Thank you very much for your help @Retired_mod ! However, I am afraid that I don't fully understand the solution, sorry! 😞

I checked your code and compared with mine and I don't see the differences. On each data stream, there is a column named `created_at` which is the time event. It works perfectly when joining two streams, but the problem raises when I try to join three (or more) streams. I understand that internally spark creates a single global watermark from the two created_at columns in a single join operation according to the doc. However, it seems that using more JOINs fails.

I tried a workaround which is writeStream the first JOIN, and then readStream and do the third JOIN. In this case it works, but it is a bit messy. I just wonder if I am doing something wrong or is it the expected behaviour?

Thank you very much!