Skip to content

Decide if this project fits

timeseries-table-format manages local time-series tables built from immutable Parquet segments. It tracks coverage, rejects overlapping appends, and queries committed segments with DataFusion SQL.

A good fit

Use it when:

  • New time-series Parquet files arrive over time.
  • You want one managed table root instead of custom file-discovery code.
  • Each complete entity identity should have at most one row per index interval.
  • You want SQL results as pyarrow.Table or pyarrow.RecordBatchReader objects.
  • You need occasional backfills or corrections to selected values by complete row key.

Choose another tool when

  • You only need ad hoc queries over a few files. Query Parquet directly with a tool such as DuckDB or Polars.
  • You need frequent low-latency point updates or a central database server. Use a database; keyed updates here rewrite affected Parquet segments.
  • You need object storage, small-file compaction, column dropping/renaming, automatic schema merging, or merge operations. Use a lakehouse format designed for those workflows.

The current release supports local filesystems, append ingestion, and explicit keyed updates. If that matches your workload, continue with Installation.