BGP graceful restart doesn't retain routes when configured per-peer about frr HOT 7 CLOSED

Okay, thanks for reporting. Will check it.

from frr.

Graceful bgpd stop does not trigger graceful restart, that's expected. You have to kill bgpd.

from frr.

SIGINT and SIGKILL produce the same result. In both cases, bgpd (technically Linux) disconnects TCP with proper termination handshake (FIN/ACK) without sending BGP notification. So, the peer shouldn't be able to distinguish those two cases.

from frr.

Ah, you mean per-peer? Then it might be a bug. I tested (and we have a test) global, it works fine.

from frr.

Yes, right. Global works fine and tested, but per-peer doesn't work for me and is untested. That's why I made this PR (#15827)

from frr.

I tested GR per-neighbor, and it works fine. It does not work for you because you have neighbor 192.168.255.1 graceful-restart before the real neighbor is created, and when vtysh parses the config, this statement is just ignored. Put it after the neighbor is defined with remote-as ....

from frr.

Oh, I see. Thanks for investigating it. My bad. I always forget that FRR's config is a series of commands and not a declarative configuration.

from frr.