perf: optimize count(*)

[waynexia]

I hereby agree to the terms of the GreptimeDB CLA.

Refer to a related PR or issue link (optional)

What's changed and what's your intention?

TL;DR: 5x

Before

MySQL [(none)]> select count(*) from phy;
+------------+
| COUNT(*)   |
+------------+
| 2582870657 |
+------------+
1 row in set (4 min 38.347 sec)

After

MySQL [(none)]> select count(*) from phy;
+------------+
| COUNT(*)   |
+------------+
| 2582870657 |
+------------+
1 row in set (50.387 sec)

This patch adds a new optimizer rule that converts count(*) to count(<TIME INDEX>). This optimization is based on the fact that our underlying storage engine scans faster on time index column than primary key column. Reading time index column does not need decoding phase like in primary key column. This rule is extended from the one from datafusion and overrides it.

This patch also changes some logic in range plan, making it able to handle aggr expr with alias.

Checklist

• I have written the necessary rustdoc comments.
• I have added the necessary unit tests and integration tests.
• This PR does not require documentation updates.

[codecov]

Codecov Report

Attention: Patch coverage is 73.52941% with 27 lines in your changes are missing coverage. Please review.

Project coverage is 85.27%. Comparing base (81f3007) to head (db5b55e).
Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3845      +/-   ##
==========================================
- Coverage   85.60%   85.27%   -0.33%     
==========================================
  Files         954      955       +1     
  Lines      163325   163426     +101     
==========================================
- Hits       139808   139369     -439     
- Misses      23517    24057     +540     

[fengjiachun]

LGTM

[evenyag]

LGTM

[killme2008]

This patch has some corner cases that don't handle properly, for example:

Create two tables:

create table "HelloWorld" (a string, b timestamp time index);

create table test (a string, b timestamp time index);

Insert some rows:

insert into "HelloWorld" values ("a", 1) ,("b", 2);

insert into test values ("c", 1) ;

1. First case: doesn't handle table names in case-sensitive:

mysql> select count(*) from “HelloWorld”;
ERROR 1815 (HY000): DataFusion error: No field named helloworld.b. Valid fields are “HelloWorld”.a, “HelloWorld”.b.

2. Second case: doesn't handle sub query with count(*):

mysql> select count(*) from (select count(*) from test where a = 'a');
ERROR 1815 (HY000): DataFusion error: No field named test.b. Valid fields are "COUNT(*)".

The above examples work before this patch. @waynexia @fengjiachun @evenyag