HPA scaling event (likely scaling down event) seems to affect analysisruns

[schlags]

We're noticing some odd behavior in our pre/post-analysisruns for BlueGreen.
It seems to be a relation where if the HPA scales down the desired pod count in the RS, the reconciliation of pre/post analysis will either terminate the running analysisrun and continue with the rollout, or it will see a completed analysisrun and throw a "Collision" log. Eventually spinning up a .1 and sometimes a .2 AnalysisRun.

The former ("Run Terminated" with Success status) is very concerning since it allows our deployment to continue without completing analysis.

I haven't been able to find any conclusive documentation around how this can be mitigated. Any thoughts?

[gavin-db]

Allegedly duplicate: #2938 (comment)