Using pacemaker VIP addresses in dhcpd.conf

[Jorge Merlino]

This issue can occur in HA installations where there are VIP addresses assigned by pacemaker on the same VLAN where DHCP is enabled. MAAS could choose a secondary VIP address and use it in the address and/or peer address parameters on the failover DHCP configuration on dhcpd.conf. In the event of a failover, the IP will be moved somewhere else and this configuration would be incorrect.

It seems that MAAS does not check for this when building the failover configuration in the make_failover_peer_config method on dhcpd.conf.

I was able to reproduce it in snap version 3.2.8-12050-g.fc2c8d020.

[Christian Grabowski (cgrabowski)]

Can reproduce. MAAS does not currently have a way of differentiating a VIP from pacemaker and a regularly assigned IP on an interface.

Separate, if using the Ansible Playbooks, the VIP is only for the Postgres primary. If you can separate the rack controllers from those hosts (ideal for HA), this can be worked around.

[Jorge Merlino (jorge-merlino)]

Thank you for reviewing it. If it is not possible to tell the difference maybe we should at least add that to the documentation to raise awareness.

[Jerzy Husakowski (jhusakowski)]

We think this self-heals after a while, when MAAS observes the appropriate changes and regenerates the dhcp config. Does that not happen? Can you attach appropriate logs from when this problem occurs?

Note that MAAS HA improvements are on the roadmap, and this particular issue may be handled with the bulk of other HA-related ones.

[Jorge Merlino (jorge-merlino)]

It self-heals when the VIP moves to another host but when the VIP is in the host MAAS might use it. Is that what you mean?