[07:56:39] <pfischer>	 o/
[07:56:45] <pfischer>	 ottomata: thanks!
[09:57:26] <pfischer>	 lunch
[10:01:23] <dcausse>	 ebernhardson: 11% is the average on every window or the total of duplicated events over the time period or your whole dataset ?
[10:15:24] <dcausse>	 lunch
[10:23:38] <gehel>	 lunch 2
[12:16:34] <dcausse>	 errand
[13:24:26] <ebernhardson>	 dcausse: 11% is total, basically the number of unique (time bucket, wikiid, page_id) tuples over the dataset vs the total dataset
[13:24:57] <dcausse>	 ebernhardson: thanks, so perhaps a single big window could drive most of the dups?
[13:25:30] <ebernhardson>	 dcausse: seems like it, yea. we mostly have to decide how long we are willing to let events wait around
[14:44:51] <gehel>	 pfischer: didn't you also have a presentation ready?
[16:20:57] <pfischer>	 gehel: Yes I did. I guess I filled in the form too late. Anyway, those slides are still there. Maybe another time.
[16:27:07] * ebernhardson struggles to understand what flinks is going to do if there are sequential windows in a processing pipeline
[16:29:48] <ebernhardson>	 i suppose i'm wonder what the time assigned to the output of the first aggregation is. first deduped event, last? or the time when the window closes?
[16:44:15] * ebernhardson is just bad at reading :P  Apparently its window end timestamp - 1.
[17:50:49] <dcausse>	 dinner
[18:48:37] <gehel>	 inflatador: fyi, we just repooled wdqs2021
[18:48:44] <gehel>	 s/2021/2012/