ActianCorp / spark-vector

Repository for the Spark-Vector connector
Apache License 2.0
20 stars 9 forks source link

Fix for Issue II-5795 #77

Closed fgropengieser closed 4 years ago

fgropengieser commented 4 years ago

The hardcoded number of partitions in conjunction with limiting the number of executors to significantly less than number of partitions leads to a problem on x100 side. If the filled data stream output queues are not fetched then a point is reached, where a partition queue is completely filled up. In this situation, no more input tuples can be processed and hence we have a deadlock situation.

The fixe tries to mitigate the problem by setting the number of partitions to the "expected" number of executors for the job.