So, to consider the classic query process:
1. A router marks a route active.
2. It determines there is no FS, so it puts the metric to infinity, and marks the route to be sent as a query.
3. A query is built including this route. When the query is transmitted, a timer is set--the SIA timer.
4. When the neighbor receives the query, it acknowledges receiving the query.
So, we know the neighbor has received the query. If they don't provide an answer to the query by the time the timer wakes up, the neighbor is marked SIA, and taken down. The original reason for this chain of events was:
1. We know the neighbor is still there.
2. We know we still have two way, reliable communication with the neighbor.
3. But, we know the neighbor isn't answering the query.
Hence, what we know is we've stepped outside the bounds of the state machine, as defined by DUAL. Once you step outside the state machine, the only reliable way to fix things is to restart the state machine--even though it's ugly.
EIGRP doesn't do this any longer. Instead, every so often, a router waiting on a reply to a query will send an "SIA-Query." If that packet is acknowledged, the timer is reset so we won't go SIA. _If_ you are in "new code," you shouldn't ever see SIAs, really--at least if you see one, you know something is _really_ wrong, it's not something you should see in "normal operation."
Remember, though, that the entire concept of resetting the neighbor when a route is marked SIA has nothing to do with the routing table--those are different things, in the code. The removal of the SIA route from the routing table and the neighbor reset are set off by the same event, but they are different parts of that process. The static route is called in because of the route removal on the route being marked SIA, not because of the neighbor being torn down.
HTH.
:-)
Russ
