[DRILL-2758] Memory leak in parquet writer when writing billions of records

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 0.8.0
• Fix Version/s: 0.9.0
• Component/s: Storage - Parquet
• Assignee: Steven Phillips
• Reporter: Aman Sinha
• Created: 10/Apr/15 21:13
• Updated: 14/Apr/15 01:49
• Resolved: 14/Apr/15 01:49

Description

Encountered the following memory leak when running a CTAS creating parquet data. This is on a large data set, so I cannot provide the reproduction here. Several billions of records are written and all fragments except 1 show FINIISHED state in the profile. The 1 fragment that is in RUNNING state shows 0 records written. The jstack on that node showed no activity and cpu was idle. However, drillbit.log shows the following memory leak:

2015-04-10 18:11:40,337 [2ad7efdb-37ad-743e-3fe4-54a9d158d843:frag:1:40] WARN  o.a.d.e.w.fragment.FragmentExecutor - Failure while closing out resources
java.lang.IllegalStateException: Attempted to close accountor with 1 buffer(s) still allocatedfor QueryId: 2ad7efdb-37ad-743e-3fe4-54a9d158d843, MajorFragmentId: 1, MinorFragmen
tId: 40.


        Total 1 allocation(s) of byte size(s): 1153433, at stack location:
                org.apache.drill.exec.memory.TopLevelAllocator$ChildAllocator.buffer(TopLevelAllocator.java:234)
                org.apache.drill.exec.store.parquet.ParquetDirectByteBufferAllocator.allocate(ParquetDirectByteBufferAllocator.java:45)
                parquet.bytes.CapacityByteArrayOutputStream.allocateSlab(CapacityByteArrayOutputStream.java:74)
                parquet.bytes.CapacityByteArrayOutputStream.initSlabs(CapacityByteArrayOutputStream.java:88)
                parquet.bytes.CapacityByteArrayOutputStream.<init>(CapacityByteArrayOutputStream.java:69)
                parquet.column.values.plain.PlainValuesWriter.<init>(PlainValuesWriter.java:48)
                parquet.column.ParquetProperties.getValuesWriter(ParquetProperties.java:109)
                parquet.column.impl.ColumnWriterImpl.<init>(ColumnWriterImpl.java:81)
                parquet.column.impl.ColumnWriteStoreImpl.newMemColumn(ColumnWriteStoreImpl.java:68)
                parquet.column.impl.ColumnWriteStoreImpl.getColumnWriter(ColumnWriteStoreImpl.java:56)
                parquet.io.MessageColumnIO$MessageColumnIORecordConsumer.<init>(MessageColumnIO.java:124)
                parquet.io.MessageColumnIO.getRecordWriter(MessageColumnIO.java:315)
                org.apache.drill.exec.store.parquet.ParquetRecordWriter.newSchema(ParquetRecordWriter.java:165)
                org.apache.drill.exec.store.parquet.ParquetRecordWriter.updateSchema(ParquetRecordWriter.java:141)
                org.apache.drill.exec.physical.impl.WriterRecordBatch.setupNewSchema(WriterRecordBatch.java:162)
                org.apache.drill.exec.physical.impl.WriterRecordBatch.innerNext(WriterRecordBatch.java:113)
                org.apache.drill.exec.record.AbstractRecordBatch.next(AbstractRecordBatch.java:142)
                org.apache.drill.exec.physical.impl.validate.IteratorValidatorBatchIterator.next(IteratorValidatorBatchIterator.java:118)
                org.apache.drill.exec.physical.impl.BaseRootExec.next(BaseRootExec.java:68)
                org.apache.drill.exec.physical.impl.SingleSenderCreator$SingleSenderRootExec.innerNext(SingleSenderCreator.java:99)