PeeringDB Plus

A globally distributed, read-only mirror of PeeringDB data. PeeringDB Plus incrementally syncs PeeringDB objects on a regular schedule (escalating to a periodic full re-fetch as a safety net), stores the result in SQLite with edge replication via LiteFS on Fly.io, and serves the same dataset through six coexisting API surfaces.

Live instance: https://peeringdb-plus.fly.dev

---

Why

The upstream PeeringDB API is the canonical source for peering coordination data, but it is single-region and rate-limited. PeeringDB Plus offers:

• Low-latency reads from the nearest Fly.io region (LiteFS replicates SQLite transactions to every replica).
• Multiple wire formats from a single dataset: drop-in PeeringDB API compatibility, OpenAPI REST, GraphQL, ConnectRPC/gRPC, MCP, and a Web UI all read from the same ent.Client.
• Built-in observability: the server sends OpenTelemetry traces (sampled per route), metrics, and structured logs.
• One Go binary: the server needs no cgo. LiteFS runs as a separate FUSE process.

API surfaces

All six surfaces are mounted on the same HTTP server and read from the same SQLite database.

Surface	Path	Description
Web UI	/ui/	templ + htmx search, detail pages, ASN comparison; content-negotiates terminal vs browser
GraphQL	/graphql	gqlgen via entgql; interactive playground on GET, queries on POST
REST	/rest/v1/	OpenAPI-compliant (entrest); spec at /rest/v1/openapi.json
PeeringDB Compat	/api/	Drop-in replacement for www.peeringdb.com/api -- same paths, same envelope
ConnectRPC / gRPC	/peeringdb.v1.*/	Get / List / Stream RPCs for all 13 entity types; reflection + health checks enabled
MCP	/mcp	Read-only tools, resources, and prompts for network research agents

GET / returns a JSON service-discovery document to API clients. It redirects browsers to the Web UI and sends ANSI-colored help text to terminal clients such as curl. Each response has an HTTP Link header that points to the agent documents.

Most responses carry a weak ETag that changes when the database changes. All nodes send the same ETag for the same database version. A client that sends If-None-Match gets 304 Not Modified while the data stays the same.

See docs/API.md for each API, with filter semantics, ordering, divergences, and the response memory budget for /api lists.

Quick start

Run a mirror

For an operator-managed mirror, follow Getting Started. It covers the standalone container, source build, first sync, and verification. These operator workflows contact upstream PeeringDB.

Develop locally

Follow Development: Local setup for the locked toolchain and a fixture-backed server. Do not point development servers or load tests at upstream PeeringDB.

Verify it's working

curl -s http://localhost:8080/healthz                    # liveness
curl -sI http://localhost:8080/readyz                    # readiness (200 once first sync completes)
curl -s http://localhost:8080/api/net/1 | head -c 500    # PeeringDB-compatible API
curl -sO http://localhost:8080/skills/peeringdb-plus.zip # Origin-aware Agent Skill

Open http://localhost:8080/ui/ (Web UI) and http://localhost:8080/graphql (GraphQL playground) in a browser.

For operator setup and first-sync details, see docs/GETTING-STARTED.md.

Usage examples

PeeringDB-compatible API

# List networks
curl https://peeringdb-plus.fly.dev/api/net

# Fetch a specific network by PeeringDB numeric ID
curl https://peeringdb-plus.fly.dev/api/net/42

# Search with query parameters. Lists accept limit, skip, fields, since, and q.
# Lists and single-object requests accept depth. See the API reference for limits.
curl 'https://peeringdb-plus.fly.dev/api/net?q=cloudflare&limit=5'

ConnectRPC / gRPC

ConnectRPC, gRPC, and gRPC-Web all share the same endpoints. Reflection (grpcreflect.NewHandlerV1 / V1Alpha) and the standard gRPC health check service are enabled, so grpcurl and gRPC-aware load balancers work out-of-the-box.

# List all services
grpcurl -plaintext localhost:8080 list

# Get a single network by ID
buf curl --protocol grpc --http2-prior-knowledge \
  http://localhost:8080/peeringdb.v1.NetworkService/GetNetwork \
  -d '{"id": 42}'

# Stream the networks that match a filter (server streaming, no pagination)
buf curl --protocol grpc --http2-prior-knowledge \
  http://localhost:8080/peeringdb.v1.NetworkService/StreamNetworks \
  -d '{"asn": 15169}'

Streams accept the List* filters except id, and also since_id and updated_since. If you do not set since_id or updated_since, the pdbplus-total-count response header gives the number of matching rows. The server also sends this value as grpc-total-count, which is a deprecated name. By default, the server stops a stream after 60 seconds (PDBPLUS_STREAM_TIMEOUT). A client can cancel a stream at any time. See docs/API.md § Streaming semantics.

