You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@spark.apache.org by "James Baker (JIRA)" <ji...@apache.org> on 2017/07/05 14:21:00 UTC
[jira] [Created] (SPARK-21319)
UnsafeExternalRowSorter.RowComparator memory leak
James Baker created SPARK-21319:
-----------------------------------
Summary: UnsafeExternalRowSorter.RowComparator memory leak
Key: SPARK-21319
URL: https://issues.apache.org/jira/browse/SPARK-21319
Project: Spark
Issue Type: Bug
Components: SQL
Affects Versions: 2.1.0, 2.0.0, 2.2.0, 2.3.0
Reporter: James Baker
When we wish to sort within partitions, we produce an UnsafeExternalRowSorter. This contains an UnsafeExternalSorter, which contains the UnsafeExternalRowComparator.
The UnsafeExternalSorter adds a task completion listener which performs any additional required cleanup. The upshot of this is that we maintain a reference to the UnsafeExternalRowSorter.RowComparator until the end of the task.
The RowComparator looks like
{code:java}
private static final class RowComparator extends RecordComparator {
private final Ordering<InternalRow> ordering;
private final int numFields;
private final UnsafeRow row1;
private final UnsafeRow row2;
RowComparator(Ordering<InternalRow> ordering, int numFields) {
this.numFields = numFields;
this.row1 = new UnsafeRow(numFields);
this.row2 = new UnsafeRow(numFields);
this.ordering = ordering;
}
@Override
public int compare(Object baseObj1, long baseOff1, Object baseObj2, long baseOff2) {
// TODO: Why are the sizes -1?
row1.pointTo(baseObj1, baseOff1, -1);
row2.pointTo(baseObj2, baseOff2, -1);
return ordering.compare(row1, row2);
}
}
{code}
which means that this will contain references to the last baseObjs that were passed in, and without tracking them for purposes of memory allocation.
We have a job which sorts within partitions and then coalesces partitions - this has a tendency to OOM because of the references to old UnsafeRows that were used during the sorting.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org