Andras Istvan Nagy created KYLIN-4246:
-----------------------------------------
Summary: Wrong results from real-time streaming when an optional
field is used as a dimension
Key: KYLIN-4246
URL: https://issues.apache.org/jira/browse/KYLIN-4246
Project: Kylin
Issue Type: Bug
Components: Real-time Streaming
Affects Versions: v3.0.0-alpha
Reporter: Andras Istvan Nagy
We run into an issue which seems to be related to the real-time streaming
receiver.
We have an optional field in the kafka messages, that is, in some cases it has
a value, in other cases it is missing from the JSON message. This field is
defined as a dimension and is used in the queries. We were assuming that when
this field is missing from the JSON message, its value will be interpreted as
null (as is the case with the Kylin batch engine).
The results for queries that include this field are correct for those segments
that have been rebuilt with the Kylin batch engine, but incorrect when they are
built by the streaming receiver.
E.g. in the example query below (simplified version of our actual queries),
optional_field is an optional field, and mandatory_field always has a value. In
our case, the below query should produce 0 as output, because we have no such
records where optional_field has a value (is not null), and
mandatory_field='X'. Still, for those segments that are coming from the
streaming receiver, we get non-0 values.
{{select count(*)}}
{{from movement_events}}
{{where mandatory_field='X'}}
{{and optional_field is not null;}}
Is this a known issue? Can we avoid this somehow, without changing the source
events?
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
