BGP Route Oscillation and Route Leak Analyzer for Azure Virtual WAN

Detects BGP route oscillation and valley-free route leak violations in Azure Virtual WAN multi-hub topologies, then evaluates whether proposed solutions resolve each issue.

Domain Knowledge

Oscillation Detection (Preference Cycle)

A BGP oscillation occurs when ASes form a mutual routing preference cycle: e.g., AS A prefers routes via AS B, while AS B prefers routes via AS A. The preference graph is a directed functional graph (out-degree ≤ 1 per node). A cycle of any length causes persistent non-convergence. Detection uses DFS-based traversal with path tracking.

Output: oscillation_cycle and affected_ases are deterministically ordered (rotated so the smallest ASN appears first).

Route Leak Detection (Gao-Rexford Valley-Free Model)

Under the Gao-Rexford valley-free routing principle (RFC 7908):

• Customer-learned routes → may export to anyone (customers, peers, providers)
• Peer-learned routes → may export to customers ONLY
• Provider-learned routes → may export to customers ONLY

A route leak = exporting peer-learned or provider-learned routes to a peer or provider.

Solution Evaluation Rules

Fixes OSCILLATION:

1. Explicitly stop one cycle participant from preferring the other (stop preferring routes via)
2. Route preference hierarchy that deprioritizes peer routes (breaks cycle edge)
3. Export policy filtering peer-learned routes between cycle participants
4. Data-plane override: UDR or Routing Intent

Fixes ROUTE LEAK:

1. Export policy blocking provider-learned routes to peers (source-side)
2. BGP community no-export on provider routes (source-side)
3. Ingress filtering rejecting routes with provider AS in path (receiver-side)
4. RPKI origin validation (receiver-side)
5. Data-plane override: UDR or Routing Intent

Fixes BOTH: UDR and Routing Intent (forwarding layer overrides)

Fixes NEITHER:

• Timer tuning (keepalive/holdtime), route dampening, ECMP, max-prefix limits
• Waiting for convergence, restarting BGP, prefix-list filtering
• Generic AS-PATH/MED heuristics, generic route maps without specific config
• Disabling BGP sessions / shutting down peering / removing gateways (connectivity destruction — not acceptable)

KEY DISTINCTION: Filtering provider-learned routes fixes LEAKS but NOT oscillation. Filtering peer-learned routes fixes OSCILLATION but NOT leaks.

Usage

import sys
sys.path.insert(0, '/app/environment/skills/evo-azure-bgp-oscillation-route-leak/scripts')
from utils import generate_report

report = generate_report(data_dir='/app/data/', output_dir='/app/output/')

Scripts

• scripts/utils.py — All utility functions for data loading, detection, evaluation, and report generation