flink-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Aljoscha Krettek <aljos...@apache.org>
Subject Re: How to preserve KeyedDataStream
Date Tue, 03 Nov 2015 14:06:43 GMT
Hi,
where are you storing the results of each window computation to? Maybe you could also store
it from inside a custom WindowFunction where you just count the elements and then store the
results.

On the other hand, adding a (1) field and doing a window reduce (à la WordCount) is going
to be way more efficient because we only have to keep one element per window (the current
reduced tuple) instead of all the tuples, as we have to for a fold or WindowFunction. If you
want you can also combine a reduce and WindowFunction:
WindowedStream.apply(ReduceFunction<T> preAggregator, WindowFunction<T, R, K, W>
function)

here, the ReduceFunction does the WordCount-like counting while in the WindowFunction you
get the final result and store it inside your model.

Let me know if you need more information.

Cheers,
Aljoscha
> On 03 Nov 2015, at 11:28, Martin Neumann <mneumann@sics.se> wrote:
> 
> Hej,
> 
> I want to do the following thing:
> 1. Split a Stream of incoming Logs by host address. 
> 2. For each Key, create time based windows
> 3. Count the number of items in the window
> 4. Feed it into a statistical model that is maintained for each host
> 
> Since I don't have fields to sum upon, I use a (window) fold function to count the number
of elements in the window. (Maybe there is a better way to do this, or it could be part of
the primitives)
> My problem is now that I get back a DataStream so the distribution by key is lost. Is
there a way to preserve the distribution by key? Currently I only store the count of element
in the windows so I cannot simple do byKey again.
> 
> I could fold into tuples that have the count and also contain the host address but that
feels clumsy.
> 
> Any hints are welcome.
> 
> 
> cheers Martin


Mime
View raw message