Similar to what has been done for eqord in T167841#5414996
Even if eqdfw and knams have more redundancy to their neighbor caching POP. Advertising the site's prefixes only if the router receives a proper contributing route would prevent an outage in case the network POP becomes an island.
We should also investigate why cr2-knams has set routing-options aggregate route 91.198.174.224/28.
Similarly 91.198.174.192/27 is probably not needed on cr2-knams as it doesn't talk directly to LVS
[edit routing-options rib inet6.0 aggregate route 2620:0:860::/46] - policy BGP_aggregate_contributors_eqiad; + policy BGP_from_LVS; [edit routing-options rib inet6.0 aggregate route 2620:0:860::/48] - policy BGP_aggregate_contributors_codfw; + policy BGP_from_LVS; [edit routing-options aggregate route 208.80.152.0/22] - policy BGP_aggregate_contributors_eqiad; + policy BGP_from_LVS; [edit routing-options aggregate route 208.80.152.0/23] - policy BGP_aggregate_contributors_codfw; + policy BGP_from_LVS;
Note that the /22 and /46 even if present in the aggregates are not advertised to the world (not in the bgp-out4/6 prefix-list).
Mentioned in SAL (#wikimedia-operations) [2020-03-25T09:56:35Z] <XioNoX> change aggregate policy for 2620:0:860::/46 on cr2-eqdfw - T236785
Mentioned in SAL (#wikimedia-operations) [2020-03-25T10:19:46Z] <XioNoX> change aggregate policy for v4 prefixes on cr2-eqdfw - T236785
[edit routing-options rib inet6.0 aggregate route 2620:0:862::/48] - policy BGP_aggregate_contributors; + policy BGP_from_LVS; [edit routing-options aggregate route 91.198.174.0/24] - policy BGP_aggregate_contributors; + policy BGP_from_LVS;
Note that the other advertised aggregates (2a02:ec80::/32 and 185.15.56.0/22) keep BGP_aggregate_contributors as they don't have any LVS contributing prefix.
Mentioned in SAL (#wikimedia-operations) [2020-03-25T10:37:52Z] <XioNoX> change aggregate policy for 2620:0:862::/48 on cr3-knams - T236785
Previous change rolled back as:
* 2620:0:862::/48 Self I
was not being advertised to the world anymore
While there are contributing LVS routes:
2620:0:862:ed1a::1/128
*[BGP/170] 2w5d 10:23:40, MED 0, localpref 100, from 91.198.174.244
AS path: 64600 I, validation-state: unverifiedWhich also match the as-path regex:
cr3-knams# run show route aspath-regex "^(65002|65001)? 64600.*" | match 2620:0:862:ed1a::1 2620:0:862:ed1a::1/128
Mentioned in SAL (#wikimedia-operations) [2020-03-26T08:27:41Z] <XioNoX> troubleshot v6 conditional advertisement from cr3-knams - T236785
Mystery solved.
The reason for the route to not be accepted was:
Inactive reason: Unusable path
This was due to:
rib inet6.0 aggregate route 2620:0:862:ed1a::/64
Causing the contributing BGP routes to be "caught" in the above aggregate.
In consequences, 2620:0:862::/48 was only "seeing" 2620:0:862:ed1a::/64 which didn't match the policy rule.
Similarly, as mentioned in the description the following need to be removed for the v4 equivalent:
[edit routing-options aggregate] - route 91.198.174.224/28; - route 91.198.174.192/27;
As they are not in use.
Mentioned in SAL (#wikimedia-operations) [2020-03-26T09:00:52Z] <XioNoX> push v4 conditional advertising on cr3-knams - T236785