MaxParkedNodes to control cluster size limits

kaarthis commented 4 months ago

Customers can use MaxParkedNodes property to define the max. no. of blocked nodes allowed during an upgrade, thereby also define the maximum size a cluster (i.e current size + #undrainablenodes) can grow during and after an upgrade. Max-surge will no longer be used to define the max limits going forward with this proposal.

max-surge wont be used to define the size by which cluster grows . This is controlled by maxundrainablenodes/MaxParkedNodes .

Quota check on prevalidation will be based off maxundrainablenodes/MaxParkedNodes i.e current size + maxundrainablenodes/MaxParkedNodes should satisfy quota otherwise user has to make amends on either quota or reduce the maxundrainablenodes/MaxParkedNodes.

Default size of maxundrainablenodes/MaxParkedNodes is same as Max-Surge today if not set i.e 10%. Maximum value is 100% or tolerate upto full cluster size going blocked.

maxundrainablenodes/MaxParkedNodes is a percentage between 1 and 100 (inclusive). Default is 10%.

Max-surge today is a percentage between 1 and 100 (inclusive). It cannot be greater than the maxundrainablenodes/MaxParkedNodes. Users will be getting an error to correct this on update validations.

Every upgrade iteration whenever there is a blocked node it eats into the 'max-surge' capacity by that many blocked / parked nodes.

eg: if maxundrainablenodes/MaxParkedNodes is 10%