Symptom

In multisite back-to-back scenario in sites A and B. Where there is no full-mesh. e.g. Spine A1 is connected to Spine B1 only. Spine A2 is connected to Spine B2 only As in multisite white paper figure 51: https://www.cisco.com/c/en/us/solutions/collateral/data-center-virtualization/application-centric-infrastructure/white-paper-c11-739609.html When Spine B2 reloads: - Routes from this spine are advertised with a high metric in OSPF and ISIS so this route is not taken. This works well. - Routes from A2 spine to B2 form and are shared with the site's leaves as soon as OSPF between A2 and B2 come up. Causing leaves in site A to route to A2 and then B2, before B2 has had a chance to converge COOP. This causes packet loss after the spine is reloaded. Since the OSPF/ISIS redistribution mechanisms are shared with multipod, this scenario also applies to back-to-back multipod topologies.

Conditions

- ACI multisite/multipod is configured in a back-to-back scenario. - There is no full mesh. - One of the spines in either site is reloaded.

Workaround

If performing a scheduled reload of a spine, shutting down the link from it's directly connected spine on the other site/pod before issuing the reload.

Further Problem Description