Following a rancher HA setup from the docs: https://rancher.com/docs/rancher/v2.x/en/installation/how-ha-works/ I will need to build a new HA Rancher Cluster and use NGINX as a LB of choice.
However using one NINGX will introduce a single point of failure, so I am thinking to put 2x LBs and use DNS round robin.
Will this kind of setup work with rancher?
Thank you
Yes that can work, however if one of the NGINX servers goes down, then the DNS wiil still send requests to the node that is down, and half of the requests will still not work.
The better solution (and what we are using) is to use Keepalived on the NGINX servers to have a VIP that is shared between the two servers. If the server that currently owns the VIP goes down, then the other server will assume the IP and continue working.
Perfect, thank you. I see now that NGINX also have their own HA solution with nginx-ha-keepalived and nginx-sync so I guess I might go with than one, but thank you for the confirmation.
The keepalived solution (without using NGINX Plus and its keepalived-ha-module) works fine but active sessions get lost on master/backup failures. Usually not a big problem if a client is capable of repeating invocations…
Using pacemaker and corosync seems to be an option too.
Didn’t try yet but… maybe someone needs it.