Adding BGP Prefix Length Statistics to BIRD Exporter

When managing a BGP full routing table containing approximately 220,000 IPv6 routes, network engineers need visibility into prefix length distribution. I sought to understand how many /48, /32, and /44 prefixes exist within my routing table, but discovered that the popular bird_exporter for Prometheus didn't provide this granularity.

The Challenge

The bird_exporter tool offered excellent metrics for BGP session health and route counts, yet lacked the ability to break down routes by prefix length. A direct approach—querying all routes and parsing their lengths—created a critical problem: Unix socket buffer overflow when processing 220,000 routes generating approximately 880,000 lines of output.

The Solution

Rather than parsing complete route details, I leveraged BIRD v2's efficient count functionality. Specific queries like:

show route table master6 where net ~ [::/0{48,48}] primary count

returned only essential data—the count of matching prefixes.

Key implementation details included:

• Querying specific prefix lengths using BIRD's filter syntax
• Using the primary filter to count unique prefixes without duplication from multiple peers
• Parsing BIRD v2's response format: extracting the filtered count from responses like "1007-197991 of 220785 routes"

Results

The implementation exports Prometheus metrics tracking prefix length distribution across routing tables, with separate metrics for IPv4 and IPv6. The data revealed diverse prefix allocations, with /48 prefixes comprising 99,344 routes and /32 prefixes totaling 25,763 routes in the master6 table.

Technical Insights

Three lessons learned: leverage built-in count capabilities rather than parsing, respect system buffer limitations, and validate solutions against real-world datasets. The complete implementation is available on GitHub.