Ordering of events in structured streaming window aggregates in append mode

It's not in the architecture (yet). Many posts here on SO allay that notion.

As cricket_007 states in numerous posts, you can best leave the sorting to the down-stream system(s) in general. It is also more flexible that way, the whole notion of the RDBMS is that fixing the data sort order is less valid - clustering aside.

If you look at this use case https://mapr.com/blog/real-time-analysis-popular-uber-locations-spark-structured-streaming-machine-learning-kafka-and-mapr-db/ then you see that sorting plays no role. That said, I see many requests, but many can achieve the goal without sorting.

Single Partition Topics are an outcome for lower volumes, providing the sort order is set by the 'producer' and that order is acceptable. Also, you can consider KSQL and writing to a single partition KAFKA Topic to read in from subsequently or KAFKA Streams with Java, Scala.

I think the issue is that as from a post I saw a few years ago:

The basic tenet of structured streaming is that a query should return the same answer in streaming or batch mode. We support sorting in complete mode because we have all the data and can sort it correctly and return the full answer. In update or append mode, sorting would only return a correct answer if we could promise that records that sort lower are going to arrive later (and we can't). Therefore, it is disallowed.