[Serve] Address incremental memory leak due to _PyObjScanner

[simon-mo]

Signed-off-by: simon-mo [email protected]

Why are these changes needed?

Ray user @AbhishekBose reported a memory leak in Ray Serve and I was able to pin it down to the the leak in the PyObjScanner object. Previously we had to deal with this in other places. But it seems like the fix was incomplete.

This PR apply the same fix to the other places where _PyObjScanner is involved. Someone from the Ray Serve team should follow up with a release test.

Related issue number

https://discuss.ray.io/t/dagdriver-memory-increasing-for-basic-dags/8758/1

Checks

• I've signed off every commit(by using the -s flag, i.e., git commit -s) in this PR.
• I've run scripts/format.sh to lint the changes in this PR.
• I've included any doc changes needed for https://docs.ray.io/en/master/.
• I've made sure the tests are passing. Note that there might be a few flaky tests, see the recent failures at https://flakey-tests.ray.io/
• Testing Strategy
  • Unit tests
  • Release tests
  • This PR is not tested :(

[simon-mo]

I have confirmed running user's reproduction:

• before: memory growth from 120 to 300 in 10 minutes
• after: memory stable around 144MB for 10 minutes.

[simon-mo]

@sihanwang41 should decide whether we make a patch release for this