On Sat, Mar 7, 2015 at 4:17 PM, Jean-Pascal Billaud <j...@tellapart.com> wrote: > So given this let's go a bit further. Imagine my static factory provides a > stats collector that my various map() code would use to export some metrics > while mapping tuples. This stats collector comes with a timer that flush the > stats buffer once every 5mn for instance. Given that executor JVM should not > be reinitialized over the life of the DStream in practice, it is totally > reasonable to assume that this timer will be able to do his job. Right?
Ignoring failures, the executors should not be restarted over the life of the streaming app, yes. You will have, potentially, many executors at once of course. > Given what you said earlier that totally makes sense. In general is there any > spark architecture documentation other than the code that gives a good > overview of the thing we talked about? http://spark.apache.org/docs/latest/cluster-overview.html This is entirely applicable to streaming too. --------------------------------------------------------------------- To unsubscribe, e-mail: user-unsubscr...@spark.apache.org For additional commands, e-mail: user-h...@spark.apache.org
