ARROW-17252: [R] Intermittent valgrind failure (explicit StopProducing approach)

[paleolimbot]

No description provided.

[paleolimbot]

@github-actions crossbow submit test-r-linux-valgrind

[github-actions]

https://issues.apache.org/jira/browse/ARROW-17252

[github-actions]

⚠️ Ticket has not been started in JIRA, please click 'Start Progress'.

[github-actions]

Revision: e0fab9b

Submitted crossbow builds: ursacomputing/crossbow @ actions-cc5998cdea

Task	Status
test-r-linux-valgrind

[pitrou]

This looks like a clean approach, but hopefully it will solve the issue as well :-)

[paleolimbot]

@github-actions crossbow submit test-r-linux-valgrind

[github-actions]

Revision: eb8194d

Submitted crossbow builds: ursacomputing/crossbow @ actions-60deab85dd

Task	Status
test-r-linux-valgrind

[pitrou]

@paleolimbot Did you favor another approach? It's not clear here :-)

[paleolimbot]

Thanks - I should have left a note. This approach never solved the issue! Reverting part of the PR that introduced the errors (#13773) and waiting for IO thread pool jobs to finish (#13779) both seem to work. I think we are going with the first option as a patch, although there is a PR in the query engine to support better cancelling of scans (#13782), after which the call to StopProducing() should work (if the leaks we're seeing are, in fact, a result of pending tasks on the IO thread pool, which I think they are). I envision #13706 as the PR that will fix this for good.