v2.0: Reclaims more old accounts in clean (backport of #4044)

[mergify]

Problem

In clean_accounts(), if there are old storages and old accounts beyond what is expected, we do not always reclaim them. This is because when scanning the candidates during clean, if the account's latest slot in the slot list is not in the uncleaned_roots list, we don't collect reclaims.

This can happen if there are candidates from old/ancient slots, where their newest entry is in a slot that is not being visited by clean. We would find multiple entries for this candidate in the index, yet determine not to collect reclaims.

Summary of Changes

Collect reclaims on candidates that have slot lists greater than one entry, and if they are all cleanable (i.e. rooted).

Justification to Backport

Without this change, nodes running with --disable-accounts-disk-index will collect excessive storages, because startup index generation does not correctly mark duplicate pubkeys when the disk index is disabled. But now clean does go and inspect those old storages, yet it won't reclaim the old versions of the accounts because the newest version of the account is not in the uncleaned_roots list. And because we don't reclaim these old versions, we cannot purge old storages that are all dead.

(Historically the uncleaned_roots were all the most recent slots, so it made sense to check for the newest versions there. Now, we clean really old slots too, which means the newest version of accounts in these old slots may be elsewhere (i.e. not just in uncleaned_roots))

---

This is an automatic backport of pull request #4044 done by [Mergify](https://mergify.com).

[mergify]

Cherry-pick of 3d43824 has failed:

On branch mergify/bp/v2.0/pr-4044
Your branch is up to date with 'origin/v2.0'.

You are currently cherry-picking commit 3d43824154.
  (fix conflicts and run "git cherry-pick --continue")
  (use "git cherry-pick --skip" to skip this patch)
  (use "git cherry-pick --abort" to cancel the cherry-pick operation)

Changes to be committed:
	modified:   accounts-db/src/accounts_index.rs

Unmerged paths:
  (use "git add/rm <file>..." as appropriate to mark resolution)
	both modified:   accounts-db/src/accounts_db.rs
	deleted by us:   accounts-db/src/accounts_db/tests.rs

To fix up this pull request, you can check it out locally. See documentation: https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/reviewing-changes-in-pull-requests/checking-out-pull-requests-locally

[HaoranYi]

lgtm.
One question, are those test code necessary for backporting?

[brooksprumo]

One question, are those test code necessary for backporting?

The tests aren't necessary for the logic changes, but seems like a good idea to also keep the tests for backports in general, right?

[HaoranYi]

The core logic change looks good to me.

The test is only partially test what we intended to test. There is a follow up PR which improve these tests.

#4147

Since we are backporting the new tests here, maybe we want to update the tests here to use 4147 in this back port PR?

[brooksprumo]

Since we are backporting the new tests here, maybe we want to update the tests here to use 4147 in this back port PR?

Yeah, that's a good point. I'm not sure if it's best to pull the test fixes into this PR, or to not and then backport the fixes on their own.

[HaoranYi]

yeah.

I think the goal for include these tests in the backport is to make sure we are testing the clean for old slots even if the new slot is not added. For the test to exercise this, we really need 4147.

However, creating another backport for just the sake of these tests doesn't seem to worth it,
My suggestion is just to cherry-pick 4147 on top of this backport...

[brooksprumo]

I'm going to merge this backport as-is, and then backport the tests on their own. This keeps the PRs on master/v2.1/v2.0 the same, which is useful for any debug/ops-related reverts/bisects/etc.