Open-source project
vectordotdev/vector avatar
vectordotdev/vector

Vector's Own Benchmark Table, and the Two Rows It Loses

A high-performance observability data pipeline.

22,870 stars2,387 forksRustMPL-2.0

At a glance

What is it?
Vector is a Rust agent and aggregator that collects, transforms and routes logs and metrics from one config. It leads most of its own throughput table, trails FluentBit on regex parsing, and still lists traces as coming soon.
Who is it for?
Adopt Vector when your bottleneck is moving bytes between hosts: on the project's own table it leads file to TCP by a wide margin and beats Logstash on every row shown. Do not adopt it for traces, which the principles list still marks as coming soon, nor for regex heavy parsing, where FluentBit's 20.5mib/s beats Vector's 13.2mib/s.
Can I use it commercially?
Yes, with conditions. MPL-2.0 is a weak copyleft licence: you can use it inside commercial and closed-source software, but if you distribute changes to its own files, you must publish those changes under the same licence.
Is it still maintained?
Yes. The repository last received commits 2 days ago.
What is it written in?
Mainly Rust, according to GitHub's language statistics.

Answers come from the project's GitHub data, last synced on November 13, 2026, and from our analysis. They are not legal advice.

From daily snapshots

Star history

  • vectordotdev/vector22,870 ★ · +321
Star history
Repository2026-09-132026-11-13
vectordotdev/vector22.5k22.9k

321 Stars in the last 61 days, measured from the daily GitHub snapshot. Compare with other projects

Editorial analysis

One binary in two roles, and two data types it does not carry

Vector collects, transforms and routes logs and metrics, and the same executable plays either role: the docs describe deployment as an agent or an aggregator, which is the difference between sitting on one host and sitting between many. That makes it a candidate to replace two pieces of your stack rather than one, and it is also what makes its benchmark table awkward to read, because a single Vector column is lined up against agents such as Filebeat and FluentBit and against heavier forwarders such as Logstash and Splunk HF in the same row. The claim that Vector is a complete platform carries a footnote in the principles list: metrics are marked beta and traces are marked coming soon. A pipeline designed around traces today has nowhere to put them, and the metrics half of the pitch ships with a beta label in the release you would install. Treat traces as a roadmap item.

The throughput table leads four rows and loses two

Read the performance table row by row and the up to 10x faster than every alternative claim holds in some rows and not others. TCP to Blackhole is 86mib/s for Vector against 64.4mib/s for FluentBit and 40.6mib/s for Logstash. File to TCP is 76.7mib/s against 40.1mib/s for Splunk UF and 3.1mib/s for Logstash, the widest gap on the page. TCP to HTTP is 26.7mib/s against 19.6mib/s for FluentBit and under 1mib/s for FluentD. Then two rows go the other way. Regex Parsing is 13.2mib/s for Vector against 20.5mib/s for FluentBit, and the project flags that row as the exception to its protocol baseline. TCP to TCP is 69.9mib/s against 70.4mib/s for Splunk UF, a loss by half a megabit per second. The n/a cells say as much as the numbers do: they mark tests with no result for the competing tool. If your bottleneck is parsing rather than moving bytes, the regex row is the one to plan around.

Blank correctness cells are missing results, not losses

The second table covers correctness rather than speed, and its own caption admits the cases are not exhaustive. What is visible is still telling. Disk Buffer Persistence is marked for Vector and Filebeat, carries a warning for Logstash, and leaves FluentBit and FluentD empty. File Rotate (copytruncate) has a mark for Vector, Splunk UF and Splunk HF and nothing at all for the rest. Process (SIGHUP) marks Vector, Splunk UF and Splunk HF, warns for Logstash, and leaves Filebeat, FluentBit and FluentD blank. The caption presents these as fundamental differences in quality and attention to detail, but the table does not say what a blank cell or a warning mark means. The single case worth acting on is the copytruncate row, because that is the behaviour of a logging library that copies and then truncates a file under a live handle. Nobody's rotation pattern is in this table, so it does not answer the question you actually have.

MPL-2.0, publish = false, and maintainers on Datadog's payroll

The licence is MPL-2.0 and the Cargo manifest sets publish = false, so the crate is not distributed through a package registry. Two things follow from the repository layout for anyone doing a compliance pass. The third party licence surface is tracked in the tree rather than left implicit: there is a LICENSE-3rdparty.csv, a licenses/ directory, license-tool.toml, deny.toml for cargo-deny, and NOTICE shipped in the deb assets. And Vector is maintained by Datadog's Community Open Source Engineering team, which the project states plainly rather than burying. Hold that next to the listed use case of transitioning vendors without disrupting workflows: the people merging the pull requests are employed by one of the vendors your pipeline can route to. That is a fact for an architecture review, and the copyleft question belongs to whoever handles licence compliance rather than to a README.

