After supporting insert-only transactional hive full acid tables #19518, #19419, this PR support transactional hive full acid tables.
Support hive3 transactional hive full acid tables.
Hive2 transactional hive full acid tables need to run major compactions.
this pr impacts tpch q16 Agg strategy, but no performance issue
this pr improves tpcds sf100
before:
cold 141 sec
hot 133 sec
after:
code 137 sec
hot 128 sec
Issue Number: close#20669
RewriteInPredicateRule may cast InPredicate expr's two child to the same type, for example: where cast(age as char) in ('11'), the type of age is int, RewriteInPredicateRule will cast expr's two child type to int. As in the example above, child 0 will be such struct:
```
child 0: type: int
|--- child: type : char
|-- child: type : int
```
Due to the RewriteInPredicateRule cast the type of the expr to int, it will reanalyze stmt, but it will reset stmt first before reanalyze the stmt, and reset opt will change child 0 to such struct:
```
child: type : char
|-- child: type : int
```
It cause two child's type will be cast to varchar in func castAllToCompatibleType, the logic of RewriteInPredicateRule will be useless.
In 1.1-lts and 1.2-lts, such case " where cast(age as char) in ('11')" can't work well, because func castAllToCompatibleType will cast int to char but int can't cast to char(master can work well because func castAllToCompatibleType will cast int to varchar in such case).
```
MySQL [test]> select user_id from test_cast where cast(age as char) in ('45');
ERROR 1105 (HY000): errCode = 2, detailMessage = type not match, originType=INT, targeType=CHAR(*)
```
the formula used to compute ndv after filter implies that the new rowCount is smaller than the original rowCount. When we apply this formula to join, we should add branch if new row count is bigger than original row count.
when new row count is bigger, the ndv is not changed.
we have some prunning path logical in cascades framework. However it do not work as we expected. if we do prunning on one Group, then maybe we need to do thousands of times optimization on its parent without any success result. This PR remove these prunning provisionally. We will add prunning back when we re-design it.
The precision handling for division with DECIMALV3 is as follows (excluding cases where division increases precision):
(p1, s1) / (p2, s2) ----> (p1 + s2, s1)
However, due to precision loss in division, it is considered to increase the precision of the left operand:
(p1, s1) / (p2, s2) =====> (p1 + s2, s1 + s2) / (p2, s2) ----> (p1 + s2, s1)
However, the legacy optimizer repeats the analyze and substitute steps for an expression, which can result in the accumulation of precision:
(p1, s1) / (p2, s2) =====> (p1 + s2, s1 + s2) / (p2, s2) =====> (p1 + s2 + s2, s1 + s2 + s2) / (p2, s2)
To address this, the previous approach was to forcibly convert the left operand of DECIMALV3 calculations. This results in rewriting the expression as:
(p1, s1) / (p2, s2) =====> cast((p1, s1) as (p1 + s2, s1 + s2)) / (p2, s2)
Then, during the substitution step, a check is performed. If it is a cast expression, the expression modified by the cast is extracted:
cast((p1, s1) as (p1 + s2, s1 + s2)) =====> (p1, s1)
protected Expr substituteImpl(ExprSubstitutionMap smap, ExprSubstitutionMap disjunctsMap, Analyzer analyzer) {
if (isImplicitCast()) {
return getChild(0).substituteImpl(smap, disjunctsMap, analyzer);
}
This way, there won't be repeated analysis, preventing the continuous increase in precision. However, if the left expression is a constant (literal), theoretically, the precision would continue to increase. Unfortunately, the code that was removed in this PR (#19926) obscured this issue.
for (Expr child : children) {
if (child instanceof DecimalLiteral && child.getType().isDecimalV3()) {
((DecimalLiteral)child).tryToReduceType();
}
}
An attempt will be made to reduce the precision of literals in the expressions. However, this code snippet can cause such a bug.
mysql [test]>select cast(1 as DECIMALV3(16, 2)) / cast(3 as DECIMALV3(16, 2));
+-----------------------------------------------------------+
| CAST(1 AS DECIMALV3(16, 2)) / CAST(3 AS DECIMALV3(16, 2)) |
+-----------------------------------------------------------+
| 0.00 |
+-----------------------------------------------------------+
1.00 / 3.00, due to reduced precision, becomes 1 / 3.
