Set limits for tablet nodes leads to full cluster upgrade with downtime

[Arustinal]

how to reproduce

1. create cluster with default spec without requests for tablet nodes
2. wait for setup
3. set requests
4. see the whole system goes down

Is it really required?

[l0kix2]

I will try to reproduce, according to my expectations tablet nodes spec change shouldn't end in other components update.
Are you sure core image or some master-related configuration wasn't changed?

[l0kix2]

Oh, I see, we are consider tablet nodes as a component that triggers full update indeed.
Will look if it is necessary.

[l0kix2]

I don't think the updating all the components because of tablet nodes is necessary at this point. I assume originally it was done to control downtime via isFullUpdate flag — so if you update stateful components like master, data nodes, tablet nodes — you need to set isFullUpdate=true.
If we change current behaviour, it may break some users flows.
We will try to offer more granular mechanism for updates of different components in the future.