[Enhancement] Reduce the number of disk accesses while parsing a segment footer

[sduzh]

Search before asking

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

Version

61c9d11

What's Wrong?

In the current implementation, parsing a segment footer needs two disk accesses:
https://github.com/apache/incubator-doris/blob/61c9d11fdb72dccf94876ae3706e7ed492622807/be/src/olap/rowset/segment_v2/segment.cpp#L96
If there are a large number of segment files, this will hurt the performance significantly. I have written a demo program to reduce the number of I/Os by reading more data at once, and the query results improved from 47s to 31 seconds. However, since the size of the footer can vary over a wide range, it is difficult to decide how much data is appropriate to read at once.

What You Expected?

Try to parse segment footer with just one disk access.

How to Reproduce?

No response

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

[sduzh]

This is not a bug report, but I cannot find a suitable issue template to use.

[sduzh]

I will submit a PR as soon as possible, but since I can only submit the code during off-hours and have been busy lately, I can't give a specific deadline. If anyone is interested in this issue and has a good idea, feel free to submit a PR.