Improve PQL regex query performance

[chunshao90]

Describe This Problem

The following regex query is very slow.

sum (rate(http_requests_counter{url=~"/api/query|/plugin/prometheus/api/v1/query.*"}[3m])) 

But the performance is very good after splitting the query.

sum (rate(http_requests_counter{url="/plugin/prometheus/api/v1/query"}[3m])) 
sum (rate(http_requests_counter{url="/api/query"}[3m])) 

Proposal

Improve PQL regex query performance.

Additional Context

No response

[jiacai2050]

For simple case, I think we can add a datafusion OptimizerRule to convert regex filter to equal filter. Eg:

select * from demo where name ~ '1|2';
select * from demo where name in ('1', '2');

Related changes:
https://github.com/CeresDB/ceresdb/blob/17b40c63b50e7ea50b22be651f4a8c07a48dbc00/query_engine/src/context.rs#L89

[jiacai2050]

https://crates.io/crates/regex-syntax

This crate allow us to program based on regex syntax.

[tanruixiang]

apache/datafusion#6487 partially solved this issue.

[tanruixiang]

apache/datafusion#7185 Now we can support string which include _.
And #1128 make the PQL regex expression optimizeable.