Table of Contents

Stored field compression

Stored fields are written in blocks under the .fdt extension and compressed.

Choose a policy

var config = new IndexWriterConfig
{
    CompressionPolicy = FieldCompressionPolicy.Deflate, // default
    StoredFieldBlockSize = 16,                          // docs per block
};
Policy Package Notes
None Core No compression. Fastest write, largest disk.
Deflate (default) Core BCL DeflateStream. Good ratio, no native deps.
Brotli Core BCL BrotliStream. Better ratio, slower writes.
Lz4 Rowles.LeanCorpus.Compression.LZ4 Very fast, modest ratio.
Snappy Rowles.LeanCorpus.Compression.Snappy Similar speed to LZ4.
Zstandard Rowles.LeanCorpus.Compression.Zstandard Better ratio than LZ4, still fast.

The policy is recorded in the segment header, so reads tolerate mixed segments.

Optional codec packages

Install and register a codec package to use LZ4, Snappy, or Zstandard:

// In standard .NET the module initialiser registers the codec automatically.
// In Native AOT, call Register() explicitly at startup.
Lz4Compression.Register();
SnappyCompression.Register();
ZstandardCompression.Register();

Block size

StoredFieldBlockSize (default 16) controls how many documents share a compression block. Larger blocks compress better but cost more on single-document retrieval.

Trade-offs

  • Indexing throughput: None > Lz4 ≈ Snappy > Deflate > Zstandard > Brotli.
  • On-disk size: Brotli ≈ Zstandard < Deflate < Lz4 ≈ Snappy < None.
  • Retrieval cost scales with block size, not policy.

See also

Edit this page