Just use a singleton or static variable. It will be a simple per-JVM value that is therefore per-executor.
On Thu, Jun 14, 2018 at 6:59 AM Nikodimos Nikolaidis <niknik...@csd.auth.gr> wrote: > Hello community, > > I am working on a project in which statistics (like predicate > selectivity) are collected during execution. I think that it's a good > idea to keep these statistics in executor level. So, all tasks in same > executor share the same variable and no extra network traffic is needed. > Also, I am not especially interested in thread safety, it's not a big > deal if some updates are lost - we are trying to see the general trend. > > This could be done, for example, with an in-memory data structure store > server like Redis in each worker machine. But, could it be done in Spark > natively? > > thanks, > nik > > > --------------------------------------------------------------------- > To unsubscribe e-mail: dev-unsubscr...@spark.apache.org > >