Skipping scale-up if MachineTypeNotAvailableAnnotation is present on MachineDeployment

[hardikdr]

What this PR does / why we need it: This PR enables autoscaler to skip the scale-up if MachineTypeNotAvailable annotation is present on the machine-deployment.
This annotation is set by the MCM, when certain machine-types are not out of capacity and can't be further scaled-up.

As an outcome of this change, autoscaler will pick up alternate node-group if one is not capable of scaling.

Related to : #35
Which issue(s) this PR fixes:
Fixes #

Special notes for your reviewer:

Release note:

Autoscaler avoids scaling-up the MachineDeployment if MachineTypeNotAvailableAnnotation is present

[prashanth26]

Hi @hardikdr ,

Were these changes tested with a real setup?

[hardikdr]

Yes, I have tested this with the related changes at MCM #454, works fine.

[prashanth26]

hold

[CLAassistant]

All committers have signed the CLA.

[jsravn]

Is this the correct approach? From looking at it, upstream autoscaler already handles this but you have to return unregistered instances in the autoscaler cloudprovider interface, which this fork isn't doing properly.

[hardikdr]

@jsravn That's true, and also one of the reasons why I stopped working on this approach for now.
The interface in this version of CA, isn't mature enough, and I am rebasing the autoscaler to the latest version here: https://github.com/gardener/autoscaler/tree/machine-controller-manager-provider. I'll then be looking at this topic again and find a more natural way to fix it.

[hardikdr]

/hold

[jsravn]

@hardikdr Ah cool. I was also going to try rebasing. I should have a PR soon which fixes it to return unregistered instances - so at least this functionality will work as expected in the existing code (if it fails to scale up the machinedeployment, autoscaler will give up after max-node-provision-time and try the next one).

[hardikdr]

Sounds great. We are discussing this issue at length here: #35.

Also, you might want to make sure, your changes are on top of the machine-controller-manager-provider, which is also a default branch now.

[jsravn]

@hardikdr FYI I opened a PR at #58.

[prashanth26]

We no longer need this PR right? Thanks to this - #58?

[hardikdr]

No, we don't need this anymore.
/close