问题
BASIC PROBLEM: I want to submit N + 1 jobs to an LSF-managed Linux cluster in such a way that the (N + 1)-st "wrap-up" job is not run until all the preceding N jobs have finished.
EXTRA: If possible, it would be ideal if I could arrange matters so that the (N + 1)-st ("wrap-up") job receives, as its first argument, a value of 0 (say) if all the previous N jobs terminated successfully, and a value different from 0 otherwise.
This problem (or at least the part labeled "BASIC PROBLEM") is vastly simpler than what LSF's bsub appears to be designed to handle, so I have a hard time wading through the voluminous documentation for bsub to figure out the simplest way to do what I want to do.
What would be the simplest
bsubcommands to achieve this arrangement?
To be more concrete, what would I have to replace the various ??? slots below to ensure that wrapup is executed only after all the foo jobs have finished (ideally with an argument that reflects the ending status of the foo jobs)?
bsub -q someq ??? foo 1
bsub -q someq ??? foo 2
bsub -q someq ??? foo 3
bsub -q someq ??? wrapup [???]
回答1:
To expand on Michael Closson's answer, what you're looking for here is bsub's -w option, which allows you to submit a job that will only be scheduled if some dependency condition is met.
The most common conditions to use is the exit status of some other job, if you name each of your "foo $i" jobs with -J:
bsub -q someq -J "job_1" foo 1
bsub -q someq -J "job_2" foo 2
bsub -q someq -J "job_3" foo 3
Then you can submit another job that depends on the exit status of these jobs as follows:
bsub -q someq -w "done(job_1) && done(job_2) && done(job_3)" wrapup
This tells LSF to only schedule "wrapup" if the jobs named job_1, job_2, and job_3 terminate with DONE status. You can also use job-id's instead of job names, or specify the specific status you want to test for with expressions like
done("job_1") // termination status is DONE
exit("job_1") // termination status is EXIT
ended("job_1") // termination status is EXIT or DONE
And combine these with logical operators &&, ||, !
来源：https://stackoverflow.com/questions/12944237/lsf-bsub-how-to-specify-a-single-wrap-up-job-to-be-run-after-all-others-fin