Labelling this PR as size/S

GCE e2e test build/test passed for commit f674109.

• Build Log
• Test Artifacts
• Internal Jenkins Results

👍 Looks good to me!

LGTM

Did the mergebot die or something? :)

@paralin it isn't merging anything because the scalability tests are not currently passing. See the "Google Internal e2e" tab on http://submit-queue.k8s.io/

This PR prevents us from using DaemonSet for the things we want to use it for.

I agree. This was a quick fix for pathological behavior because we don't have he correct configuration knobs but this will be reverted before DaemonSet leaves alpha. Is 2 of #17297 (comment) preferable as a short term fix?

@bgrant0607 as Mike says it didn't work properly anyway, so this is a good fix to make things at least work without errors as an interim until there's a better solution.

@mikedanese If this gets reverted before DaemonSet graduates to Beta, I'm fine with that. Please add that to #15310.

For self-hosted clusters, (2) is what we want. The master node should be configured like the other nodes.

For clusters where the master isn't on the same network, such as GKE, I agree it doesn't make as much sense.

Perhaps we can leverage #17151 / #17190 to control where daemon pods land by default without needing the full node admission control matrix described in #3885, since this really is more about node selection than schedulability.