January 2019
Beginner to intermediate
154 pages
4h 31m
English
If your jobs are running slowly, and from Spark UI you find out that some of the tasks are running slower when compared to others, then you might consider enabling speculation execution for these tasks. If the property spark.speculation is set to true, then Spark will identify the slow running tasks and run the speculative tasks on other nodes to complete the job more quickly. The tasks that finish first are accepted and the second task gets killed by Spark.
The following code sets this property to true:
#Pythonconf.set("spark.speculation","true")
Read now
Unlock full access