Hi Martin, thanks for your prompt response! I suppose, that you are right with your guess. Maybe the SEG does not work correctly...
In the meanwhile I have tested the following: After submitting a job via globus-ws, the job will be executed by lsf. Afterwards there is an entry in the logfile of lsf (/opt/hptc/lsf/top/work/hptclsf/logdir/lsb.acct). This entry seems to be equal (or similar) to the "old" lsf logfile. Mainly the version number of lsf is different. In /opt/globus/gt4/etc/globus-lsf.conf the right location of the lsf logfile is given: "log_path=/opt/hptc/lsf/top/work/hptclsf/logdir". How can I run the SEG manually? I have tried "/opt/globus/gt4/libexec/globus-scheduler-event-generator -s LSF", but without success and with an error message: globus_scheduler_event_generator: Unable to dlopen module "/opt/globus/gt4/lib/libglobus_seg_LSF_gcc64dbg.la": file not found Btw: We use Globus 4.0.8 (I did not mentioned it in the last post. Regards, Benjamin > -----Ursprüngliche Nachricht----- > Von: Martin Feller [mailto:[email protected]] > Gesendet: Freitag, 30. Juli 2010 14:50 > An: Löhnhardt, Benjamin > Cc: [email protected] > Betreff: Re: [gt-user] globus-ws with lsf does not work > > Hi, > > Just an educated guess: > I assume the problem is the scheduler event generator (SEG). > In Gram4, and I think also in Gram5, the SEG is responsible for > telling Gram about the status of the jobs in the job manager. > If the SEG doesn't tell Gram about job status, the job doesn't make > any progress from Gram's perspective. > > I think the SEG works on the log files of the job managers to get the > status information about the jobs. > If something changed in the logging format of the job manager, the SEG > may not be able to get the information anymore. > To confirm this I would probably run the SEG by hand and submit a > Gram job to old/new LSF and check if the SEG actually spits out > information on job status as it is processed in the job manager. > > Martin > > Löhnhardt, Benjamin wrote: > > Hello, > > > > we have a problem with globus and LSF as job manager. When we execute > a > > globus job via globus-ws on a client, on the server the job will be > handled > > by the lsf job manager "normally". The job even will be executed by > lsf > > itself, so that the test script was run on the server. However, the > client > > does not notice that the script was run successfully and waits. The > output on > > the client: > > > > -bash-3.1$ globusrun-ws -submit -s -F https://nimrod.med.uni- > goettingen.de > > -Ft LSF -c /tmp/test.sh > > Delegating user credentials...Done. > > Submitting job...Done. > > Job ID: uuid:c06227cc-9bc6-11df-80d4-00215af48192 > > Termination time: 07/31/2010 10:39 GMT > > > > We have updated the lsf system on the server from 6.2 to 7.0. Has > anybody a > > hint why the client is waiting of a response by the server? How can > we fix > > this issue? > > > > Regards, > > Benjamin > > > > > >
