pick from #37235
the algorithm for computing stats for "expr1 and expr2" predicate is as
following:
1. compute output stats of expr1 based on input stats. the result stats
is denoted by leftStats
2. compute stats of expr2 based on leftStats after step1, leftStats
should be normalized to avoid abnormal cases, such as ndv > rowCount or
numNulls > rowCount
Issue Number: close #xxx
<!--Describe your changes.-->
## Proposed changes
Issue Number: close #xxx
<!--Describe your changes.-->
* When creating a backupJob with huge of tables in a database, it can
cause backupJob editlog size over 2GB and bdbje will throw exception
because of ByteBuffer overflow
## Proposed changes
Issue Number: close #xxx
<!--Describe your changes.-->
pick #36632
Most users only care about the size of **max_batch_interval**, but in
order to achieve an interval effect, they have to configure
**max_batch_rows** and **max_batch_size** according to the
characteristics of the data. By adjusting these two default values,
users do not need to worry about configuration in most scenarios.
## Proposed changes
Issue Number: close #xxx
<!--Describe your changes.-->
pick from master #37069
add a new session variable: use_max_length_of_varchar_in_ctas
In CTAS (Create Table As Select), if CHAR/VARCHAR columns do not
originate from the source table, whether to set the length of such a
column to MAX, which is 65533. The default is true.
cherry-pick #36828 to branch-2.1
The requirement for predicate pushdown through the window operator is
that the partition by slots of the window contains all slots in the
predicate. The original implementation of doris only allows predicate
pushdown with one slot. This PR relaxes this restriction and allows for
predicate pushdown with multiple slots. The same applies to the
predicate pushdown of the LogicalPartitionTopN operator. The following
sql is an example.
select
*
from
(
select
row_number() over(partition by id, value1 order by value1) as num,
id,
value1
from
push_down_multi_column_predicate_through_window_t ) t
where
abs(id + value1)<4
and num <= 2;
Co-authored-by: feiniaofeiafei <moailing@selectdb.com>
The Analyzer of NereidsPlanner use different rules to analyze normal plan and view, to prevent the plans in the views analyze multiple times, because some rules can not apply multiple times, say, decimal type coercion, if this rule applied multiple times, it will generate a wrong result.
But this design is trick. Normally, after process the LogicalView, the whole plan tree in the LogicalView should not contains unbound plan, but the current situation is not like this. And this problem block development of some rules, so I refactor it:
1. the Analyzer will not traverse the children of the LogicalView
2. After link the LogicalView to the outer plan tree, the whole plan tree of the LogicalView will not contains unbound plan
3. analyze view and table use the same rules, keep it simple
## Proposed changes
pick from #36782
support explain like:
explain delete from T where A=1
Issue Number: close #xxx
<!--Describe your changes.-->
(cherry picked from commit dc369cd13096dbb90700f7fbf8f35a9059d9906f)
## Proposed changes
Issue Number: close #xxx
<!--Describe your changes.-->
