Fork me on GitHub

Is there any documentation on how :reduce tasks work? Is there any difference between a terminal :reduce task and an terminal (i.e., :output) task with plugin :onyx.peer.function/function?


I have a workaround for above watermark triggering issue. I can explain it and maybe you can tell if that is a good idea or not. I know the incoming data are chronological. I want fixed windows of these data. (I also know that there are no empty windows). Watermark triggering does not work in my specific case, I know now. What I do to get the same behaviour, though, is the following: In an upstream task I emit an additional segment for each processed segment, where I include just a special keyword and a timestamp by <the same time of the original segment minus the window range> to make sure it will end up in the previous window. With a punctuation trigger I check if an incoming segment has the special keyword, and if yes, I trigger. Are there any ways in which this can break or is a bad idea otherwise? E.g., does this still work if you scale up the number of peers?


This approach will have problems if you scale up to more peers and don’t emit a segment that will reach each peer.


Which can be hard to do if you have a grouping-fn, and extremely difficult if not because the peer it’ll be sent to is randomised then.