Timelake

Credits

Timelake is built on other open-source projects and borrows ideas from many more. We take ideas, never code. Thank you to everyone behind them.

Built on

ProjectWhat it does in Timelake
Apache IcebergThe table format
Apache ParquetThe data files
Apache ArrowData in memory
Apache DataFusionThe query engine
iceberg-rust, object_storeIceberg types, and every read and write to storage
DuckDBReading tables with plain SQL

Ideas we borrowed

IdeaFrom
Size-tiered compactionApache Cassandra
Tables readable with no catalog serviceIceberg's Hadoop catalog
Staging a backfill on a branch, then publishing itIceberg branches
Rollup tables of per-bucket aggregatesApache Druid
Answering a query from a matching rollupApache Calcite
A time window's first, smallest, largest and last points (m4)Apache IoTDB
Object-store reads split into parallel requests of a few MiBDurner, Leis, and Neumann, VLDB 2023

Benchmarks, fonts, and components

WhatFrom
Benchmark queriesTSBS, ClickBench, TPC-H via DuckDB
FontsGeist, by Vercel and basement.studio
UI componentsshadcn/ui

Licenses are in NOTICE. Our notes on these projects, and on the ones we're studying, are in PRIOR_ART.md.