Unexpected behavior of 'clear adj-rib out' and 'flush adj-rib out' with route-refresh feature enabled

[moonsky219]

Describe the bug

When route-refresh feature is enabled, neighbor x.x.x.x clear adj-rib out does not withdraw routes in its adj-rib out table, actually I didn't observe any BGP messages with tcpdump. neighbor x.x.x.x flush adj-rib out triggers an OPEN message to switch instead of resending routes in its adj-rib out table.

To Reproduce

Steps to reproduce the behavior:

• Add route-refresh in config and establish BGP session between exabgp and switch x.x.x.x.
• Send several routes to the switch.
• Try commands neighbor x.x.x.x clear adj-rib out and neighbor x.x.x.x flush adj-rib out

Expected behavior

With route-refresh capability negotiated between ExaBGP and switch, neighbor x.x.x.x clear adj-rib out should still withdraw all routes in adj-rib out table. neighbor x.x.x.x flush adj-rib out should still resend all routes in adj-rib out table.

Environment (please complete the following information):

• OS: Debian GNU/Linux 9.13 (stretch)

• Version 4.2.18 with Python 3.7.3

  neighbor x.x.x.x {
      local-address y.y.y.y;
      router-id y.y.y.y;
      local-as 1;
      peer-as 2;
      description "xxx";
  
      family {
              ipv4 unicast;
      }
  
      capability {
              add-path send/receive;
              route-refresh;
      }
  
      api {
              processes [ announce-routes ];
      }
  }

Additional context

• Both Route refresh capability (2) and Enhanced route refresh capability (70) are negotiated.
• I also tried to fire neighbor x.x.x.x clear adj-rib out and neighbor x.x.x.x flush adj-rib out after the BGP session is established with route-refresh, I notice that all following neighbor x.x.x.x announce route ... seems not working.

[thomas-mangin]

Thank you for the report, I will look into it when I can. I will first have to build a set of tests for the CI and then debug it. It will not be immediate. What is the impact of this issue on your work?

[moonsky219]

Hi @thomas-mangin, thank for your response. This issue blocked our efforts to improve our network availability. It would be very nice if you can look into the issue at your convenience time.

[thomas-mangin]

we now have a test for this and it is correctly failing (after fixing other rib related things) - just to indicate that I am working on the issue, but it is not yet finished.

[thomas-mangin]

Could you re-perform your tests with the latest master please.

[thomas-mangin]

Closing the issue. Feel free to re-open if you find that the tests api-rib and api-rr-rib did not cover your issue reported.