<--Describe your changes.-->
Fix some bugs of orc lazy materialization(#18615)
- Fix issue causing column size to continuously increase after `execute_conjuncts()` by calling `Block::erase_useless_column()`.
- Fix partition issues of orc lazy materialization.
- Fix lazy materialization will not be used when the predicate column is inconsistent with the orc file.
If a query hits a materialized view that has row storage enabled, but the row storage column is not present in the materialized view, it will result in a query crash. Therefore, it is necessary to include the row storage column when creating the materialized view, and serialize the row storage column during the execution of SchemaChange.
we should not plan any Filter or Project above CteAnchor, because there are project or filter under anchor sometimes.
and the whole plan can not translate to a valid plan for BE.
1.move PushdownFilterThroughCTEAnchor and PushdownProjectThroughCTEAnchor into PUSH_DOWN_FILTERS rule set
2.move PushdownFilterThroughProject before MergeProjectPostProcessor
Support user-defined variables.
After this PR, we can use `set @a = xx` to define a user variable and use it in the query like `select @a`.
the changes of this PR:
1. Support the grammar for `set user variable` in the parser.
2. Add the `userVars` in `VariableMgr` to store the user-defined variables.
3. For the `set @a = xx`, we will store the variable name and its value in the `userVars` in `VariableMgr`.
4. For the `select @a`, we will get the value for the variable name in `userVars`.
* [fix](load) in strict mode, return error for load and insert if datatype convert fails
Revert "[fix](MySQL) the way Doris handles boolean type is consistent with MySQL (#19416)"
This reverts commit 68eb420cabe5b26b09d6d4a2724ae12699bdee87.
Since it changed other behaviours, e.g. in strict mode insert into t_int values ("a"),
it will result 0 is inserted into table, but it should return error instead.
* fix be ut
* fix regression tests
1. update all date related functions' signatures order.
1.1. if return value need to be compute with time info, args with datetimev2 at the top of the list, followed by datev2, datetime and date
1.2. if return value need to be compute with only date info, args with datev2 at the top of list, followed by datetimev2, date and datetime
2. Priority for use datev2, if we must cast date to datev2 or datetime/datetimev2
This commit support a function allows return a field column in named struct column.
Since the function can return any type, this commit also supports ANY_STRUCT_TYPE
and ANY_ELEMENT_TYPE.
There should be 2 kinds of ScanNode:
OlapScanNode
ExternalScanNode
The Backends used for ExternalScanNode should be controlled by FederationBackendPolicy.
But currently, only FileScanNode is controlled by FederationBackendPolicy, other scan node such as MysqlScanNode,
JdbcScanNode will use Mix Backend even if we enable and prefer to use Compute Backend.
In this PR, I modified the hierarchy of ExternalScanNode, the new hierarchy is:
ScanNode
OlapScanNode
SchemaScanNode
ExternalScanNode
MetadataScanNode
DataGenScanNode
EsScanNode
OdbcScanNode
MysqlScanNode
JdbcScanNode
FileScanNode
FileLoadScanNode
FileQueryScanNode
MaxComputeScanNode
IcebergScanNode
TVFScanNode
HiveScanNode
HudiScanNode
And previously, the BackendPolicy is the member of FileScanNode, now I moved it to the ExternalScanNode.
So that all subtype ExternalScanNode can use BackendPolicy to choose Compute Backend to execute the query.
All all ExternalScanNode should implement the abstract method createScanRangeLocations().
For scan node like jdbc scan node/mysql scan node, the scan range locations will be selected randomly from
compute node(if preferred).
And for compute node selection. If all scan nodes are external scan nodes, and prefer_compute_node_for_external_table
is set to true, the BE for this query will only select compute nodes.
1. derive analytic node stats, add support for rank()
2. filter estimation stats derive updated. update row count of filter column.
3. use ColumnStatistics.orginal to replace ColumnStatistics.orginalNdv, where ColumnStatistics.orginal is the column statisics get from TableScan.
TPCDS 70 on tpcds_sf100 improved from 23sec to 2 sec
This pr has no performance downgrade on other tpcds queries and tpch queries.
