How I did it radv sends a good-bye packet when the service is stopped, which causes a IPv6 route update on SoC side. And this update leads to an interface bouncing and causes traffic disruption even though the ToR device might already be isolated. This PR is to mitigate the traffic disruption issue during planned maintenance, by killing radv instead of stopping. So the cease packet won't be sent. How to verify it Verified on dev clusters: Traffic disruption was no longer reproducible. radv took the killing path if knob was off, radv would take the stopping path sign-off: Jing Zhang zhangjing@microsoft.com |
||
---|---|---|
.. | ||
arp_update | ||
asic_status.py | ||
asic_status.sh | ||
bgp.sh | ||
configdb-load.sh | ||
core_cleanup.py | ||
database.sh | ||
gbsyncd-platform.sh | ||
gbsyncd.sh | ||
lldp.sh | ||
mark_dhcp_packet.py | ||
mgmt-framework.sh | ||
radv.sh | ||
service_mgmt.sh | ||
snmp.sh | ||
sonic-netns-exec | ||
supervisor-proc-exit-listener | ||
swss.sh | ||
syncd_common.sh | ||
syncd.sh | ||
teamd.sh | ||
telemetry.sh | ||
update_chassisdb_config | ||
write_standby.py |