I‘m using JSONL for long term storage, workers written in C for fast daily summaries and DuckDB for time series analysis.

It‘s a breath, because my data stays readable for the next decades and I can reprocess it whenever new tools are available