[j-nsp] BGP import policy not refreshing properly

[Will Orton]

I have 2 POPs each with a connection to a common upstream. The upstream 
is sending me MEDs, but lots of routes have (missing or 0) MEDs and I 
want to reset those to a fixed value so I can tweak them later.

So I have an import policy on each BGP session like so:

term setall-meds {
    from metric 0;
    then {
        metric 30000;
    }
}
term def {
    then {
        local-preference 110;
        accept;
    }
}
term rej {
    then reject;
}


I apply this on both routers and get, for example:

At POP A (M10i 9.3R1.7):
A Destination        P Prf   Metric 1   Metric 2  Next hop        AS path
* 64.152.0.0/13      B 170        110          0 >(TO POP B)      3356 I
                     B 170        110      30000 >(UPSTREAM AT A) 3356 I

At POP B (M10 9.3R3.8):
A Destination        P Prf   Metric 1   Metric 2  Next hop        AS path
* 64.152.0.0/13      B 170        110          0 >(UPSTREAM AT B) 3356 I


So the M10 at POP B doesn't appear to be applying the import policy and 
setting the MED to 30000. POP A as a result picks the route through B.  
(Yes, I waited more than the 15 minutes for POP B's CPU to go back to 
idle so the RE-333-768 churned through the whole table).

This resolved itself with a hard clear of the BGP session to the upstream 
at POP B. 'soft-intbound' clear at B didn't do it (other than pegging the 
RE CPU for another 15 minutes).

Any ideas? JUNOS bug? Old/decrepit RE getting non-deterministic 
with age? Do I really have to hard-clear the BGP session on the 'B' 
router any time I change the import policy now? :/


-Will Orton