1

groupwin

I use the meaning in esper:

This view groups events into sub-views by the value returned by the specified expression or the combination of values returned by a list of expressions.

I think it is that you have the ability to operate by group,not stream(the group by is used to control how aggregations are grouped.)

unaligned window

In google's dataflow ,unaligned windows means:

By unaligned windows, we mean windows which do not span the entirety of a data source, but instead only a subset of it, such as per-user windows.

And does these mean the same thing? Set<event> XXXX = eventStream.GroupByKey(XXX);

Community
  • 1
  • 1
陶加涛
  • 11
  • 4

1 Answers1

0

Grouped windows: there is no relationship to aggregations and aggregations are operating only by the group-by clause if there is one. Grouped window for Esper is a way to define the subsets of events that are considered. That is useful various use cases, for example match-recognize pattern matching over a subset of events or just running some function over some subset of events or comparing previous events in the same group with "prev".

In general ability to "operate per-group" in Esper is done by contexts ie. context partition per key.

user3613754
  • 816
  • 1
  • 5
  • 5
  • Thanks very much for your answer. http://stackoverflow.com/questions/41498805/how-to-describe-this-complex-scene-by-cep In context,I can not do like that:EventA.context.paritionA.FieldB == EventB.context.paritionB.FieldE. I can not do the correlation between two context's patititon. Any way,how do you think about that context (patition per key) and the unaligned window? Are they the same thing? – 陶加涛 Jan 08 '17 at 07:37
  • Contexts and context partitions actually come from the book Event Processing in Action (see the docs). So the book would probably answer this. I don't know Google Dataflow. Each context partition is independent of others and can have its own lifetime. So going across context partitions is not a thing. Context partitions can be anything aligned or unaligned and they are not windows because they don't necessarily retain events. – user3613754 Jan 09 '17 at 23:22
  • Thanks you very much.it seems that I have to key by the stream before I pass it to the cep engine. By the way,groupwin() group seems OK,but it seems cannot be used in pattern – 陶加涛 Jan 10 '17 at 01:14