RIRRd
View on GitHub Storage that stays local
An embedded database for your mirrored objects, indexes and source state. Pick the sources and object classes your deployment needs.
Set up your first mirror
Open-source beta
Your routing registry,
Your routing registry,
built in Rust.
Mirror IRR data, look up routes and AS-sets, and build filters for BIRD or FRR. An Internet Routing Registry daemon with embedded storage and a console included.
Open-source project · Full IRRd parity is still in development.
Run it on your infrastructure
The pieces you need, together.
Choose your sources and resource budget. RIRRd runs as one daemon, with no PostgreSQL or Redis to manage.
A console included
Browse source health, explore registry objects and copy query commands. WHOIS, HTTP and GraphQL serve the same stored data.
Explore the live consoleFits your router workflow
Use bgpq4 to generate IPv4 and IPv6 IRR filters for BIRD or FRR. Inspect and test candidate policies before applying them.
Follow the router guideStart small. Choose what you mirror.
Installation, cache budgets, source configuration and upgrade guidance in one place.
Latest development work: mirrored ingestion and bounded queries across all 19 RPSL classes, query memory limits, and verified database compaction. These are separate from the historical comparison below. Read the development review and full-catalog storage measurements.
Measured, with context
A closer look at the footprint.
100,000 synthetic objects, on the same Linux VM. Compared with pinned official IRRd 4.5.4, with RPKI disabled in both profiles.
Snapshot import
MedianSeconds · lower is better · three runs each
Idle memory
Process PSSMiB · lower is better · summed process memory
Database storage
AllocatedMiB · lower is better · includes PostgreSQL WAL
Queries per second & response time
Eight clients · reused connections
| Workload | RIRRd queries/s | IRRd queries/s | RIRRd p95 | IRRd p95 |
|---|
Historical development measurements, not a capacity promise for the current beta. These workloads cover a shared subset of features; they do not establish full IRRd parity.
Dataset, memory accounting & limitations
These measurements use the exact revisions listed in the raw results. Both systems used the same synthetic fixture and recorded host conditions. Memory includes the reference application and its PostgreSQL/Redis services; filesystem cache outside processes is excluded.
Real-world datasets and sustained load need separate sizing. Authentication, policy, replication and release qualification remain in development. See the testing page for recorded verification and beta status for the remaining gaps.
Prefer to try it first?
Meet the community network.
The software lives here. Public nodes, hosting credits and operator submissions live at irrd.tools.
United Kingdom · UK 01
Hosted by HYEHOST LLC
An experimental public node serving IRR queries over IPv4 and IPv6. Explore the data before setting up a server of your own.
Visit the community network- Stored objects
- —
- Loaded sources
- —
Loading live node counters.
An independent project
Built with curiosity.
Shared with the community.
RIRRd is a passion project by Lostepic (AaranCloud). It began in a personal repository, with roughly three months of work before being shared publicly. It is now open source for anyone to use, learn from, modify and contribute to. Development and maintenance happen as time allows.
Project background & respect for other implementations
RIRRd is one of several Rust IRR implementations. The idea of writing an IRR daemon in Rust is not unique, and this project does not claim otherwise. It is not intended to take credit from, undermine or detract from other projects. We respect their work and welcome different approaches to the same problem.
RIRRd is an independent implementation. No code was copied from another Rust IRR project. Protocol compatibility work uses official IRRd as its reference; compatibility claims are limited to the documented tests and evidence.
Read the repository disclaimerHelp shape what comes next
Contributions are welcome.
Built with AI assistance, and open to human and AI-assisted contributions with review and tests. Report reproducible bugs, query differences or performance regressions. Keep credentials and private registry data out of public issues.