You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
we have three types of different hardware for controller(cpu only), worker(cpu only) and gpu_worker(with gpus).
Feature
it would be nice to have the same group snippets for baremetal too(as for digital ocean), so we can use controller_clc_snippets or worker_clc_snippets, .... for our types of hardware
Tradeoffs
no need to define each element of a group again
The text was updated successfully, but these errors were encountered:
On cloud providers, controllers and workers should be homogeneous so clusters accept controller_clc_snippets and worker_clc_snippets lists. On bare-metal, machines may be heterogeneous, so a key value mapping is used instead.
If your bare-metal machines happen to be identical hardware right now, that's a temporary state - at some point you'll buy some shiny new machines, but not want to throw away the old ones. In fact, your example where some workers are CPU only and others may need GPU customizations is already an example.
If you find some machines uses the same snippets, just organize them into a list.
I have no plan to complicate bare-metal by supporting both a map and list input for the same information, when the clc_snippets map already handles cases.
Feature Request
we have three types of different hardware for controller(cpu only), worker(cpu only) and gpu_worker(with gpus).
Feature
it would be nice to have the same group snippets for baremetal too(as for digital ocean), so we can use controller_clc_snippets or worker_clc_snippets, .... for our types of hardware
Tradeoffs
no need to define each element of a group again
The text was updated successfully, but these errors were encountered: