[jira] [Created] (CASSANDRA-12707) JVM out of memory when querying an extra-large partition with lots of tombstones

["nicolas ginder (JIRA)" <ji...@apache.org>]

nicolas ginder created CASSANDRA-12707:
------------------------------------------

             Summary: JVM out of memory when querying an extra-large partition with lots of tombstones
                 Key: CASSANDRA-12707
                 URL: https://issues.apache.org/jira/browse/CASSANDRA-12707
             Project: Cassandra
          Issue Type: Bug
          Components: Core
            Reporter: nicolas ginder


We have an extra large partition of 40 million cells where most of the cells were deleted. When querying this partition, Cassandra runs out of memory as tombstones fill up the JVM heap. After debugging one of the large SSTable we found that this part of the code loads all the tombstones.
In org.apache.cassandra.db.filter.QueryFilter
...
public static Iterator<Cell> gatherTombstones(final ColumnFamily returnCF, final Iterator<? extends OnDiskAtom> iter)
    {
...
while (iter.hasNext())
                {
                    OnDiskAtom atom = iter.next();

                    if (atom instanceof Cell)
                    {
                        next = (Cell)atom;
                        break;
                    }
                    else
                    {
                        if(returnCF.deletionInfo().rangeCount() > DatabaseDescriptor.getTombstoneFailureThreshold()){
                            throw new TombstoneOverwhelmingException();
                        }
                        returnCF.addAtom(atom);
                    }
                }
...




--
This message was sent by Atlassian JIRA
(v6.3.4#6332)