timestamp of "arrival", but it will
> > produce sorted output as you "order by rowTime". Hope it helps.
> >
> > Best,
> > Stefano
> >
> > -Original Message-
> > From: Xingcan Cui [mailto:xingc...@gmail.com]
> > Sent: Wednesday, Ap
: Xingcan Cui [mailto:xingc...@gmail.com]
> Sent: Wednesday, April 12, 2017 8:11 AM
> To: dev@flink.apache.org
> Subject: Re: Question about the process order in stream aggregate
>
> Hi everybody,
>
> thank you all for your help.
>
> @Fabian I also check the DataStream tha
Xingcan Cui [mailto:xingc...@gmail.com]
Sent: Wednesday, April 12, 2017 8:11 AM
To: dev@flink.apache.org
Subject: Re: Question about the process order in stream aggregate
Hi everybody,
thank you all for your help.
@Fabian I also check the DataStream that translated from the query and try to
figure out
;
> The order of records is changed because of the connection between source
> and first map function. Here, records are distributed round robin to
> increase the parallelism from 1 to n. The parallel instances of map might
> forward the records in different order to the ProcessFunction that
ent order to the ProcessFunction that computes the
aggregation.
Hope this helps,
Fabian
Von: Stefano Bortoli
Gesendet: Dienstag, 11. April 2017 14:10
An: dev@flink.apache.org
Betreff: RE: Question about the process order in stream aggregate
Hi Xingcan,
Are you using parallelism 1 for the test? pr
Hi Xingcan,
Are you using parallelism 1 for the test? procTime semantics deals with the
objects as they loaded in the operators. It could be the co-occuring
partitioned events (in the same MS time frame) are processed in parallel and
then the output is produced in different order.
I suggest y
