HTCondor Project List Archives



[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Condor-devel] [condor-fw] RFC: convention for advertising SlotTypeID integer from SLOT_TYPE_N



Hi,

Perhaps related (or perhaps thread hijacking) - it would be really nice if the schedd recorded the name of the remote slot, instead of SlotID@Hostname.

Our admins have commented that, since we started the transition to p-slots, debugging has become much harder because we don't know which of the 24 running jobs on a host is the one they wanted to look at.

See below.  All jobs claim to be running in "slot1".

Brian

[root@t3 ~]# condor_q -run -g | grep red-d16n15
440478.0   cmsprod088      4/3  12:16   0+23:45:37 slot1@xxxxxxxxxxxxxxxxxxxxxxxxxx
440704.0   cmsprod088      4/3  14:24   0+21:53:13 slot1@xxxxxxxxxxxxxxxxxxxxxxxxxx
441013.0   cmsprod088      4/3  15:26   0+20:46:28 slot1@xxxxxxxxxxxxxxxxxxxxxxxxxx
441323.0   cmsprod088      4/3  17:26   0+16:45:06 slot1@xxxxxxxxxxxxxxxxxxxxxxxxxx
441342.0   cmsprod088      4/3  17:28   0+16:35:52 slot1@xxxxxxxxxxxxxxxxxxxxxxxxxx
441510.0   cmsprod088      4/3  19:44   0+10:59:47 slot1@xxxxxxxxxxxxxxxxxxxxxxxxxx
442242.0   cmsprod088      4/4  03:36   0+00:10:25 slot1@xxxxxxxxxxxxxxxxxxxxxxxxxx
442400.0   uscmsPool2190   4/4  04:12   0+09:22:24 slot1@xxxxxxxxxxxxxxxxxxxxxxxxxx
442567.0   uscmsPool1848   4/4  05:16   0+06:48:09 slot1@xxxxxxxxxxxxxxxxxxxxxxxxxx
442763.0   uscmsPool1404   4/4  07:21   0+06:35:18 slot1@xxxxxxxxxxxxxxxxxxxxxxxxxx
442792.0   uscmsPool1404   4/4  07:22   0+06:31:35 slot1@xxxxxxxxxxxxxxxxxxxxxxxxxx
471529.0   cmsprod088      3/31 02:54   3+08:07:42 slot1@xxxxxxxxxxxxxxxxxxxxxxxxxx
471770.0   cmsprod088      3/31 05:45   3+04:19:02 slot1@xxxxxxxxxxxxxxxxxxxxxxxxxx
472002.0   cmsprod088      3/31 08:55   2+23:43:04 slot1@xxxxxxxxxxxxxxxxxxxxxxxxxx
472778.0   cmsprod088      3/31 18:38   2+11:32:22 slot1@xxxxxxxxxxxxxxxxxxxxxxxxxx
473191.0   cmsprod088      4/1  00:43   2+05:11:47 slot1@xxxxxxxxxxxxxxxxxxxxxxxxxx
481098.0   uscmsPool1404   4/4  07:21   0+06:42:44 slot1@xxxxxxxxxxxxxxxxxxxxxxxxxx
481941.0   cmsprio155      4/4  13:03   0+01:00:13 slot1@xxxxxxxxxxxxxxxxxxxxxxxxxx
835011.0   uscmsPool1428   4/3  04:02   1+10:02:58 slot1@xxxxxxxxxxxxxxxxxxxxxxxxxx
835041.0   cmsprod088      4/3  04:14   1+05:02:16 slot1@xxxxxxxxxxxxxxxxxxxxxxxxxx
836327.0   cmsprod088      4/3  15:55   0+20:41:18 slot1@xxxxxxxxxxxxxxxxxxxxxxxxxx
836462.0   cmsprod088      4/3  17:10   0+16:48:41 slot1@xxxxxxxxxxxxxxxxxxxxxxxxxx
837702.0   uscmsPool2703   4/4  11:16   0+00:23:46 slot1@xxxxxxxxxxxxxxxxxxxxxxxxxx
837948.0   uscmsPool283    4/4  12:43   0+00:00:21 slot1@xxxxxxxxxxxxxxxxxxxxxxxxxx
[root@t3 ~]# condor_q -g -l 441342.0 | grep RemoteHost

On Apr 4, 2012, at 2:05 PM, Erik Erlandson wrote:



From: Erik Erlandson <eje@xxxxxxxxxx>
Subject: [condor-fw] RFC: convention for advertising SlotTypeID integer from SLOT_TYPE_N
Date: April 4, 2012 1:28:56 PM CDT
To: Condor Flightworthy Team <condor-fw@xxxxxxxxxxx>
Reply-To: eje@xxxxxxxxxx


As part of upcoming RFE:
https://condor-wiki.cs.wisc.edu/index.cgi/tktview?tn=2905

I'm planning to also advertise the 'slot type id' on slot ads, which is
defined as:

SlotTypeID = n  # where n corresponds to SLOT_TYPE_n

The proposed convention for dynamic slots is that they will advertise:
SlotTypeID = -n  # my parent p-slot is of type n

One consequence of that scheme is that if anybody ever wished to
directly submit against a particular p-slot type, they would have to do
this:

requirements = (SlotTypeID == 4  ||  SlotTypeID == -4)

The reason is that the requirements have to match both the p-slot and
then the d-slot after it is created.

Currently, I'm of the mind that it ought to be preferred practice to use
slot properties and resources in one's requirements, and let the
negotiation do its job, as opposed to directly requesting slot types
(although I did find submitting to specific slot types useful in
constructing my tests for the new features).


_______________________________________________
condor-fw mailing list
condor-fw@xxxxxxxxxxx
https://lists.cs.wisc.edu/mailman/listinfo/condor-fw


_______________________________________________
Condor-devel mailing list
Condor-devel@xxxxxxxxxxx
https://lists.cs.wisc.edu/mailman/listinfo/condor-devel