statistic: the default value of CMSketch is not initialized correctly when using Analyze

[qw4990]

Development Task

How to reproduce:

create table t (a int, key(a));
insert into t values (0);
insert into t values (0);
...
insert into t values (0); // insert 2048 rows

insert into t values (1);
insert into t values (2);
insert into t values (3);

analyze table t with 0 topn;
explain select * from t where a = 1;
explain select * from t where a = 2;
explain select * from t where a = 3;

The estimated rows for a = 1/2/3 is zero, which is not accurate.

This is a bad case for the heuristic rule of removing noise when estimating point get:

		noise := (c.count - uint64(c.table[i][j])) / (uint64(c.width) - 1)
		if uint64(c.table[i][j]) < noise {
			vals[i] = 0
		} else {
			vals[i] = c.table[i][j] - uint32(noise)
		}

[qw4990]

In this case, we should use the default value of this CMSketch, but it is zero. We should investigate more to know why the default value is zero in this case.

[qw4990]

After reviewing the code, I find that the default value of CMSKetch is not initialized correctly in Analyze while Fast Analyze handles it rightly.
So if we use Fast Analyze, you can see the accurate estRows:

... // insert data

mysql> set tidb_enable_fast_analyze=1;
Query OK, 0 rows affected (0.00 sec)

mysql> analyze table t with 0 topn;
Query OK, 0 rows affected (0.02 sec)

mysql> explain select * from t where a = 3;
+------------------------+---------+-----------+---------------------+-------------------------------+
| id                     | estRows | task      | access object       | operator info                 |
+------------------------+---------+-----------+---------------------+-------------------------------+
| IndexReader_6          | 1.00    | root      |                     | index:IndexRangeScan_5        |
| └─IndexRangeScan_5     | 1.00    | cop[tikv] | table:t, index:a(a) | range:[3,3], keep order:false |
+------------------------+---------+-----------+---------------------+-------------------------------+
2 rows in set (0.00 sec)

[qw4990]

I will fix this issue soon.