Flink message & state lifecycle.

2016-01-13 Thread Andrew Coates
Hi, I'm trying to understand how the lifecycle of messages / state is managed by Flink, but I'm failing to find any documentation. Specially, if I'm using a windowed stream and a type of trigger that retain the elements of the window to allow for processing of late data e.g. ContinousEventTimeTri

Re: Flink message & state lifecycle.

2016-01-15 Thread Andrew Coates
a new result is emitted. > */ >public static EventTimeTrigger accumulating(AbstractTime > allowedLateness) { > return new EventTimeTrigger(true, allowedLateness.toMilliseconds()); >} > } > > You can specify a lateness and while that time is not yet reached the

Re: Flink message & state lifecycle.

2016-01-15 Thread Andrew Coates
> fires. Only after the > > * watermark passes the specified lateness are the window elements > discarded, without > > * emitting a new result. If a late element arrives within the > specified lateness > > * the window is computed again and a new result is emit

Re: Flink message & state lifecycle.

2016-01-15 Thread Andrew Coates
is smart enough, configurable enough, or customisable enough to allow *where *the ageing state is kept. Thoughts? Thanks! Andy On Fri, 15 Jan 2016 at 15:51 Andrew Coates wrote: > Hi Aljoscha, > > Thanks for the info! > > Andy > > On Fri, 15 Jan 2016 at 10:12 Aljoscha Krettek wrote:

Re: Flink message & state lifecycle.

2016-01-16 Thread Andrew Coates
on s state backend > that stores state in HDFS TFiles (if I’m not mistaken) and he also > previously contributed the DB state backend that can store state in a SQL > data base. > > Cheers, > Aljoscha > > On 15 Jan 2016, at 16:56, Andrew Coates wrote: > > > > > >

Re: Does Kafka connector leverage Kafka message keys?

2016-04-13 Thread Andrew Coates
Hi Stephan, If we were to do that, would flink leverage the fact that Kafka has already partitioned the data by the key, or would flink attempt to shuffle the data again into its own partitions, potentially shuffling data between machines for no gain? Thanks, Andy On Sun, 10 Apr 2016, 13:22 Ste

Re: streaming join implementation

2016-04-14 Thread Andrew Coates
Extending on what Henry is asking... What if data can be more that a day late, or in a more streaming nature, what if updates can come through for previous values? This would obviously involve storing a great deal of state. The use case I'm thinking of has large large volumes per day. So an extern