[Enhancement] unbalanced resouce overhead between several BEs when loading

[TangSiyang2001]

Search before asking

• I had searched in the issues and found no similar issues.

Description

env: 1FE + 3BE 84c 130G for load 3 * 500G high performance SSD
scenario: insert 4800000000 lines data into TPCH lineitem table.

CPU

Mem

Solution

Actually, the node with heavy resource overhead is suffering from back pressure of the add batch RPC, due to the low responce from the target BE which is adding batch data.

Finally, we found that the read lock for checking broken tablet (writers) maybe inreasonable and add to the overhead, making it very slow though with a very light behavior.

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