[torqueusers] Non-cummulative pbsnodes -o command
jwang at dataseekonline.com
Thu Feb 14 15:58:22 MST 2008
Well, it's definitely non-cumulative with the -o option on both the 2.2.1
and 2.0.0p7 installs here. The -r option is cumulative, I use the -r in a
loop on the node names to reset the nodes and return them to being online as
it only marks them down till the next update from pbs_mom, if we use the -c
option to return the nodes to the online state, it may erroneously mark the
node online for a while if the node is non-responsive (actually I think the
-c would be just fine for returning nodes to service too).
I compiled the 2.2.1 installation myself so I know that it is in fact a
2.2.1 installation (unless Cluster Resources misnames their packages for
download). I do find it interesting that the 2.0.0p7 installation actually
states that the behaviour would be non-cumulative in the man page but I have
no idea if that was from the installation or a modified man page by the
prior support staff.
The intent is to wipeout the images of the legacy cluster once we work out
an image that we are happy with on the new cluster. There's plenty on the
legacy cluster that had not been well documented and was implemented in a
non-standard fashion, apparently the previous staffing took the BOFH
mentality to heart.
On 2/14/08 1:47 PM, "Garrick Staples" <garrick at usc.edu> wrote:
> On Thu, Feb 14, 2008 at 12:02:17PM -0600, John Wang alleged:
>> I have two clusters, one is running torque 2.0.0p7 and the other torque
>> 2.2.1. Both versions behave with the "pbsnodes -o" effect being
>> non-cummulative thereby requiring all nodes known to be down to be specified
>> on the command line. The 2.0.0p7 man page on pbsnodes actually state that
>> this would be the behaviour. The man page reads as follows:
>>> ³It is important that all the nodes known to be down are given as arguments
>>> the command line. This is because nodes which are not listed are assumed
>>> be UP and will be indicated as such if they were previously marked DOWN.²
> That is for -r, not -o. -o doesn't marks nodes offline, not down.
> I gaurantee you that -o does not work that way in your 2.2.1 install.
> I'm not going to comment on your ancient 2.0 install.
>> The 2.0.0p7 installation is a legacy install that predates the current
>> support staff so the history of that installation is unknown but the 2.2.1
>> installation was compiled from source files obtained from Cluster Resources
>> with default options.
> Well, upgrade from the ancient version to a current version?
> torqueusers mailing list
> torqueusers at supercluster.org
More information about the torqueusers