// the find
vectordotdev/vector
A high-performance observability data pipeline.
Vector is a Rust-based observability pipeline that collects, transforms, and routes logs, metrics, and traces to any backend. It runs as either an edge agent or a central aggregator, replacing the patchwork of Filebeat + Logstash + whatever else. Maintained by Datadog but vendor-neutral by design.
The performance numbers are real and reproducible — the test harness is public and the benchmarks are honest enough to show FluentBit beating Vector on regex parsing. VRL (Vector Remap Language) is a genuinely good idea: a purpose-built, sandboxed scripting language for transforms that catches errors at compile time rather than silently dropping events at 3am. Disk-buffered delivery with at-least-once guarantees is built in, not bolted on — file rotation edge cases that trip up Filebeat are handled correctly. The unified logs+metrics model means one config, one binary, one thing to operate.
Traces support is still marked 'coming soon' in the README despite the project being years old — if you need distributed tracing in the pipeline, you're not there yet. The configuration surface is enormous; a non-trivial topology with multiple sources, transforms, and sinks becomes a YAML maintenance burden fast, and the error messages when you misconfigure a component can be cryptic. Being Datadog-maintained cuts both ways: roadmap decisions are influenced by what Datadog needs internally, and there's a non-zero risk of the open-source version being steered toward Datadog's commercial interests over time. Compile times are brutal if you're building from source — full debug build on a modern machine is a coffee break.