I had a server problem that use is ejecting my job. The server adm sent me an email explaining the problem.
"Basically what is happening is as follows:
When the job execution, a large amount of information is written to standard output. If the program was run on a standard PC, this data would be sent directly to the computer screen. How it is being run through the PBs, unlike the screen, this data is written to a file belonging to the queue manager itself. Over time, this file grows significantly, to the point of taking up all the "/ var" the server on which the job runs. To solve this problem, you should look at the Rosetta documentation a way to suppress this information, or if they are needed, a way of direct writing to another file located in the "/ data"."
and now? any tips?
If you are in MPI mode and want the data, -mpi_tracer_to_file [filename_stub] will write each process's standard out to its own file.
You can also use -mute all to supress nearly all output to standard out. -mute and -unmute take tracer names as arguments (the stuff that prepends most likes of output, like "core.pack.pack_rotamers" or whatever). So you can -mute all to supress all output then -unmute certain types of output you may be interested in.
Another option is to use the "-out:levels" option to change how "chatty" the logging output is. The default level is "info", which may be more than you need for production runs. Typically for production runs, I turn most things down to "warning", and then selectively increase for specific tracers I'm interested in. (For example, setting "core.init" and "protocols.jd2.JobDistributor" to "info" includes reproducibility information, like commandlines, random number seeds, and the completion status of individual jobs.) - valid settings, in order of increasing verbosity are 'error', 'warning', 'info', 'debug', 'trace'
The benefit of using "-out:levels" over mute/unmute is that mute will mute all messages, including warnings and errors. Having the warnings present, even for tracers you're normally not interested in, can help later on to track down what might be going wrong with your runs,