Configuring Graceful Restart and Long-lived Graceful Restart¶
Graceful restart and long-lived graceful restart BGP helper modes are supported for the TF control node and XMPP helper mode.
Application of Graceful Restart and Long-lived Graceful Restart¶
Whenever a BGP peer session is detected as down, all routes learned from the peer are deleted and immediately withdrawn from advertised peers. This causes instantaneous disruption to traffic flowing end-to-end, even when routes kept in the vrouter kernel in the data plane remain intact.
Graceful restart and long-lived graceful restart features can be used to alleviate traffic disruption caused by downs.
When configured, graceful restart features enable existing network traffic to be unaffected if TF controller processes go down. The TF implementation ensures that if a TF control module restarts, it can use graceful restart functionality provided by its BGP peers. Or when the BGP peers restart, TF provides a graceful restart helper mode to minimize the impact to the network. The graceful restart features can be used to ensure that traffic is not affected by temporary outage of processes.
Graceful restart is not enabled by default.
With graceful restart features enabled, learned routes are not deleted when sessions go down, and the routes are not withdrawn from the advertised peers. Instead, the routes are kept and marked as ‘stale’. Consequently, if sessions come back up and routes are relearned, the overall impact to the network is minimized.
After a certain duration, if a downed session does not come back up, all remaining stale routes are deleted and withdrawn from advertised peers.
The graceful restart and long-lived graceful restart features can be enabled only for BGP peers in TF 3.2.
BGP Graceful Restart Helper Mode¶
The BGP helper mode can be used to minimize routing churn whenever a BGP session flaps. This is especially helpful if the SDN gateway router goes down gracefully, as in an rpd crash or restart on an MX Series Junos device. In that case, the contrail-control can act as a graceful restart helper to the gateway, by retaining the routes learned from the gateway and advertising them to the rest of the network as applicable. In order for this to work, the restarting router (the SDN gateway in this case) must support and be configured with graceful restart for all of the address families used.
The graceful restart helper mode is also supported for BGP-as-a-Service (BGPaaS) clients. When configured, contrail-control can provide a graceful restart or long-lived graceful restart helper mode to a restarting BGPaaS client.
The following are highlights of the graceful restart and long-lived graceful restart features.
Configuring a non-zero restart time enables the ability to advertise graceful restart and long-lived graceful restart capabilities in BGP.
Configuring helper mode enables the ability for graceful restart and long-lived graceful restart helper modes to retain routes even after sessions go down.
With graceful restart configured, whenever a session down event is detected and a closing process is triggered, all routes, across all address families, are marked stale. The stale routes are eligible for best-path election for the configured graceful restart time duration.
When long-lived graceful restart is in effect, stale routes can be retained for a much longer time than that allowed by graceful restart alone. With long-lived graceful restart, route preference is retained and best paths are recomputed. The community marked LLGR_STALE is tagged for stale paths and re-advertised. However, if no long-lived graceful restart community is associated with any received stale route, those routes are not kept, instead, they are deleted.
After a certain time, if a session comes back up, any remaining stale routes are deleted. If the session does not come back up, all retained stale routes are permanently deleted and withdrawn from the advertised peer.
XMPP Helper Mode¶
TF supports for long-lived graceful restart (LLGR) with XMPP helper mode. Graceful restart and long lived graceful restart can be enabled using the Tungsten Fabric WebUI or by using the provision_control script.
The helper modes can also be enabled via schema, and can be disabled
selectively in a contrail-control node for BGP or XMPP sessions by
gr_helper_disable in the
/etc/contrail/contrail-control.conf configuration file.
Graceful restart parameters are configured in the
global-system-config of the schema. They can be configured by means
of a provisioning script or by using the Tungsten Fabric WebUI.
Configure a non-zero restart time to advertise for graceful restart and long-lived graceful restart capabilities from peers.
Configure helper mode for graceful restart and long-lived graceful restart to retain routes even after sessions go down.
Configuration parameters include:
disablefor all graceful restart parameters:
bgp-helper-enableto enable graceful restart helper mode for BGP peers in contrail-control
xmpp-helper-enableto enable graceful restart helper mode for XMPP peers (agents) in contrail-control
The following shows configuration by a provision script.
/opt/contrail/utils/provision_control.py --api_server_ip 10.xx.xx.20 --api_server_port 8082 --router_asn 64512 --admin_user admin --admin_password <password> --admin_tenant_name admin --set_graceful_restart_parameters --graceful_restart_time 60 --long_lived_graceful_restart_time 300 --end_of_rib_timeout 30 --graceful_restart_enable --graceful_restart_bgp_helper_enable
The following are sample parameters:
-set_graceful_restart_parameters --graceful_restart_time 300 --long_lived_graceful_restart_time 60000 --end_of_rib_timeout 30 --graceful_restart_enable --graceful_restart_bgp_helper_enable
When BGP peering with Juniper Networks devices, Junos must also be explicitly configured for graceful restart/long-lived graceful restart, as shown in the following example:
set routing-options graceful-restart set protocols bgp group <a1234> type internal set protocols bgp group <a1234> local-address 10.xx.xxx.181 set protocols bgp group <a1234> keep all set protocols bgp group <a1234> family inet-vpn unicast graceful-restart long-lived restarter stale-time 20 set protocols bgp group <a1234> family route-target graceful-restart long-lived restarter stale-time 20 set protocols bgp group <a1234> graceful-restart restart-time 600 set protocols bgp group <a1234> neighbor 10.xx.xx.20 peer-as 64512
The graceful restart helper modes can be enabled in the schema. The
helper modes can be disabled selectively in the
contrail-control.conf for BGP sessions by configuring
gr_helper_disable in the
The following are examples:
/usr/bin/openstack-config /etc/contrail/contrail-control.conf DEFAULT gr_helper_bgp_disable 1
/usr/bin/openstack-config /etc/contrail/contrail-control.conf DEFAULT gr_helper_xmpp_disable 1
service contrail-control restart
For more details about graceful restart configuration, see https://github.com/tungstenfabric/docs/blob/master/wiki/tf-controller/Graceful-Restart.md.
Cautions for Graceful Restart¶
Be aware of the following caveats when configuring and using graceful restart.
Using the graceful restart/long-lived graceful restart feature with a peer is effective either to all negotiated address families or to none. If a peer signals support for graceful restart/long-lived graceful restart for only a subset of the negotiated address families, the graceful restart helper mode does not come into effect for any family in the set of negotiated address families.
Because graceful restart is not yet supported for contrail-vrouter-agent, the parameter should not be set for
graceful_restart_xmpp_helper_enable. If the vrouter agent restarts, the data plane is reset and the routes and flows are reprogrammed anew, which typically results in traffic loss for several seconds for new and /existing flows.
Graceful restart/long-lived graceful restart is not supported for multicast routes.
Graceful restart/long-lived graceful restart helper mode may not work correctly for EVPN routes, if the restarting node does not preserve forwarding state for EVPN routes.