[
https://issues.apache.org/jira/browse/KAFKA-10277?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17166561#comment-17166561
]
Matthias J. Sax commented on KAFKA-10277:
-----------------------------------------
{quote}Could I pick this up?
{quote}
Absolutely.
{quote} it also looks like there's no check to enforce that the {{keySelector}}
returns a non-null join key.
{quote}
Well, there is (at least partly). Exactly the line you highlighted. If the
returned `mappedKey` is `null`, we don't call `getValueOrNull()`, ie, if the
key is `null` we don't do the table lookup. And I think we need to keep this
check. – But you are right that we would still emit the input record if we use
a leftJoin what sound incorrect.
Also note, that the Processor is used to stream-table and stream-globalTable
join and thus we need to ensure that it works correctly for both cases. For the
stream-table join, the provided `keyMapper` is just the key-identify function
`(k,v) -> k`.
But I would defer the detailed discussion for the PR review?
> Relax non-null key requirement for KStream-GlobalKTable joins
> -------------------------------------------------------------
>
> Key: KAFKA-10277
> URL: https://issues.apache.org/jira/browse/KAFKA-10277
> Project: Kafka
> Issue Type: Improvement
> Components: streams
> Reporter: Matthias J. Sax
> Assignee: Joel Wee
> Priority: Major
> Labels: beginner, newbie
>
> In general, Kafka Streams requires that key and value are non-null for
> aggregations and join input records. While this requirement is reasonable in
> general, for KStream-GlobalKTable joins it's questionable for the stream
> input record key. The join is based on a `keySelector` and it seems to be
> sufficient to require that the `keySelector` returns a not-null join-key for
> the stream record.
> We should consider to relax the non-null key requirement for stream-side
> input records.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
