[Enhancement] statistics error lead to OOM when multi-table join scenario

[yx-keith]

Search before asking

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

Description

In multi-table join scenario, The result of intermediate join will be used as the input of the subsequent joins.
During the interval between updating statistics, when we update data, statistics is not collected in time, If we run a multi-table join query within this time interval, the optimizer may choose a poor strategy because there is no correct statistics.

for example:
select * from example_tbl t1 join example_tbl02 t2 on t1.city=t2.city and t1.city="成都" join example_tbl03 t3 on t1.city=t3.city;

this is plan：

in this case, When the last statistical information was collected, there were no rows containing '成都', but before the next statistics collection, rows containning '成都' was inserted， wrong statistics lead to poor execution plan:
example_tbl02 will be broadcasted to other node to join, but the actual situation is example_tbl02 has many many rows contains '成都'，which may cause OOM during broadcast distribution.

Solution

654ead9

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