Fix segmentation faults with enabled keep_descriptor_pool_after_request

[simonberger]

Fixes #16894

It is tested on command execution only, where it fixes the issues of reported memory leaks, heap corruption and segmentation fault.

I could not run tests on my Mac because of #16944

[hakimio]

@haberman take a look?

[haberman]

Hmm, looks like there are some test failures unrelated to your change. Want to rebase and push again?

[simonberger]

Hmm, looks like there are some test failures unrelated to your change. Want to rebase and push again?

@haberman Sure, done

[MrMage]

Looks like only some internal "copybara" check failed. Is that a blocker, or can this be merged?

@bshaffer would it be possible for you to merge this in your function as an owner?

[haberman]

All PRs must be committed internally first, because the internal repo is the source of truth. If we tried to merge the PR directly in GitHub I think it might get reverted by the next CL that syncs to GitHub.

The Copybara failure should be temporary. Internally the CL is in @honglooker's court to review and resolve any issues with the internal CI.

[MrMage]

Thank you for the insights! Looks like this was now merged internally and synced by Copybara.

[simonberger]

I already asked in the related issue about if a backport to 25.x should be done what I personally do not really care about, but I just realized this fix currently would not do it in a potential 27.2 release.
Is this something you do internally or shall I create a backport fix for 27.x and possibly 25.x. Some more clear expectations for the different types of pull requests would be good. I read in the guide everything shall target main first which is unlike most other repositories I know. What shall follow after is a bit uncertain to me.

Edit:
Created backport to 27.x here #17232 in case it is not required/wanted just close it.