Network & Infrastructure / BGP Policy & Internet Edge
BGP update backlog and slow convergence without session loss
After an upstream event, the session stays established while the route table converges much more slowly than expected. A large update backlog can hide behind a healthy peer-state indicator.
Environment and inputs: BGP speakers and route reflectors; policy syntax, RPKI integration and address-family behavior depend on the implementation. Diagnostic/design example; not tested on production equipment.
Collect the evidence
Record input/output queues, update processing rates, memory, CPU, policy complexity and FIB programming delay. Compare timestamps for route reception, selection and hardware installation.
Explain the result
Convergence includes control processing and data-plane programming. An empty TCP retransmission counter does not prove route-policy evaluation or line-card installation is keeping pace.
Plan the fix
Reduce unnecessary policy work or route volume based on measured bottlenecks, and review supported platform capacity. Avoid shorter keepalives as a remedy for a processing backlog.
Check the result and rollback
Replay representative churn in a lab and measure end-to-end forwarding recovery. Revert if optimizations alter routing intent, and retain queue and installation evidence for platform capacity decisions.