Details
-
Bug
-
Status: Closed
-
Critical
-
Resolution: Fixed
-
None
Description
Problem:
Some corrupted parquet files are generated and exceptions will be thrown when read.
e.g.
Caused by: org.apache.parquet.io.ParquetDecodingException: Can not read value at 0 in block -1 in file <FilePath>
at org.apache.parquet.hadoop.InternalParquetRecordReader.nextKeyValue(InternalParquetRecordReader.java:251)
at org.apache.parquet.hadoop.ParquetReader.read(ParquetReader.java:132)
at org.apache.parquet.hadoop.ParquetReader.read(ParquetReader.java:136)
at org.apache.hudi.common.util.ParquetReaderIterator.hasNext(ParquetReaderIterator.java:49)
at org.apache.hudi.common.util.queue.IteratorBasedQueueProducer.produce(IteratorBasedQueueProducer.java:45)
at org.apache.hudi.common.util.queue.BoundedInMemoryExecutor.lambda$null$0(BoundedInMemoryExecutor.java:112)
at java.util.concurrent.FutureTask.run(FutureTask.java:266)
at java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:511)
... 4 more
Caused by: org.apache.parquet.io.ParquetDecodingException: could not read page Page [bytes.size=1054316, valueCount=237, uncompressedSize=1054316] in col required binary col
at org.apache.parquet.column.impl.ColumnReaderImpl.readPageV1(ColumnReaderImpl.java:599)
at org.apache.parquet.column.impl.ColumnReaderImpl.access$300(ColumnReaderImpl.java:57)
at org.apache.parquet.column.impl.ColumnReaderImpl$3.visit(ColumnReaderImpl.java:536)
at org.apache.parquet.column.impl.ColumnReaderImpl$3.visit(ColumnReaderImpl.java:533)
at org.apache.parquet.column.page.DataPageV1.accept(DataPageV1.java:95)
at org.apache.parquet.column.impl.ColumnReaderImpl.readPage(ColumnReaderImpl.java:533)
at org.apache.parquet.column.impl.ColumnReaderImpl.checkRead(ColumnReaderImpl.java:525)
at org.apache.parquet.column.impl.ColumnReaderImpl.consume(ColumnReaderImpl.java:638)
at org.apache.parquet.column.impl.ColumnReaderImpl.<init>(ColumnReaderImpl.java:353)
at org.apache.parquet.column.impl.ColumnReadStoreImpl.newMemColumnReader(ColumnReadStoreImpl.java:80)
at org.apache.parquet.column.impl.ColumnReadStoreImpl.getColumnReader(ColumnReadStoreImpl.java:75)
at org.apache.parquet.io.RecordReaderImplementation.<init>(RecordReaderImplementation.java:271)
at org.apache.parquet.io.MessageColumnIO$1.visit(MessageColumnIO.java:147)
at org.apache.parquet.io.MessageColumnIO$1.visit(MessageColumnIO.java:109)
at org.apache.parquet.filter2.compat.FilterCompat$NoOpFilter.accept(FilterCompat.java:165)
at org.apache.parquet.io.MessageColumnIO.getRecordReader(MessageColumnIO.java:109)
at org.apache.parquet.hadoop.InternalParquetRecordReader.checkRead(InternalParquetRecordReader.java:137)
at org.apache.parquet.hadoop.InternalParquetRecordReader.nextKeyValue(InternalParquetRecordReader.java:222)
... 11 more
Caused by: java.io.EOFException
at java.io.DataInputStream.readFully(DataInputStream.java:197)
at java.io.DataInputStream.readFully(DataInputStream.java:169)
at org.apache.parquet.bytes.BytesInput$StreamBytesInput.toByteArray(BytesInput.java:286)
at org.apache.parquet.bytes.BytesInput.toByteBuffer(BytesInput.java:237)
at org.apache.parquet.bytes.BytesInput.toInputStream(BytesInput.java:246)
at org.apache.parquet.column.impl.ColumnReaderImpl.readPageV1(ColumnReaderImpl.java:592)
How to reproduce:
We need a way which could interrupt one task w/o shutdown JVM. Let's say, speculation. When speculation is triggered, other tasks working at the same executor will have the risk to suffer a wrong parquet generation. This will not always result in corrupted parquet file. Nearly half of them will throw exception while there is few tasks succeed without any signal.
RootCause:
ParquetWriter is not thread safe. User of it should apply proper way to guarantee that there is not concurrent call to ParquetWriter.
In the following code:
We call both write and close to parquet writer concurrently. Data may being written while we call close. In close method, compressor (a class used by parquet to do compressing which has a stateful data structure insied) will be cleared and payback to a pool for following reuse. Due to the concurrent write mentioned above, data may be continued pushed to compressor even though we have them cleared. Besides, there is a mechanism inside compressor which tries to check some invalid use. That's why some of invalid usage will throw exception rather than generate corrupted parquet.
Validation:
Current solution is validated by production environment. A signal is that when this fix applied is that there should be no task failed due to some error like "BlockCompressorStream: write beyond end of stream". The reason is that BlockCompressorStream checking mechanism will not be triggered by concurrent write.
