Compression

The single-byte compression_type in a Chunk Header selects the compression format for every Page in that Chunk.

Value	Name	Page data format
0	UNCOMPRESSED	Raw bytes
1	SNAPPY	Snappy raw block
2	GZIP	One gzip member
3	LZO	LZO block (C++ implementation, enabled at build time)
4	SDT	Reserved; no codec
5	PAA	Reserved; no codec
6	PLA	Reserved; no codec
7	LZ4	LZ4 raw block
8	ZSTD	Zstandard frame
9	LZMA2	LZMA2 stream in an XZ container

Each Page is compressed independently; compression state is not shared across Pages. The Page Header's uncompressed_size records the uncompressed payload length. compressed_size records the stored body length after compression and any encryption, including Page AEAD framing and tags.

The numeric values are stable wire identifiers rather than the ordinal positions of a compressor library. Values 3 through 6 are legacy identifiers from earlier TsFile versions. The Java implementation does not define or implement them and rejects these bytes while parsing a Chunk Header. The C++ implementation retains them as LZO, SDT, PAA, and PLA; only LZO has a codec (enabled at build time), while SDT, PAA, and PLA remain reserved without a codec. A reader that does not provide the codec cannot decode the affected Chunk.

Page-local Framing

Let a Page start at P and its Header occupy H bytes. Read exactly compressed_size bytes for its body:

[P + H, P + H + compressed_size)

This is the stored length, including any encryption framing. After decryption, UNCOMPRESSED already contains the encoded Page payload; other methods need decompression. Either path must produce exactly uncompressed_size bytes. With length-preserving encryption or no encryption, UNCOMPRESSED has equal stored and uncompressed lengths. Page AEAD adds framing, so that equality no longer holds.

Pages do not share compression history. A reader can seek directly to a selected Page or skip its body using the stored length, without decompressing earlier Pages.

Read Pipeline and Validation

The Page Header is read before the payload, so the reader knows both bounds before allocating or decompressing:

read Page Header
  -> read exactly compressed_size bytes
  -> authenticate/decrypt according to the file encryption header or legacy properties
  -> decompress according to compression_type
  -> require exactly uncompressed_size bytes
  -> decode the Page streams

Apply the length checks in Binary Representation before allocating buffers. An unavailable codec prevents decompression. A scanner can still use data_size to skip the Chunk if it parses the raw Header fields, even when the codec registry rejects the identifier; it must not treat that identifier as UNCOMPRESSED.