Resolution:     FIXED           Issue type:    DEFECT
                                   Target milestone: not determined
      Assigned to:    ernst (ernst)
      QA Contact:     andreas
          URL:
       * Summary:     Memory leak in sge_schedd
   Status whiteboard:
      Attachments:

     Issue 243 blocks:
   Votes for issue 243:


   Opened: Mon Apr 29 05:20:00 -0700 2002 
------------------------


Memory leak in sge_schedd

According to the code discussion with Patrik Koch we have identified
several errors in sge_schedd process:

> Code discussion
> ===============
>
> - sge_process_events.c, event_handles_default_scheduler()
>   line 650: if (is_running)... possible???
>   How can it be that tasks of a newly added job are already running?
>   And a few lines above at the beginning of sgeE_JOB_ADD:
>   How can the job and the relevant task be already in the joblist?
>   line 653: at_inc_job_counter() leaves priority_group_list or
>     user list (PGR_subordinate_list) unsorted!

I also would assume that the job/task should not already be in the
joblist. I will fix it.

> - sge_process_events.c, event_handles_default_scheduler()
>   line 1559: sgeE_JATASK_DEL
>   if ja_task is enrolled and the only task of the job -> job is
>       removed from lists.job_list
>   if ja_task is not enrolled but the only task of the job -> job is
>       not removed from lists.job_list ???

I will fix it. (==> Memory leak)

> - sge_job_schedd.c, split_job()
>   line 508: job=NULL;
>   line 525: remaining tasks -> lCopyElem(job) ==>> LERROR !
>   line 567: if (job) -> always false !
>   ??? What should be done with remaining tasks? Are they possible?

I don't think that remaining tasks are possible in the moment. If they
occure, they should stay in the source list if they are not needed for
scheduling decisions or if it is not necessary to generate scheduling
messages for the "qstat -j" output.

I will fix the error you found.

   ------- Additional comments from ernst Mon Apr 29 05:21:54 -0700 2002 -------
Started.

   ------- Additional comments from ernst Mon Apr 29 06:33:29 -0700 2002 -------
Fixed.

   ------- Additional comments from ernst Wed May 8 03:24:11 -0700 2002 -------
Review has been done by Andreas.
#35 fixed IZ245: qhost -l h=<hostname> does not work andreas
Description

[Imported from gridengine issuezilla http://gridengine.sunsource.net/issues/show_bug.cgi?id=245]

        Issue #:      245              Platform:     All          Reporter: andreas (andreas)
       Component:     gridengine          OS:        All
     Subcomponent:    clients          Version:      V53_alpha1      CC:
                                                                            [_] uddeborg
                                                                            [_] Remove selected CCs
        Status:       REOPENED         Priority:     P4
      Resolution:                     Issue type:    DEFECT
                                   Target milestone: 6.0
      Assigned to:    andreas (andreas)
      QA Contact:     roland
          URL:
       * Summary:     qhost -l h=<hostname> does not work
   Status whiteboard:
      Attachments:
                      Date/filename:                             Description:                  Submitted by:
                      Wed Jan 30 06:40:00 -0700 2008: patch.diff That''s the diff (text/plain) andreas

     Issue 245 blocks:
   Votes for issue 245:


   Opened: Tue May 7 07:21:00 -0700 2002 
------------------------


DESCRIPTION:
Though one would expect it to work the 'hostname'
attribute can't be used
with qhost for selecting particular hosts.

WORK AROUND:
# qhost | grep <hostname>

SUGGESTED FIX:
The 'hostname' attribute is not in the 'host'
complex but in the 'queue' complex. For this
reason an exechost does not have a 'hostname'
attribute.To fix this problem the 'hostname'
attribute must be moved into the 'host' complex
and the host information must be filled into this
complex attribute from the exechost.

   ------- Additional comments from uddeborg Fri Jun 17 01:55:57 -0700 2005 -------
This problem still remains in 6.0u5 beta

   ------- Additional comments from sgrell Tue Dec 6 08:37:08 -0700 2005 -------
Changed subcomponent.

Stephan

   ------- Additional comments from roland Tue Nov 14 08:29:51 -0700 2006 -------
WORKAROUND:
qhost -h hostlist

   ------- Additional comments from dom Tue Jan 29 11:57:54 -0700 2008 -------
fixed in maintrunk and V61_BRANCH

   ------- Additional comments from andreas Wed Jan 30 06:39:15 -0700 2008 -------
Fix

http://gridengine.sunsource.net/servlets/ReadMsg?list=cvs&msgNo=9557

appears problematic. The attached patch could solve the actual problem.

   ------- Additional comments from andreas Wed Jan 30 06:40:44 -0700 2008 -------
Created an attachment (id=147)
That''s the diff
#36 fixed IZ246: Load values on HP multi-CPU machines not correct andy
Description

[Imported from gridengine issuezilla http://gridengine.sunsource.net/issues/show_bug.cgi?id=246]

        Issue #:      246              Platform:     HP       Reporter: andy (andy)
       Component:     gridengine          OS:        HP-UX
     Subcomponent:    kernel           Version:      5.3         CC:    None defined
        Status:       VERIFIED         Priority:     P3
      Resolution:     FIXED           Issue type:    DEFECT
                                   Target milestone: 5.3p1
      Assigned to:    joga (joga)
      QA Contact:     andreas
          URL:
       * Summary:     Load values on HP multi-CPU machines not correct
   Status whiteboard:
      Attachments:

     Issue 246 blocks:
   Votes for issue 246:


   Opened: Tue May 7 07:35:00 -0700 2002 
------------------------


On HP multi-CPU machines the load values retrieved
by SGE are already dived by the number of
processors (most probably already done by the
system calles, since the "uptime" command has a
similar behavior).

This happens on HP-UX 10 and HP-UX 11.

Therefore the load values (load_* and np_load*)
are too small compared to all other supported SGE
architectures).

Workaround: use load scaling

   ------- Additional comments from joga Wed Jun 5 00:29:18 -0700 2002 -------
fixing it for 5.3.1 and maintrunc.

   ------- Additional comments from joga Wed Jun 5 00:59:49 -0700 2002 -------
Load values were divided by the number of processors in the code
retrieving them for the hp11 architecture.

Fixed for 5.3 branch and maintrunc.

Should still be verified by a user having access to such a machine.

   ------- Additional comments from joga Wed Oct 2 02:25:54 -0700 2002 -------
A user (Laurent Aymeric) reported correct behaviour on a 6 cpu machine.
Thanks Laurent!
1 2 3 4 5 6 7 8 9 10 11 12 13
Note: See TracQuery for help on using queries.