nodeAffinity is not correctly updated with DeepDerivative

[zshi-redhat]

https://bugzilla.redhat.com/show_bug.cgi?id=1914066

[fedepaol]

Ouch! Couple of questions:

• Is this because the nodeAffinity is reset when deleting the policy (and as a 0 field is not compared)?
• Should we replace the other occurrences of DeepDerivative too?
• Did you check if this brings back the unwanted extra reconciliation loops?

[fedepaol]

Typo: "chnange".
I also wonder if we should delete the comment abovee

[zshi-redhat]

I can delete previous comment if you prefer, just thought it might be useful for reference if later we hit the endless loop issue again.

[fedepaol]

I don't have a strong opinion, I guess it's fine since the comment below explains it pretty clearly.

[zshi-redhat]

fixed typo

[zshi-redhat]

Ouch! Couple of questions:

• Is this because the nodeAffinity is reset when deleting the policy (and as a 0 field is not compared)?

NodeAffinity is not updated when deleting policy, the legacy nodeAffinity remains in the daemonset spec.

• Should we replace the other occurrences of DeepDerivative too?

I didn't know any issue with the other one, I think we can keep it as it is for now.

• Did you check if this brings back the unwanted extra reconciliation loops?

Yes, there is no endless loop in sriov operator log with DeepEqual.

[fedepaol]

LGTM (apart from the typo)

[fedepaol]

LGTM

[pliurh]

With DeepEqual, I suppose we will hit the issue which @fedepaol tried to fix since it will be always false. Can we add a condition to check whether the nodeAffinity should be removed from the spec of DS, instead of using DeepEqual?

[zshi-redhat]

With DeepEqual, I suppose we will hit the issue which @fedepaol tried to fix since it will be always false. Can we add a condition to check whether the nodeAffinity should be removed from the spec of DS, instead of using DeepEqual?

Done