Page MenuHomePhabricator

Netbox: manage VRRP priorities
Open, LowPublic

Description

This is a follow up from T311218: Netbox: use FHRP Groups feature
The sub-task allowed us to have a cleaner IPAM by getting rid of duplicate VRRP records (from one per device to a single VRRP group).

It also allows us to manage VRRP priorities (currently set to 0 in Netbox).

eqiad/codfw routers uses VRRP pinning to artificially balance traffic through the two main eqiad<->codfw transports.
Other sites use the default value of 100.

Driving the VRRP priority from Netbox would help us improve our automation, specifically when we need to drain a router for maintenance (eg. set a lower priority on the node to "depool").

2 reasons to set this task as a low priority:

  • Moving the values to Netbox (when applicable) is easy, but the tooling to change the values on all the interfaces would requires either a custom script or a cookbook and could be more significant work. For a task that we don't perform often. This could however be combined with other similar tasks such as T260355: Transport circuits draining automation.
  • The new DC design is moving away from VRRP, so this work would become deprecated in the next few years

Event Timeline

ayounsi created this task.
Restricted Application added a subscriber: Aklapper. · View Herald Transcript

Good stuff!

I was just actually looking at this for the cloudsw, to try to get better balancing of the uplinks from c8/d5 (as currently d5 being used for both cloud and prod realm and discards are high). The templates for those don't currently have any mechanism to set priority. If that's needed there I'll maybe have a look at doing the homer changes.

But even there we are moving away from VRRP there also, to the per-row subnets. Overall it's worth doing as the VRRP / VC switches will be around for a while, bug agreed probably not too high priority.