Storage Format

TSEdge stores data in ordinary files under a database directory. The layout is intentionally simple so it can be inspected, tested and explained without a separate server process.

database_dir/
  manifest.txt
  wal.log
  series/
    air.temperature/
      metadata.txt
      segment_000001.tse
      segment_000002.tse

Files

• manifest.txt marks the directory as a TSEdge database and stores basic database metadata.
• wal.log stores accepted points that have not yet been flushed into segment blocks.
• metadata.txt stores per-series metadata.
• segment_*.tse files store append-only compressed blocks.

Segments and blocks

Each segment file is an append-only sequence of compressed blocks. A block is the unit of compression and query skipping.

Blocks contain enough metadata to answer a simple question quickly: can this block contain points in the requested time range? If not, the reader can skip the compressed payload and move to the next block.

The payload contains two compressed streams:

• timestamp stream
• value stream

Multi-byte integer fields are encoded explicitly in little-endian order. This avoids silently depending on the host CPU byte order.

Block header

Field	Type	Purpose
magic	u32	block signature
version	u32	block format version
point_count	u32	number of points
min_timestamp	i64	block time lower bound
max_timestamp	i64	block time upper bound
timestamp_size	u32	compressed timestamp stream size
value_size	u32	compressed value stream size
payload_size	u32	total compressed payload size
min_value	f64	minimum value
max_value	f64	maximum value
sum_value	f64	sum of values

The current implementation also stores a compression type and a reserved field in the binary block header. Readers validate these fields before decoding so malformed files fail cleanly.

Block-level min_value, max_value, sum_value and point_count allow fully covered aggregate queries to avoid decompressing payloads.