Over in kubernetes/kubernetes#132216, Dims managed to reduce the dead code in the Kubernetes binaries.

I had a look and all the libraries that prevent dead code elimination all come from upstream k/k, so we should be able to leverage these benefits in the VPA once k/k 1.34 is out.

Dims also wrote a script to ensure that this problem doesn't come back (see kubernetes/kubernetes#132290)

I gave it a test, and came up with these results:

| Binary               | Before (bytes) | After (bytes) | Reduction (bytes) | Percentage Reduction |
|----------------------|----------------|---------------|-------------------|----------------------|
| admission-controller | 73,996,402     | 43,305,218    | 30,691,184        | 41.48%               |
| recommender          | 74,984,130     | 43,873,874    | 31,110,256        | 41.49%               |
| updater              | 74,460,562     | 43,481,554    | 30,979,008        | 41.60%               |

Once k/k 1.34 is out, I'll ensure that we use these benefits and have the automation in place to not regress.

/area vertical-pod-autoscaler
/assign
/triage accepted