Spark: Broadcast variables: It appears that you are attempting to reference SparkContext from a broadcast variable, action, or transforamtion

Question

Class ProdsTransformer:

    def __init__(self):  
      self.products_lookup_hmap = {}
      self.broadcast_products_lookup_map = None

    def create_broadcast_variables(self):
      self.broadcast_products_lookup_map = sc.broadcast(self.products_lookup_hmap)

    def create_lookup_maps(self):
    // The code here builds the hashmap that maps Prod_ID to another space.

pt = ProdsTransformer ()
pt.create_broadcast_variables()  

pairs = distinct_users_projected.map(lambda x: (x.user_id,    
                         pt.broadcast_products_lookup_map.value[x.Prod_ID]))

I get the following error:

"Exception: It appears that you are attempting to reference SparkContext from a broadcast variable, action, or transforamtion. SparkContext can only be used on the driver, not in code that it run on workers. For more information, see SPARK-5063."

Any help with how to deal with the broadcast variables will be great!

Answer 1

By referencing the object containing your broadcast variable in your map lambda, Spark will attempt to serialize the whole object and ship it to workers. Since the object contains a reference to the SparkContext, you get the error. Instead of this:

pairs = distinct_users_projected.map(lambda x: (x.user_id, pt.broadcast_products_lookup_map.value[x.Prod_ID]))

Try this:

bcast = pt.broadcast_products_lookup_map
pairs = distinct_users_projected.map(lambda x: (x.user_id, bcast.value[x.Prod_ID]))

The latter avoids the reference to the object (pt) so that Spark only needs to ship the broadcast variable.