Two release lines move at once: v0.58.0 and vdev-v0.3.25

Three recent tags tell the story: v0.61.0 on 2026-10-29, v0.60.0 on 2026-10-19, and vdev-v0.3.30 on 2026-10-09, while the manifest sits at 0.62.0-dev. Two lines ship from one repository, so version numbers from this project are not one sortable sequence. Rust is pinned by rust-version = "1.95" with edition 2024, and the release profile is tuned hard, with lto = "fat", codegen-units = 1 and debug symbols off, so a source build is slow by design and needs a pinned toolchain file. Pin deliberately on the artifact side too: the download link ends in /releases/latest/download/, which resolves to whatever is newest, so a build script that fetches that URL silently replaces your binary without touching your repository.

No install line in the README, and the paths the deb wants to own

The README's opening points to a quickstart and install guide rather than giving a command, and the project names four places: a quickstart guide at vector.dev/docs/setup/quickstart/, a docs tree at vector.dev/docs, a per-release download at https://vector.dev/releases/latest/download/, and container images published under the repository's GitHub package. Because publish = false, cargo install is not the path being advertised, and the deb packaging metadata shows the shape of a system install: the binary into /usr/bin/, defaults at /etc/default/vector with mode 600, and example configs under /etc/vector/examples/. Check those paths before you install. A host already running another collector may already use those paths, and /etc/default/vector is a file you will be editing by hand with mode 600 on it. The secret-backend-example binary sits behind a cargo required feature, so a default build leaves it out.

The numbers come from a separate harness, since cargo bench is disabled

The tree behind a release is large: benches/, regression/, rfcs/, tilt/ with a Tiltfile, proto/, distribution/, and AGENTS.md, CLAUDE.md and AI_POLICY.md sitting next to SECURITY.md, VERSIONING.md and RELEASES.md. The Makefile shows what it costs to work on it. Tests run through cargo nextest run, and setting COVERAGE=true swaps in cargo llvm-cov nextest --no-report, after which make coverage-report emits the lcov file. Scope, extra bench flags and the criterion output directory are overridable from the environment, and the target triple is detected from the OS, with DNSTAP_BENCHES unset on Windows. For the throughput figures themselves, read autobenches = false and the comment beside it: the benchmarks are not runnable on their own. Those numbers come from the separate vector-test-harness repository, so reproducing them means running that harness against your own hardware, not cargo bench.

Editorial conclusion

Adopt Vector when your bottleneck is moving bytes between hosts: on the project's own table it leads file to TCP by a wide margin and beats Logstash on every row shown. Do not adopt it for traces, which the principles list still marks as coming soon, nor for regex heavy parsing, where FluentBit's 20.5mib/s beats Vector's 13.2mib/s. Before installing, check two things: that /usr/bin/vector and /etc/default/vector do not already belong to another agent on that host, and that your build does not fetch https://vector.dev/releases/latest/download/ on every run, because that URL resolves to whatever shipped most recently. Then pin a version, write the pipeline config, and watch what the disk buffer holds during your first sink outage.

Frequently asked questions

How do I install the Vector observability pipeline?

The README's opening section gives no install command. It points at the quickstart guide at vector.dev/docs/setup/quickstart/, a per-release download at https://vector.dev/releases/latest/download/, and container images in the repository's GitHub package. The manifest sets publish = false, so it is not installed from a package registry.

Does Vector handle traces yet?

No. The principles list marks logs as supported, metrics as beta and traces as coming soon. A design that depends on trace routing today has to route that data somewhere else.

Is Vector faster than FluentBit and Logstash?

On the project's own table, Vector leads on TCP to Blackhole, file to TCP and TCP to HTTP, and beats Logstash on all five rows shown. It trails FluentBit on Regex Parsing, 13.2mib/s against 20.5mib/s, and trails Splunk UF on TCP to TCP, 69.9mib/s against 70.4mib/s.

What licence is Vector under and who maintains it?

The licence is MPL-2.0, and the project states that Vector is maintained by Datadog's Community Open Source Engineering team. Third party licences are tracked in LICENSE-3rdparty.csv, licenses/, license-tool.toml and deny.toml.

What does building Vector from source require?

The manifest pins rust-version = "1.95" with edition 2024, and the release profile uses lto = "fat" with codegen-units = 1 and debug symbols disabled. Tests run through cargo nextest run, with cargo llvm-cov substituted when COVERAGE=true.

Official sources

  1. License: MPL-2.0
  2. Project website
  3. README
  4. Releases
  5. vectordotdev/vector on GitHub
Add this badge to your README

If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.

Add this badge to your README

markdown
[![Hysen Labs](https://hysenlabs.com/badge/vectordotdev-vector.svg)](https://hysenlabs.com/projects/vectordotdev-vector)

Star history chart for your README

markdown
[![Star history](https://hysenlabs.com/chart/vectordotdev-vector.svg)](https://hysenlabs.com/projects/vectordotdev-vector)

Was this useful?