I'm trying to get spark to read uncompressed thrift files from s3. So far it has not been working.
- data is loaded in s3 as uncompressed thrift files. The source is AWS Kinesis Firehose.
- I have a tool that deserializes files with no problem, so I know that thrift serialization/deserialization works.
- in spark, im using newAPIHadoopFile
- using elephantbird's LzoThriftBlockInputFormat, I am able to successfully read lzo-compressed thrift files
- I can't figure out what InputFormat should I use to read uncompressed thrift files.
Is that possible with any of the InputFormats out there? Do I have to implement my own?
I ended up writing my own custom thrift deserializer.
Needed to implement a custom InputFormat and custom RecordReader. Still surprised that such classes don't already exist in some lib. The two classes have been tested and work, but since i stopped working on the project soon after i solved this, the code is not cleaned up.
https://github.com/mklosi/thrift-deserializer