Hi Bashir,
I think it is just a case of somebody bothering to plumbing it in explicitly,
e.g.
/** Specifies row group size. By default, DEFAULT_BLOCK_SIZE. */
public Sink withRowGroupSize(int rowGroupSize) {
return toBuilder().setRowGroupSize(rowGroupSize).build();
}
and
this.writer =
AvroParquetWriter.<GenericRecord>builder(beamParquetOutputFile)
.withRowGroupSize(getRowGroupSize()) // Ze patch to set RowGroupSize
.withSchema(schema)
.withCompressionCodec(getCompressionCodec())
.withWriteMode(OVERWRITE)
.build();
Etc.
However, it might worth exploring if it can be set via the HadoopConfiguration
“parquet.block.size” property, but I’m not sure that it actually can.
We patched in something explicitly last year but didn’t contribute upstream as
there was quite a bit of activity on the ParquetIO (e.g. conversion to SDF) at
the time.
The use case we had at the time was that some downstream consumers of the
parquet (AWS S3 Select) couldn’t handle rowGroupSizes > 64MB uncompressed. I’m
sure there are other use cases out there that need this fined grained control.
Cheers, David
David Hollands
BBC Broadcast Centre, London, W12
Email: david.holla...@bbc.co.uk<mailto:david.holla...@bbc.co.uk>
From: Bashir Sadjad <bas...@google.com>
Reply to: "user@beam.apache.org" <user@beam.apache.org>
Date: Friday, 12 March 2021 at 07:58
To: "user@beam.apache.org" <user@beam.apache.org>
Subject: Setting rowGroupSize in ParquetIO
Hi all,
I wonder how I can set the row group size for files generated by
ParquetIO.Sink<https://beam.apache.org/releases/javadoc/2.20.0/org/apache/beam/sdk/io/parquet/ParquetIO.Sink.html>.
It doesn't seem to provide the option for setting that and IIUC from the
code<https://github.com/apache/beam/blob/fffb85a35df6ae3bdb2934c077856f6b27559aa7/sdks/java/io/parquet/src/main/java/org/apache/beam/sdk/io/parquet/ParquetIO.java#L1117>,
it uses the default value in
ParquetWriter.Builder<https://github.com/apache/parquet-mr/blob/bdf935a43bd377c8052840a4328cf5b7603aa70a/parquet-hadoop/src/main/java/org/apache/parquet/hadoop/ParquetWriter.java#L636>
which is 128MB. Is there any reason not to expose this parameter in ParquetIO?
Thanks
-B
