Hi, As of now, we know that kinesis does not support Upsert Stream for Sink. We are using TableAPI's SQL to generate a realtime stream, the query we are authoring generates changelog as both insert (+I) and updates (-U, +U). We observe the below error when we try to sync the upsert stream to the kinesis sink.

`doesn't support consuming update changes which is produced by node GroupAggregate`.

It will be great if the community can guide us on best practices for resolving the above issue. Thanks.

you could convert your query to a grouped result so that it’s not “update changes”--something like a top-n, or a tumbling window

suppose we are getting an event for movie watched and output expected is movie Id and its associated count on the continuous basis like if at T1 count for an id is 10 and T2 id is incremented by 1 then count 11 need to published at T2 that will be lead to an update. So, here tumbling is not exactly required. What should be done in this use case ..!!

the issue here is that flink outputs an “update” record, (+U) which Kinesis currently does not know how to consume. By wrapping these results in a window or something similar, we can effectively change it into an “insert” table which Kinesis knows how to interpret.