Hi Barry, The STDOUT files location issue can be solved by changing
Line 1046 in experiment.perl from -- `qsub $qsub_args -e $step.STDERR $step -o $step.STDOUT`; to ++ `qsub $qsub_args -e $step.STDERR -o $step.STDOUT $step`; The old line puts -o option after commands, when qsub doesn't look after commands for any more options. So the stdout files of SGE go to the home directory of the execution nodes by default. About the PATH=..... : command not found issue, is it safe to remove it from the script? I think it's trying to reset the PATH environment variable of all the execution nodes to be same as that of the head node, which is not permitted in bash as I understand. Have a nice weekend. Many thanks, Guchun On 2 February 2012 14:14, Guchun Zhang <[email protected]> wrote: > Hi Barry, > > Many thanks for the answers. They are very helpful. :-) > > Regards, > > Guchun > > > On 2 February 2012 13:53, Barry Haddow <[email protected]> wrote: > >> Hi Guchun >> >> On Thursday 02 Feb 2012 12:34:09 Guchun Zhang wrote: >> > Hi Barry, >> > >> > It still crashes after the run-giza and run-giza-inverse step is >> completed. >> > At the moment, I simply deleted the "not found" line in .digest files >> and >> > let the experiment continue. So far so good. >> >> Maybe it's the 'PATH=' statement that's causing the error messages then >> ... >> I'm not sure. If you locate the run-giza and run-giza-inverse stanzas in >> experiment.meta, and remove the 'error: not found' lines then it should >> ignore these error messages. >> >> > >> > I noticed in the log files (in steps directory) of my previous >> successful >> > runs (on a single multi-core machine) that in some steps the STDERR >> files >> > were not empty with step-being-successful kind of information and the >> > STDOUT files sometimes contained little such information. Is this >> normal? >> > >> >> Yes, the steps are quite variable in their debug messages. Some don't >> write >> anything to stdout, but most write something to stderr. experiment.perl >> just >> checks the stderr for certain types of error messages. >> >> > About the location of the returned STDOUT from SGE, I normally either >> set >> > -cwd in the script or use the option in the command. So I tried to add >> it >> > to general-parallelizer.perl. But it made no effect. Do you in any >> chance >> > know how to do it? >> >> The qsub options are probably set in experiment.perl, but I don't know the >> details, >> >> cheers - Barry >> >> > >> > Many thanks, >> > >> > Guchun >> > >> > On 1 February 2012 17:30, Guchun Zhang <[email protected]> wrote: >> > > Hi Guchun >> > > >> > > So it looks as though this line in the script >> > > >> > > echo 'starting at '`date`' on '`hostname` >> > > >> > > is giving you a harmless 'not found' error (unless it's the path >> line, as >> > > Hieu >> > > suggested). When ems parses the STDERR file it sees this, and thinks >> giza >> > > crashed. But from the output you posted, giza looks to have run >> normally. >> > > >> > > Could you try removing the following line from experiment.perl? >> > > >> > > print STEP "echo 'starting at '`date`' on '`hostname`\n"; >> > > >> > > It's line 2580 in my copy, >> > > >> > > cheers - Barry >> > > >> > > On Wednesday 01 February 2012 15:21:53 you wrote: >> > > > Hi Barry, >> > > > >> > > > Here they are. >> > > >> > > -- >> > > The University of Edinburgh is a charitable body, registered in >> > > Scotland, with registration number SC005336. >> > >> >> >
_______________________________________________ Moses-support mailing list [email protected] http://mailman.mit.edu/mailman/listinfo/moses-support
