Unified observability built for high-frequency capture
Marc-Antoine Desroches · madesroches@gmail.commicromegas.info
Capture enough detail to fix issues without reproducing them.
Micromegas refuses to choose.
One pipeline. Logs, metrics, and traces share the same path.
One pipeline, not three.
Three observability stacks is three things to learn, three things to operate, three things to pay for.
Every event shares a schema model: process_id, thread_id, time, session, plus signal-specific fields.
process_id
thread_id
time
session
Questions you can ask in one query:
Fragmented: hours hunting through three tools.Unified: one query.
900 billion events / 90 days / ~$1,750 a month
At this volume, traditional vendor SaaS pricing is orders of magnitude higher.
Micromegas was built on the opposite assumption — events should be cheap.
Same goals. Very different cost model. Everything else follows from that.
Apache Arrow — columnar, zero-copy, the lingua franca of analytics
Apache Parquet — durable columnar storage; compresses well, scans fast
DataFusion — Rust-native SQL engine, embeddable, fast
FlightSQL — gRPC-based wire protocol, language-agnostic clients
We didn't reinvent the analytics stack. We assembled it.
jsonb_path_query
jsonb_array_elements
pip install micromegas
micromegas-query "SELECT ..."
df = client.query("SELECT * FROM log_entries WHERE ...")
Your observability data joins your data warehouse, on demand.
SQL cells feed tables, charts, logs, swimlanes, flame graphs — all composable in one page.
One page, one URL — share an investigation like a dashboard.
micromegas.info
github.com/madesroches/micromegas · madesroches@gmail.com
Open source. Apache 2.0. Self-hosted on your cloud.