[
https://issues.apache.org/jira/browse/FLINK-13503?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16897004#comment-16897004
]
Jing Zhang edited comment on FLINK-13503 at 7/31/19 10:11 AM:
--------------------------------------------------------------
After discuss with [~jark] [~lzljs3620320] [~lic], we got the following
conclusion:
1. LookupJoin does not support JoinCondition like 'L.a is not distinct from
R.a' and other similar ( 'L.a = R.a or (L.a is null And R.a is null)'), which
leads to filterNulls contains false in 1.9 version. Would be fixed in 1.10
version (https://issues.apache.org/jira/browse/FLINK-13509).
2. Add contract in `LookupableTableSource` to specify the behavior when
lookupKeys contains null, it behaviors like point 2 in the jira content
3. Update `JdbcLookUpFunction` based on ad.2.
Correct me if the conclusion is wrong.
was (Author: jinyu.zj):
After discuss with [~jark] [~lzljs3620320] [~lic], we got the following
conclusion:
1. LookupJoin does not support JoinCondition like 'L.a is not distinct from
R.a' and other similar ( 'L.a = R.a or (L.a is null And R.a is null)'), which
leads to filterNulls contains false in 1.9 version. Would be fixed in 1.10
version.
2. Add contract in `LookupableTableSource` to specify the behavior when
lookupKeys contains null, it behaviors like point 2 in the jira content
3. Update `JdbcLookUpFunction` based on ad.2.
Correct me if the conclusion is wrong.
> Add contract in `LookupableTableSource` to specify the behavior when
> lookupKeys contains null
> ---------------------------------------------------------------------------------------------
>
> Key: FLINK-13503
> URL: https://issues.apache.org/jira/browse/FLINK-13503
> Project: Flink
> Issue Type: Task
> Components: Connectors / JDBC, Table SQL / API
> Affects Versions: 1.9.0, 1.10
> Reporter: Jing Zhang
> Assignee: Jing Zhang
> Priority: Minor
>
> I think we should add contract in `LookupableTableSource` to specify expected
> behavior when the lookupKeys contains null value.
> For example, one input record of eval method is (null,1) which means to look
> up data in (a,b) columns which key satisfy the requirement. there are at
> least three possibility here.
> * to ignore null value, that is, in the above example, only looks `b = 1`
> * to lookup `is value`, that is, in the above example, only looks `a is
> null and b = 1`
> * to return empty records, that is, in the above example, only looks `a =
> null and b = 1`
> In fact, there are different behavior in current code.
> For example, in Jdbc connector,
> The query template in `JdbcLookUpFunction` like:
> SELECT c, d, e, f from T where a = ? and b = ?
> If pass (null, 1) to `eval` method, it will generate the following query:
> SELECT c, d, e, f from T where a = null and b = 1
> Which always outputs empty records.
> BTW, Is this behavior reasonable?
> and the `InMemoryLookupableTableSource` behaviors like point 2 in the above
> list.
> some private connector in Blink behaviors like point 1
--
This message was sent by Atlassian JIRA
(v7.6.14#76016)
