Compile only what you need

Big ecosystem, small binary

faucet-stream is a Cargo workspace of around 90 crates — a core plus a source or sink per system. That's the menu, not the meal. Every connector is an optional feature, so you compile in only the ones your pipeline uses:

cargo install --path cli \
  --no-default-features \
  --features "source-rest,sink-jsonl,transforms"

That binary contains the REST source, the JSONL sink, and the transform engine — and nothing else. No Postgres driver, no cloud SDK, no Kafka client. Smaller binary, faster build, smaller attack surface, fewer transitive dependencies to audit.

Why this shape matters

• Cost is proportional to use. A cloud SDK or a database driver is only pulled in when you enable the connector that needs it — a config that never touches BigQuery pays nothing for the BigQuery client.
• Isolation is enforced. CI builds each connector feature on its own, so an accidental cross-dependency between connectors is caught, not shipped.
• Supply chain stays lean. Fewer compiled-in crates means fewer things in your dependency tree to review and keep patched.

Good for third-party authors too

The same design keeps the ecosystem open: a connector depends only on faucet-core, which re-exports the shared pieces (async-trait, serde_json, schemars) so authors don't chase a pile of dependencies. Anyone can publish a faucet-source-* / faucet-sink-* crate that slots into the same feature model.

Takeaway

You get a broad ecosystem without a bloated binary. Turn on the connectors you use, leave the rest on the shelf, and ship something small.