Fix three bugs:
1. Hudi slice maybe has log files only, so `new Path(filePath)` will throw errors.
2. Hive column names are lowercase only, so match column names in ignore-case-mode.
3. Compatible with [Spark Datasource Configs](https://hudi.apache.org/docs/configurations/#Read-Options), so users can add `hoodie.datasource.merge.type=skip_merge` in catalog properties to skip merge logs files.
Problem:
When inferring predicate，we lost cast of source expressions and some datatype derivation.
Example:
a = b and cast(a as targetType) = constant
(cast(a as targetType) = constant ) this expression is define as source expression.
we expect getting cast(b as targetType) = constant instead of b = constant
Reason:
When inferring predicate, we will compare original type of a and b. if they can be cast
without precision lost, a new predicate would be created. But created predicate forgot
to cast to target type
Solved:
Add cast to target type, and open make other datatype valid also.
before the lambda function Expr not implement toSqlImpl() function.
so it's call parent function, which is not suit for lambda function.
and will be have error when create view.
**Support sql cache for hms catalog. Legacy planner and Nereids planner are all supported.
Not support partition cache now, not support federated query now.**
Add 2 metrics in jdbc scan node profile:
- `CallJniNextTime`: call get next from jdbc result set
- `ConvertBatchTime`: call convert jobject to columm block
Also fix a potential concurrency issue when init jdbc connection cache pool
The origin virtual number is Math.max(Math.min(512 / backends.size(), 32), 2);, which is too small,
causing uneven cache distribution when enabling file cache.
This pr mainly has two changes:
1. add some merge processes about partition events
2. add a ut for `MetastoreEventFactory`. First add some mock classes (`MockCatalog`/`MockDatabase` ...) to simulate the real hms catalog/databases/tables/partitions, then create a event producer which can produce every kinds of `MetastoreEvent`s randomly. Use two catalogs for test, one is named `testCatalog` and the other is the `validateCatalog`, use event producer to produce many events and let `validateCatalog` to handle all of the events, but `testCatalog` just handles the events which have been merged by `MetastoreEventFactory`, check if the `validateCatalog` is equals to `testCatalog`.
If there 3 above fe nodes,
the following opeartions will cause all FE nodes down.
DROP USER revoke_test_user
DROP ROLE revoke_test_role
DROP DATABASE IF EXISTS revoke_test_db
CREATE DATABASE revoke_test_db
CREATE ROLE revoke_test_role
CREATE USER revoke_test_user IDENTIFIED BY 'revoke_test_pwd'
GRANT SELECT_PRIV ON revoke_test_db.* TO ROLE 'revoke_test_role'
GRANT 'revoke_test_role' TO revoke_test_user
SHOW GRANTS FOR revoke_test_user
REVOKE 'revoke_test_role' from revoke_test_user
SHOW GRANTS FOR revoke_test_user
DROP USER revoke_test_user
DROP ROLE revoke_test_role
DROP DATABASE revoke_test_db
mv in select materialized_view should disable show table,
because Nereids planner can output the string such as
slot#[0] in toSql() of SlotRef. Note this is only a
temporary solution, will use an expression translator later
select rank() over (partition by A, B) as r, sum(x) over(A, C) as s from T;
A is a common partition key for all windowExpressions, that is A is intersection of {A,B} and {A, C}
we could push filter A=1 through this window, since A is a common Partition key:
select * from (select a, row_number() over (partition by a) from win) T where a=1;
origin plan:
----filter((T.a = 1))
----------PhysicalWindow
------------PhysicalQuickSort
--------------PhysicalProject
------------------PhysicalOlapScan[win]
transformed to
----PhysicalWindow
------PhysicalQuickSort
--------PhysicalProject
----------filter((T.a = 1))
------------PhysicalOlapScan[win]
But C=1 can not be pushed through window.
Issue Number: close #xxx
ShowTableStmtTest.testNoDb and DropDbStmtTest.testNoPriv are unstable cases，error msg is:
java.lang.Exception: Unexpected exception, expected<org.apache.doris.common.AnalysisException> but was<mockit.internal.expectations.invocation.MissingInvocation>
we can not know what is missing ,and this issue cannot be reproduced locally,so add some log
When be load reblancer choose candidate tablets, it will try moving tablets from high load backends to low backend backends. If the higher HIGH BE has no available slot num, it should try next HIGH BE.
Sometimes FE replica's version is unreliable. FE's replica may bigger than BE's real version. Need check if BE missing version (last failed version > 0).
void throw analysis ex for unsupported type to make ext table goes nereids.
this will improve the nereids' availability for external table if unsupported type is in the basic table schema but not referenced in the real sql.
tested in external table env.
Consider the following case:
select pu.pk_ct_pu as id
from fms_rd_nc65_zb.NC65P.CT_PU pu
left join fms_rd_nc65_zb.NC65P.PUB_WF_INSTANCE pwi
on pu.pk_ct_pu = pwi.billid
and pu.vtrantypecode=pwi.billtype
left join fms_rd_nc65_zb.NC65P.SM_USER su
on pu.creator = su.cuserid
where pu.pk_ct_pu='1001A110000000K8XPVN';
PUB_WF_INSTANCE table has a BLOB type column and currently it will throw analysis exception and fallback to old optimizer, although this column is not referenced in the real sql. The old optimizer doesn't have the outer join -> inner join rule and the "pu.pk_ct_pu='1001A110000000K8XPVN'; " is not pushed down and the performance will drop down. After the pr, we add the unsupported type instead of throw exception directly, it will decide the unused case and can continue goes nereids and use all advanced optimization for the sql.
