Skip to content

[Enhancement] unbalanced resouce overhead between several BEs when loading #28564

@Hastyshell

Description

@Hastyshell

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.

image

CPU
image
image
image

Mem
image
image
image

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

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions