zebra: nhg code selection doesnot check invalid status

[hnattamaisub]

Problem description:
1)In some corner cases in zebra, “Failed to enqueue dataplane  install” happened for routes which was caused because of INVALID nhg.
2)Our NHG code logic, check all nexthop entries is valid or not,but it does not check whether the nhg itself is valid or not during our checks which in turn cause route install failures in zebra.

Fix:
Handled the nhg logic to check for validity during selection.

[mjstapp]

can you say some more about the cases where this would happen - what's the order of events that get zebra into this condition?

[hnattamaisub]

can you say some more about the cases where this would happen - what's the order of events that get zebra into this condition?

Sure, will explain more on what was happening.
Trigger: (Multiple Interface flaps sequentially between vrrp nodes in multiple node topo which has vrrp and ospf )
Issue: OSPF route was affected between some nodes. Ping failure due to lost routes.

1)When a nexthop becomes invalid (e.g., interface down), NHGs are NOT deleted immediately. Instead, they're marked as INVALID (flags=0x0) and cleaned up later.
During route churn, Zebra's RIB may temporarily have multiple route entries for the same prefix (some with valid NHGs, some with invalid NHGs). This is EXPECTED behavior due to our code design.

2)Zebra processes (rib_process) during that timeframe with invalid NHG and enqueue dataplane install failed.

Note (This is very very corner case , not hit very frequently , but this check makes sure VALID NHG is selected by zebra during that flap events , which in turn makes the installation success and avoided the issue scenario)

Please let me know if my understanding/approach is wrong or more details is needed. Thanks.

[mjstapp]

thanks for the explanation - that makes sense to me.

[donaldsharp]

without this fix what state are we in?

[hnattamaisub]

without this fix what state are we in?

Without this fix, we face "Failed to enqueue dataplane install" errors in zebra and ospf routes are not installed in some nodes causing traffic forwarding issue after interface down/up trigger. (More specifically ping failures itself).

[hnattamaisub]

ci:rerun