This change is 

LGTM

@Amey-D Thanks for the fix.

I use vim to open a file, and type update_disabled. Then use this file as metadata. It won't work.

However, echo -n update_disabled does work.

I need to figure out what is added by vim. :/

OK. Vim adds EOL at the end of each line automatically http://stackoverflow.com/questions/16222530/why-do-i-need-vim-in-binary-mode-for-noeol-to-work/16224292#16224292.

If I generate the file with echo -n update_disabled, it will work.
If I open the same file with vim, and save it, it won't work anymore.

Anyhow, at least it's clear that this PR will work! :)

However, I still think we should fix GCI if possible because this is so fragile and confusing. :/

The title of the PR should be "trailing newline" instead of whitespace :-)
Please also add comments to prevent regression since the logic is a bit fragile.

+1 to make the consumer of this metadata more robust. It'd also help if we surface the information/configuration. Could we add something to our e2e test framework to collect such information or add a test for GCI?

Agreed with @Random-Liu and @yujuhong the proper and more robust fix should at GCI side, not here. But I am ok to take this for short-term fix for running GCI on GCE.

Offline discussion with @Amey-D:

• We are going to fix this in GCI side for long term. @Amey-D is going to file a separate internal bug to track this. The fix in GCI is required for GCI is running outside GKE anyway.
• The current fix as the short term solution should be cherrypicked into both 1.4 and 1.5 release.

cc/ @kubernetes/goog-node @kubernetes/goog-image @saad-ali for 1.5 @jessfraz for 1.4

@k8s-bot test this [submit-queue is verifying that this PR is safe to merge]

thanks!

Automatic merge from submit-queue

cherry-picked in #38781

Cherry-picked to 1.5 in #40184