If your real problem is a bad client you do not want running jobs (or do not wish them to be granted all resources when they do), why not tackle just that instead of "working-around"?
Hadoop allows authorization of users, and MR schedulers also allow restricting submissions to defined queues/pools, aside of letting you configure cluster resources per user/job. HDFS also carry permission features to prevent global write access to directories, if that is also your issue. On Sun, Mar 25, 2012 at 8:08 PM, shlomi java <shlomij...@gmail.com> wrote: > hi all, > > I want to figure out, from a client of Hadoop cluster, the statuses of jobs > that are currently running on the cluster. > I need it in order to prevent the client from submitting certain jobs to > the cluster, when some certain jobs are already running on the cluster. > I know to recognize my jobs - by their name. > > How do I do it? > > I saw the JobTracker code in jobtracker.jsp. I would be happy to use its > 'runningJobs' method. > The thing is that in the JSP, which is executed on the cluster (master), we > get hold of the JobTracker from the application context. > > Is it safe to instantiate my own JobTracker, given the right conf, from the > client? > > thanks > ShlomiJ -- Harsh J
