Details
Description
After a number of calls to a map().collect() statement Spark freezes without reporting any error. Within the map a large broadcast variable is used.
The freezing can be avoided by setting 'spark.python.worker.reuse = false' (Spark 1.2) or using an earlier version, however, at the prize of low speed.
Attachments
Issue Links
- relates to
-
SPARK-3993 python worker may hang after reused from take()
- Resolved
-
SPARK-3030 reuse python worker
- Resolved
- links to