if there is already a running executor with the given executor id, the slave sends the task to the executor. otherwise, it will launch a new executor and then send the task to it.
On Tue, Jun 4, 2013 at 1:56 AM, 王国栋 <[email protected]> wrote: > Hi guys, > > I find that spark only launches one mesos-executor on each slave node. But > hadoop will launch multiple tasktrackers on each slave node as long as > resource is enough. > > I goes through the spark code. But I have some questions. > when the scheduler launch mesos tasks, slave checks the executor info which > is given by scheduler. If 1. there is no executor with the given > executorID, what will slave do? > 2. if there is executor with the given executorID, what will slave do? > > I am doing some test to try to understand the detailed process of this. > Especially the callback sequences on MesosExecutor. > > Can anybody explain this to me? I am looking forward to the discussion. > > Thanks. > > > Guodong >
