[Bug] The MetaObject lock design of fe would cause some problems with consistent meta when catalog do replay operation #6649
Description
Search before asking
- I had searched in the issues and found no similar issues.
Version
What's Wrong?
As the issue #6386 described, with the performance considered, we introduced table lock in PR #3775, use table lock to replace db lock, but it also make the problem which lock design of fe brings occurs frequently.
With taking account of performance, we operate or get some info from meta object just get the write or read lock of current object, but we may get dropped object due to lack of parent's meta lock. in normal case, we must ensure that the meta object we get should not be dropped.
What You Expected?
In order to keep performance and make meta consistent between followers and observers, we must check meta object whether has been dropped when we get the writeLock.
How to Reproduce?
make frequent stream load to one table and then drop table at the same time.
Anything Else?
No response
Are you willing to submit PR?
- Yes I am willing to submit a PR!
Code of Conduct
- I agree to follow this project's Code of Conduct