Symptom
Use case: Customer wants to dual-home non-SDA switch trunk ports with flexlink+ to trunk ports on two different SDA FE e.g. FE1 and FE2. On the non-SDA switch, flexlink+ (or other vendor equivalent technology) forwards packets to either FE1 or FE2. Flexlink+ assures active/standby uplinks on non-SDA switch. Non-SDA switch is never using active/active uplinks.
Problem statement: If non-SDA switch uplink to FE1 is active, and FE1 reloads, it is possible to hit race condition between underlay IGP and LISP. When FE1 reloads, its RLOC is withdrawn from underlay IGP. Sometimes FE and Border IGP will converge before CP IGP converges. In scenario where FE and/or Border IGP converges first, FE and/or Border will have no IGP route to FE1 RLOC and will request new RLOC from CP. As CP IGP has not yet converged, CP LISP session state to FE1 is still up, and CP returns RLOC of FE1. FE and/or Border then holds FE1 RLOC for 24 hours in LISP map-cache even though there is no underlay IGP route to FE1 RLOC. This causes complete loss/outage for traffic destined to EIDs connected to non-SDA switch.
Workaround
Manipulate underlay IGP timers to ensure CP IGP converges BEFORE FE and Border IGP. In case of OSPF or ISIS underlay IGP this can be accomplished with manual SPF timer manipulations on CP, FE and Borders.
