How to juju deploy a service to units of multiple architectures with different configurations?

Question

I have MAAS nodes of 2 different architectures (amd64 and arm64) that I need to deploy nova-compute on. I need 2 different slightly-different sets of configuration settings to work properly.

I am deploying an amd64 nova-compute node as follows:

juju deploy --config nova-amd64-config.yaml   \
            --constraints "arch=amd64"        \
            nova-compute

I also need some arm64 nova-compute nodes, and need to specify a different configuration file for those as well, i.e.:

juju deploy --config nova-rm64-config.yaml    \
            --constraints "arch=arm64"        \
            nova-compute

But, once the original nova-compute service is deployed, the only available option is to use juju add-unit.

Is it possible to juju deploy a service specifying one configuration file, then juju add-unit to other nodes, specifying a different configuration file?

Or is there a different way I should approach this?

Answer 1

Configuration values are set at the service level not at the unit level. You would either need a charm which had multiple values and able to pass the proper values through to the appropriate units or have two nova-compute services, one for arm and one for amd.