This has been fixed to a better error message when this situation happens. I am surprised no one else has ever seen this since it would of happened long ago on systems when the frontend wasn't set up correctly.
On another note, James it appears you at one point anyway were using 2.5. If you haven't already moved back to 2.4 please do. 2.5 is under development and should not be used on any production systems. The underlying structures can/will change during development and job loss can occur when upgrading to future 2.5 versions. If you have already moved back to 2.4, perfect and no worries :). Danny On 08/01/12 08:02, James Sweet wrote: > Mark/Carl, > > slurmctld is running now and talking to the bluegen. I'm currently trying out > some test jobs. > > Thanks very much to you both for your help and expertise. > > James > > On 01/08/2012 02:31, Mark Nelson wrote: >> Hi Carl, >> >> On 01/08/12 05:47, Carl Schmidtmann wrote: >>> ----- Original Message ----- >>>> Carl, 2.3 does not work on a Q so you should avoid it. >>> I was about to say the same thing for 2.4.1 ;). >>> >>> James, >>> >>> I think I have a partial answer to your problem. You need to specify a >>> FrontendNode. Without that slurmd uses the NodeName as a FrontendNode and >>> tries to add it to the hash table creating the duplicate error. After I >>> added a FrontendNode that problem went away. >> Ah yes, that is indeed your problem. Sorry, I skimmed the slurm.conf and >> it didn't jump out at me that you were missing the definition of the >> frontend node. If I comment out the appropriate line in my slurm.conf >> (FrontendName=slurm-dev State=UNKNOWN), I get exactly the Duplicated >> NodeName error that you're seeing. >> >>> Now I am trying to figure out how a bluegene.conf file, or just the blocks >>> section, can be generated by smap. Several places say to let smap create >>> the file but there is nothing in the smap man page nor the bluegene.conf >>> man page about how to do this. Slurmd won't run without bluegene.conf and >>> smap won't run without slurmd. So how can it generate the block >>> specifications to put into the bluegene.conf file? >> You should be able to start smap using the -Dc option without having any >> of the daemons running (although you will need a valid slurm.conf). And >> then when the curses interface opens up you can use the create option to >> create blocks. For example, call "create 1" eight times to create eight >> midplane blocks. Then you can use "save bluegene.conf". >> >> Oh, also, on our system we had to set the MloaderImage to be >> /bgsys/drivers/ppcfloor/boot/firmware rather than uloader in the >> bluegene.conf. >> >> Let us know how you go. >> >> Thanks, and good luck! >> Mark >> >>> I am going to create a file based on Mark's example file just to get slurmd >>> running and then see what smap will show me. >>>