GraphQL

curl -X POST http://localhost:8080/graphql \
  -H 'Content-Type: application/json' \
  -d '{"query":"{ networks(first: 3) { edges { node { id name asn } } } }"}'

MCP and Agent Skill

The Streamable HTTP MCP endpoint is available at /mcp. It supports MCP 2026-07-28 sessionless discovery and older clients that use the initialize handshake. It provides bounded directory search, detail, comparison, IP lookup, and sync freshness tools plus reusable resources and prompts.

Discover the agent interfaces:

curl -fsS http://localhost:8080/llms.txt
curl -fsS http://localhost:8080/.well-known/mcp/server-card.json
curl -fsS http://localhost:8080/.well-known/agent-skills/index.json

Download the skill through the hostname agents should connect back to:

curl -fLO http://localhost:8080/skills/peeringdb-plus.zip
curl -fsS http://localhost:8080/skills/peeringdb-plus/SKILL.md

The generated server card, skill index, llms.txt, and ZIP metadata use the request origin. Set PDBPLUS_PUBLIC_URL only when a reverse proxy does not preserve the requested Host.

Configuration

PeeringDB Plus is configured exclusively via environment variables, validated at startup with fail-fast diagnostics. Operationally-relevant defaults:

Variable	Default	Purpose
PDBPLUS_LISTEN_ADDR	:8080	HTTP listen address
PDBPLUS_DB_PATH	./peeringdb-plus.db	SQLite database file path
PDBPLUS_PEERINGDB_URL	https://www.peeringdb.com	Upstream PeeringDB API
PDBPLUS_PEERINGDB_API_KEY	(unset)	Optional -- raises rate limit and shortens default sync interval
PDBPLUS_SYNC_MODE	incremental	incremental (delta) or full (re-fetch); full is the operator escape-hatch
PDBPLUS_SYNC_INTERVAL	1h (15m if API key set)	Time between sync cycles
PDBPLUS_FULL_SYNC_INTERVAL	24h	Maximum time between full re-fetches in incremental mode. A full cycle repairs data that incremental sync cannot see. 0 disables it
PDBPLUS_RESPONSE_MEMORY_LIMIT	128MB	Memory budget for /api responses. A list whose estimated size is larger gets HTTP 413. The same budget also limits the total estimate of all /api responses in progress. A request that would go over that total gets HTTP 503. The value needs a unit suffix: KB, MB, GB, or TB (base 1024).
PDBPLUS_PUBLIC_TIER	public	Anonymous-caller tier; set users only for private deployments
PDBPLUS_PUBLIC_URL	(unset)	Optional public-origin override for generated Agent Skill metadata

The full list (sync, observability, LiteFS, Fly.io, CSP, map tiles, and privacy tiers) is in docs/CONFIGURATION.md. The server also reads the standard OTEL_* environment variables through OpenTelemetry autoexport.

Documentation

Doc	Topic
docs/GETTING-STARTED.md	Operator setup, first sync, and verification
docs/ARCHITECTURE.md	Component diagram, data flow, code-generation pipeline, middleware chain, privacy layer, sampling matrix
docs/API.md	All six API surfaces, filter semantics, ordering, cross-entity traversal, divergences
docs/CONFIGURATION.md	Full environment-variable list with validation rules
docs/DEVELOPMENT.md	Local dev workflow, code generation, conventions, sibling-file pattern
docs/TESTING.md	Test layout, fixtures, parity harness, live tests against beta.peeringdb.com
docs/DEPLOYMENT.md	Fly.io rollout, asymmetric fleet topology, LiteFS operations
docs/meta-generated-behavior.md	How upstream meta.generated behaves, and why the sync does not use it as a cursor
CONTRIBUTING.md	How to propose changes, run the verification suite, and submit a PR
CHANGELOG.md	Release notes

Technology

• Language: Go 1.27.3
• ORM / codegen: entgo. cmd/pdb-schema-generate generates the ent schemas from schema/peeringdb.json. entgql and entrest generate the GraphQL and REST layers. The protobuf definitions in proto/peeringdb/v1/ are hand-maintained, and buf generate makes the Go code from them. All six APIs read data through the same ent client.
• Database: modernc.org/sqlite (pure Go, no cgo) plus LiteFS for edge replication
• RPC: ConnectRPC -- gRPC, gRPC-Web, and the Connect protocol on the same handlers
• GraphQL: gqlgen wired through entgql
• Web UI: templ + htmx + Tailwind CSS -- zero JavaScript build toolchain
• Observability: OpenTelemetry with the stdlib log/slog bridge; per-route trace sampling; per-VM runtime metrics
• Platform: Fly.io with an asymmetric primary / replica process-group fleet
• Container base: Chainguard minimal images (cgr.dev/chainguard/go, cgr.dev/chainguard/static, and cgr.dev/chainguard/glibc-dynamic for the LiteFS image)

Development

Development owns setup, build commands, generation, and change procedures. Testing covers fixtures and test selection. Contributing lists the pre-PR checks.

Contributing

PeeringDB Plus is open source. Bug reports, feature suggestions, and pull requests are welcome -- please read CONTRIBUTING.md before opening a PR.

License

BSD 3-Clause. See LICENSE.