optimizer may make bad decisions due to pseudo stats during init stats

[xuyifangreeneyes]

Bug Report

Please answer these questions before submitting your issue. Thanks!

1. Minimal reproduce step (Required)

Here is a small example to demonstrate the issue.

I add time.Sleep(20 * time.Minute) into (*Handle).InitStats to simulate slow init stats.

mysql> create table t1(a int, b int, index idx(a));
Query OK, 0 rows affected (0.02 sec)

-- insert data into t1

mysql> analyze table t1;
Query OK, 0 rows affected, 1 warning (0.05 sec)

Then restart the tidb-server.

mysql> explain select * from t1 where a > 1;
+-------------------------+----------+-----------+---------------+--------------------------------+
| id                      | estRows  | task      | access object | operator info                  |
+-------------------------+----------+-----------+---------------+--------------------------------+
| TableReader_7           | 3333.33  | root      |               | data:Selection_6               |
| └─Selection_6           | 3333.33  | cop[tikv] |               | gt(test.t1.a, 1)               |
|   └─TableFullScan_5     | 10000.00 | cop[tikv] | table:t1      | keep order:false, stats:pseudo |
+-------------------------+----------+-----------+---------------+--------------------------------+
3 rows in set (0.00 sec)

mysql> show stats_meta;
Empty set (0.00 sec)

mysql> show stats_histograms;
Empty set (0.00 sec)

2. What did you expect to see? (Required)

We hope the optimizer avoids using pseudo stats to make decisions. For example, tidb can choose to wait for init stats to finish before providing service.

3. What did you see instead (Required)

The optimizer uses pseudo stats during init stats, which may lead to bad plans.

4. What is your TiDB version? (Required)

master

[winoros]

The behavior affects a lot. But we only want to solve problem with TiDB no less than 7.1-LTS

[winoros]

We decide to port the config setting to v6.5.
But you need to remember that v6.5's statistics initialization can be pretty long. If you open the setting, TiDB may need to wait for a long time between its startup and the time it's ready to provide service.