This is an automated cherry-pick of #39621

Signed-off-by: ekexium eke@fastmail.com

What problem does this PR solve?

Issue Number: close #39611

Problem Summary:
The sequence of events that leads to the panic is:
membuffer.set -> enlarge -> onMemChange -> memory tracker decides to panic -> deferred releasing current staging buffer in AddRecord -> revert to last checkpoint

When reverting to last checkpoint, memdb iterates over the nodes backwards, here is why it panics:
The latest block of the arena is in an intermediate state: it is just allocated but no data has been written yet. So its length is 0. While during the iteration we assume the length cannot be 0.

What is changed and how it works?

Don't let memdb call the callback when the block is in intermediate state. Instead call it after any content is inserted after a new block is allocated.

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

Please refer to Release Notes Language Style Guide to write a quality release note.

None