You probably want python-snappy

There is one case where you might prefer this library: reading Hadoop snappy streams.

The Hadoop snappy format allows a "block" to contain one or more subblocks. Real Hadoop writers (BlockCompressorStream) emit multi-subblock blocks when an input exceeds the configured buffer size.

• pysnappy's HadoopDecompressor handles both single-subblock and multi-subblock blocks.
• python-snappy's HadoopStreamDecompressor only handles single-subblock blocks; on multi-subblock input it silently returns the first subblock and stashes the rest in an internal buffer that is never surfaced. Data loss with no error.

For the encoder, pysnappy emits multi-subblock blocks by default (respecting the buffer_size argument). Pass single_subblock=True to the HadoopCompressor constructor to emit one subblock per call, which is byte-shape-compatible with python-snappy's encoder and slightly faster.