how to trigger a graceful restart for a bgp session
hi, if we just did `systemctl restart frr`, the bgpd process will be stopped by `kill -2`, and it would send cease/peer de-config notification to its peer. its peer will flush the routes insteading of make them stale. if we did `neighbor xxxxx shutdown`,then `no neighbor xxxxxx shutdown` via vtysh, it would send cease/administrative shutdown notification. graceful restart did not work too. Now, we have changed frrcommon.sh by 's/kill -2/kill -9/g', when bgpd is killed by `kill -9`, it would not send any notifications, and this time, graceful restart functionality works, its peer makes the routes be stale and forwarding is perserverd during frr restarting. however, we think this approach is ugly. So can anyone tell us what is the corrent approach to restart bgp session gracefully. we have to restart bgp session for upgrading to a new version or changing some parameters, etc. thanks! best regards, Xinyong
Xinyong, If you do "systemcl restart frr" then it would also restart zebra also right? So would that not be a problem. AFAIK right now -9 is the only way to restart BGP trigger GR. Thanks Santosh P K On 13/04/20, 9:25 AM, "frog on behalf of Xinyong Wang" <frog-bounces@lists.frrouting.org on behalf of wang.xy.chn@gmail.com> wrote: hi, if we just did `systemctl restart frr`, the bgpd process will be stopped by `kill -2`, and it would send cease/peer de-config notification to its peer. its peer will flush the routes insteading of make them stale. if we did `neighbor xxxxx shutdown`,then `no neighbor xxxxxx shutdown` via vtysh, it would send cease/administrative shutdown notification. graceful restart did not work too. Now, we have changed frrcommon.sh by 's/kill -2/kill -9/g', when bgpd is killed by `kill -9`, it would not send any notifications, and this time, graceful restart functionality works, its peer makes the routes be stale and forwarding is perserverd during frr restarting. however, we think this approach is ugly. So can anyone tell us what is the corrent approach to restart bgp session gracefully. we have to restart bgp session for upgrading to a new version or changing some parameters, etc. thanks! best regards, Xinyong _______________________________________________ frog mailing list frog@lists.frrouting.org https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Flists.frrouting.org%2Flistinfo%2Ffrog&data=02%7C01%7Csapk%40vmware.com%7C244145244b284efbe06708d7df5e85eb%7Cb39138ca3cee4b4aa4d6cd83d9dd62f0%7C0%7C1%7C637223469407547378&sdata=hbWpBPo3LJ1hOMLlGh71NWPsBcOZlEcvkgcgIbj0Pk4%3D&reserved=0
Santosh, Yeah, in most cases, zebra is started with "--retain" enabled, so stopping zebra by "kill -9" is not a problem. If a bgp session is graceful restart enabled, what if it din't send "cease/cease peer unconfig" notification in "bgp_terminate"? Thanks Xinyong On Mon, Apr 13, 2020 at 1:03 PM Santosh P K <sapk@vmware.com> wrote:
Xinyong, If you do "systemcl restart frr" then it would also restart zebra also right? So would that not be a problem. AFAIK right now -9 is the only way to restart BGP trigger GR.
Thanks Santosh P K
On 13/04/20, 9:25 AM, "frog on behalf of Xinyong Wang" <frog-bounces@lists.frrouting.org on behalf of wang.xy.chn@gmail.com> wrote:
hi,
if we just did `systemctl restart frr`, the bgpd process will be stopped by `kill -2`, and it would send cease/peer de-config notification to its peer. its peer will flush the routes insteading of make them stale.
if we did `neighbor xxxxx shutdown`,then `no neighbor xxxxxx shutdown` via vtysh, it would send cease/administrative shutdown notification. graceful restart did not work too.
Now, we have changed frrcommon.sh by 's/kill -2/kill -9/g', when bgpd is killed by `kill -9`, it would not send any notifications, and this time, graceful restart functionality works, its peer makes the routes be stale and forwarding is perserverd during frr restarting. however, we think this approach is ugly.
So can anyone tell us what is the corrent approach to restart bgp session gracefully. we have to restart bgp session for upgrading to a new version or changing some parameters, etc.
thanks!
best regards, Xinyong
_______________________________________________ frog mailing list frog@lists.frrouting.org https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Flists.frrouting.org%2Flistinfo%2Ffrog&data=02%7C01%7Csapk%40vmware.com%7C244145244b284efbe06708d7df5e85eb%7Cb39138ca3cee4b4aa4d6cd83d9dd62f0%7C0%7C1%7C637223469407547378&sdata=hbWpBPo3LJ1hOMLlGh71NWPsBcOZlEcvkgcgIbj0Pk4%3D&reserved=0
I guess right approach would be to implement RFC https://tools.ietf.org/html/rfc8538. This is not supported yet. Thanks Santosh P K On 13/04/20, 11:16 AM, "Xinyong Wang" <wang.xy.chn@gmail.com> wrote: Santosh, Yeah, in most cases, zebra is started with "--retain" enabled, so stopping zebra by "kill -9" is not a problem. If a bgp session is graceful restart enabled, what if it din't send "cease/cease peer unconfig" notification in "bgp_terminate"? Thanks Xinyong On Mon, Apr 13, 2020 at 1:03 PM Santosh P K <sapk@vmware.com> wrote: > > Xinyong, > If you do "systemcl restart frr" then it would also restart zebra also right? So would that not be a problem. AFAIK right now -9 is the only way to restart BGP trigger GR. > > Thanks > Santosh P K > > On 13/04/20, 9:25 AM, "frog on behalf of Xinyong Wang" <frog-bounces@lists.frrouting.org on behalf of wang.xy.chn@gmail.com> wrote: > > hi, > > if we just did `systemctl restart frr`, the bgpd process will be > stopped by `kill -2`, and it would send cease/peer de-config > notification to its peer. its peer will flush the routes insteading of > make them stale. > > if we did `neighbor xxxxx shutdown`,then `no neighbor xxxxxx shutdown` > via vtysh, it would send cease/administrative shutdown notification. > graceful restart did not work too. > > Now, we have changed frrcommon.sh by 's/kill -2/kill -9/g', when bgpd > is killed by `kill -9`, it would not send any notifications, and this > time, graceful restart functionality works, its peer makes the routes > be stale and forwarding is perserverd during frr restarting. however, > we think this approach is ugly. > > So can anyone tell us what is the corrent approach to restart bgp > session gracefully. we have to restart bgp session for upgrading to a > new version or changing some parameters, etc. > > thanks! > > best regards, > Xinyong > > _______________________________________________ > frog mailing list > frog@lists.frrouting.org > https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Flists.frrouting.org%2Flistinfo%2Ffrog&data=02%7C01%7Csapk%40vmware.com%7C42da0fe884884a77ad3508d7df6df88a%7Cb39138ca3cee4b4aa4d6cd83d9dd62f0%7C0%7C0%7C637223535724871521&sdata=iqrFgeGdXMaqaQYuKBlQ7IFd8q8dwra4uQC1ap8CnbE%3D&reserved=0 > >
participants (2)
-
Santosh P K -
Xinyong Wang