Network automation design and operations
What the domain tests
Network automation design and operations covers designing automation that operates safely at scale: source of truth, intent, workflow, observability, and operational handoff. It tests architecture decisions and failure behavior, not just API calls.
Core rules
- Source of truth: NetBox, NSO, or a CMDB holds intended state; devices are consumers. Automation compares intent with actual state and reconciles drift.
- Layers: inventory, rendering/templating, transport (NETCONF/RESTCONF/gNMI/CLI), orchestration, verification, and telemetry. Keep them separable so a transport change does not rewrite business logic.
- Observability: model-driven telemetry (gNMI dial-out or dial-in, NETCONF subscriptions) streams state; syslog and SNMP traps are event sources. Define thresholds and alert ownership.
- Safety: read-only mode first, blast-radius limits, canary devices, maintenance windows, approval gates, idempotent remediation, and a tested rollback path.
- Operations: runbooks, RBAC, audit logs, secret rotation, and error budgets. Automation must fail loudly and leave devices in a known state.
- Design choice: controller-based orchestration (DNA Center, NSO, vManage) for scale and policy; device-level APIs for targeted tasks and brownfield gaps.
Worked example
Design a configuration-compliance service. Nightly, read intended interface config from the source of truth, render a vendor template, fetch actual state over RESTCONF, and compare normalized YANG paths. Write drift to a report and open a ticket; do not remediate automatically in the first release. After two weeks of clean reports, enable remediation for a canary group with an approval gate and a pre-change snapshot. Reasoning: separating detection from remediation limits blast radius and builds trust in the comparison logic before it can change production.
Traps
Screen-scraping CLI when a model-driven API exists; automating without a source of truth; alerting on every telemetry counter; no rollback or pre-change backup; ignoring partial failures in a multi-device run; and granting automation accounts full admin rights.