SignalDB

Observability for your homelab — one binary, one data directory.

SignalDB stores traces, logs, metrics, and continuous profiles. It ingests OpenTelemetry natively, answers Grafana's Tempo, Loki, Prometheus, and Pyroscope APIs, and ships a built-in Explore UI — so you get a full observability stack from a single process, without running four separate systems to get four signals.

It is designed to be stupidly easy to run small and able to scale out when you outgrow small: the same binary that runs on a Raspberry Pi or a NAS with SQLite and local disk splits into independent services backed by PostgreSQL and S3-compatible object storage.

Try it: signaldb-demo.58lab.org runs SignalDB fed by the OpenTelemetry Demo shop. Sign in with [email protected] / demo (read-only).

Why homelabbers run it

• One process, no dependencies. SQLite catalog, local-disk storage, sensible defaults. No JVM, no Zookeeper, no sidecar zoo.
• Bounded disk usage. A built-in compactor enforces 30-day retention by default and compacts Parquet files in the background — it won't quietly eat your NAS.
• A UI out of the box. Logs, traces, and metrics are explorable at /ui/ without installing Grafana. When you want Grafana, the Tempo/Loki/Prometheus-compatible APIs and a native plugin are there.
• Runs on your hardware. Multi-arch images (amd64/arm64) built on Alpine; cheap object storage (MinIO, S3) is optional, not required.
• Open standards. OTLP ingest (gRPC and HTTP), Prometheus remote_write, and Apache Parquet/Iceberg on disk — your data stays in open formats you can query with anything that speaks SQL over Parquet.

Quick start

Zero config works: start SignalDB with no tenants configured and it provisions a default tenant (dataset default) on first boot, printing a fresh ingest API key once in the startup logs (docker logs signaldb). Point your SDK at it with x-tenant-id=default and that key.

To choose your own tenant and key instead, create a minimal signaldb.toml defining a tenant and an ingest API key (authentication is always on — even at home, telemetry endpoints shouldn't be open writes):

[database]
dsn = "sqlite:///data/signaldb.db"

[discovery]
dsn = "sqlite:///data/signaldb.db"

[schema]
catalog_type = "sql"
catalog_uri = "sqlite:///data/catalog.db"

[storage]
dsn = "file:///data/storage"

[[auth.tenants]]
id = "homelab"
slug = "homelab"
name = "Homelab"
default_dataset = "default"

[[auth.tenants.api_keys]]
key = "sk-homelab-change-me"
name = "Ingest key"

[[auth.tenants.datasets]]
id = "default"
slug = "default"
is_default = true

Run the monolithic image (published for amd64 and arm64):

docker run -d --name signaldb \
  -p 3000:3000 -p 4317:4317 -p 4318:4318 \
  -v signaldb-data:/data \
  -v "$PWD/signaldb.toml:/data/signaldb.toml:ro" \
  ghcr.io/cedricziel/signaldb:main --config /data/signaldb.toml

Create a user so you can sign in to the UI:

docker exec -e SIGNALDB_USER_PASSWORD=changeme signaldb \
  signaldb-cli --config /data/signaldb.toml \
  user create [email protected] --tenant homelab

Point any OpenTelemetry SDK or Collector at it:

OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:4317
OTEL_EXPORTER_OTLP_HEADERS="authorization=Bearer sk-homelab-change-me,x-tenant-id=homelab"

Then open http://localhost:3000/ui/ and sign in. Full guides: sending OTLP data · Explore UI · Grafana datasource.

What works today

SignalDB is under active development.

Signal	Ingest	Query	Maturity
Traces	OTLP gRPC/HTTP	Tempo-compatible API, TraceQL search, Explore UI waterfall	Most mature surface
Logs	OTLP gRPC/HTTP	LogQL (broad function coverage), live tail in the UI	Solid; some backend query optimizations still landing
Metrics	OTLP, Prometheus remote_write	PromQL (growing function coverage), SQL	Usable; full PromQL parity is tracked in #336
Profiles	OTLP (v1development)	Pyroscope-compatible API, Grafana flame graphs	Experimental

Nightly micro-benchmark trends (OTLP decode, WAL, Iceberg append, querier read paths, compaction) are published at cedricziel.github.io/signaldb/benchmarks — see Benchmarking for what they do and do not measure. There are no published end-to-end load numbers yet. Remaining gaps are tracked in the issue tracker.

When you outgrow one box

The monolith is one deployment mode, not a ceiling. The same binary runs each service on its own — signaldb acceptor, signaldb router, signaldb writer, signaldb querier, signaldb compactor — discovering each other through a shared catalog and communicating over Apache Arrow Flight:

• PostgreSQL replaces SQLite for the catalog and service discovery
• S3-compatible object storage (MinIO, AWS S3) replaces local disk
• Multi-tenancy with per-tenant API keys, datasets, quotas, and rate limits — isolation is built in, not bolted on
• WAL-based durability on the ingest path, so acknowledged data survives crashes

Under the hood SignalDB is built on the FDAP stack — Apache Arrow Flight, DataFusion, Arrow, and Parquet — with Apache Iceberg as the table format. See the architecture overview.

Documentation

Docs live at https://cedricziel.github.io/signaldb/ (also browsable in docs/):

• Using SignalDB — sending data, querying, Grafana, the UI
• Operating SignalDB — deployment, WAL persistence, retention and compaction
• Architecture — how it works inside
• Contributing — coding standards

Development

cargo build                # build everything
cargo test                 # run the test suite
./scripts/run-dev.sh       # monolithic mode with local file storage

Prerequisites: stable Rust (edition 2024) and the Protocol Buffers compiler. See CLAUDE.md and docs/contributing/rust.md for conventions.

License

AGPL-3.0