Aggregating results from multiple keyed window calculations per window Hi! I would appreciate any help with handling data skew in windowed calculations. I want to calculate aggregates such as

.keyBy(e => e.key).window(...)

. The

key

here is badly skewed leading to performance issues. To handle data skew, I was thinking of using the "salting" strategy explained in this article. So I would add a well-distributed key called, say,

salt

and first make aggregation

.keyBy(e => (e.key, salt(e)).window(...)

. Then I'd like to collect the results from each "salt" to produce a single result per

(key, window)

. I'm struggling to understand how to do this in Flink. I can add

.keyBy((key, window))

after the "salted" calculation to collect results, but it is difficult to produce only one result. I can fire aggregated result per salt as they come in, but this not desired in downstream. I was also thinking of adding a consecutive window like

.keyBy(key).window(# same window as above)

, but I'm a little confused how to ensure my windowing results from the first step end up in a correct window in the second step. Do I need to do some additional

assignTimestampsAndWatermarks

stuff here? Any help very much appreciated 🙏