PageDirectory: need do gc twice in restore.

[jiaqizho]

What problem does this PR solve?

Issue Number: ref #4957

Problem Summary:

• MVCC may not clean up deleted entry if that entry be refence by a deleted ref.
• For example: [p1, ref 2 -> 1, d 1, d 2]
• First time call gcInMemEntries():
  • P1 entry 0 being_ref_count is not zero, won't be cleaned.
  • R2 entry 0 being_ref_count is zero, deref the P 1, make it being_ref_count is 0. Also R2 will be clean up.
• Second time call gcInMemEntries():
  • P 1 being_ref_count is 0, it will be clean.
• If we can't clean up all of the deleted entries in restore. Then BlobStore may not restore success.

What is changed and how it works?

• call gcInMemEntries() twice to clean up all of deleted entries.

Check List

Tests

• Unit test
• Integration test
• Manual test (add detailed scripts or steps below)
• No code

Side effects

• Performance regression: Consumes more CPU
• Performance regression: Consumes more Memory
• Breaking backward compatibility

Documentation

• Affects user behaviors
• Contains syntax changes
• Contains variable changes
• Contains experimental features
• Changes MySQL compatibility

Release note

None

[ti-chi-bot]

[REVIEW NOTIFICATION]

This pull request has not been approved.

To complete the pull request process, please ask the reviewers in the list to review by filling /cc @reviewer in the comment.
After your PR has acquired the required number of LGTMs, you can assign this pull request to the committer in the list by filling /assign @committer in the comment to help you merge this pull request.

The full list of commands accepted by this bot can be found here.

Reviewer can indicate their review by submitting an approval review.
Reviewer can cancel approval by submitting a request changes review.

[JaySon-Huang]

Also add it to PageDirectoryFactory::createFromEdit

[jiaqizho]

done

[JaySon-Huang]

I mean in these lines, we should also call gcInMemEntries twice

tiflash/dbms/src/Storages/Page/V3/PageDirectoryFactory.cpp

Lines 87 to 100 in 62e7c14

	PageDirectoryPtr PageDirectoryFactory::createFromEdit(String storage_name, FileProviderPtr & file_provider, PSDiskDelegatorPtr & delegator, const PageEntriesEdit & edit)
	{
	auto [wal, reader] = WALStore::create(storage_name, file_provider, delegator, WALStore::Config());
	(void)reader;
	PageDirectoryPtr dir = std::make_unique<PageDirectory>(std::move(storage_name), std::move(wal));
	loadEdit(dir, edit);
	// Reset the `sequence` to the maximum of persisted.
	dir->sequence = max_applied_ver.sequence;
	// After restoring from the disk, we need cleanup all invalid entries in memory, or it will
	// try to run GC again on some entries that are already marked as invalid in BlobStore.
	dir->gcInMemEntries();
	if (blob_stats)

[JaySon-Huang]

Do not need this change because we have cleaned entries after decreasing the ref count in VersionedPageEntries::derefAndClean.

tiflash/dbms/src/Storages/Page/V3/PageDirectory.cpp

Lines 599 to 603 in 64a747a

	iter->second.being_ref_count -= deref_count;
	// Clean outdated entries after decreased the ref-counter
	// set `normal_entries_to_deref` to be nullptr to ignore cleaning ref-var-entries
	return cleanOutdatedEntries(lowest_seq, /*normal_entries_to_deref*/ nullptr, entries_removed, page_lock);

The test cases will be merged into #4987