In terms of setup, we've written the Scheduler/surrounding system in Scala. For every job we want to run, we instantiate a new scheduler and MesosSchedulerDriver, and asynchronously call the normal driver.run(). The next event we expect to happen is for the scheduler's registered() method to be called. This normally occurs as expected, but after some time running multiple jobs, there will be approximately a 30-minute delay between the call to driver.run() and the registered() method being called (based on logs).
Another thing we've noticed is that the number of threads used by the process increases as more jobs are run. Does calling driver.stop() terminate any threads launched for calling the native Mesos code through JNI? Or are additional steps required? One more possible weirdness -- would anything odd happen if driver.stop() is called more than once for a particular driver? Thanks! On Sat, Oct 4, 2014 at 12:04 AM, Tim Chen <[email protected]> wrote: > I don't think there should be any functional problem running multiple > schedulers on the same machine, > > can you describe the buggy behaviors you're seeing and what the setup is > like? > > Tim > > > > On Fri, Oct 3, 2014 at 10:54 PM, Colleen Lee <[email protected]> wrote: > >> We've been designing our system on top of Mesos, and in our current >> design, we've been allocating a new Scheduler (and MesosSchedulerDriver) >> for each new job. Are there any problems with running multiple schedulers >> on a single machine, or multiple schedulers run under the same process? >> We've noticed some strange/buggy behavior with this setup, and we were >> wondering if this could be the cause. Thanks! >> > >
