[
https://issues.apache.org/jira/browse/KAFKA-7911?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16764351#comment-16764351
]
Matthias J. Sax commented on KAFKA-7911:
----------------------------------------
I did not think about this in detail yet. I guess there are some open design
question. Setting the timezone in `windowedBy` would imply that one might need
to specify it multiple times if there are multiple windowed operations.
For the second proposal, it seems you set it for a specific input stream. This
might allow for some flexibility, however, I am wonder on the impact of a
stream-stream join if both streams have different timezones set?
Also, could we assume that all input records use the same timestamp encoding or
not? If yes, we could just use config to set the timezone.
Looking at the it from 10,000ft, is seems that the timezone is a per-stream
information. Thus, does it make sense to set it on an operator?
As pointed out, this Jira requires a KIP, thus, it would be good to think about
those question (and maybe more), and write a concrete proposal (including
rejected alternatives) so the community can discuss. I can imagine that this
ticket does not describe all use cases and corner cases yet.
> Add Timezone Support for Windowed Aggregations
> ----------------------------------------------
>
> Key: KAFKA-7911
> URL: https://issues.apache.org/jira/browse/KAFKA-7911
> Project: Kafka
> Issue Type: New Feature
> Components: streams
> Reporter: Matthias J. Sax
> Priority: Minor
> Labels: needs-kip
>
> Currently, Kafka Streams only support UTC timestamps. The impact is, that
> `TimeWindows` are based on UTC time only. This is problematic for 24h
> windows, because windows are build aligned to UTC-days, but not your local
> time zone.
> While it's possible to "shift" timestamps as a workaround, it would be better
> to allow native timezone support.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
