flink-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Vinay Patil <vinay18.pa...@gmail.com>
Subject Re: [Discussion] Query regarding Join
Date Wed, 29 Jun 2016 11:18:46 GMT
Hi Aljoscha,

This clears a lot of doubts now.
So now lets say the stream paused for a while or it stops completely on
Friday , let us assume that the last message did not get processed and is
kept in the internal buffers.

So when the stream starts again on Monday , will it consider the last
element that is in the internal buffer for processing ?
 How much time the internal buffer can hold the data or will it flush the
data after a threshold ?

I have tried using AssignerWithPunctuatedWatermarks and generated the
watermark for each event, still getting one record less.


Regards,
Vinay Patil

On Wed, Jun 29, 2016 at 2:21 PM, Aljoscha Krettek <aljoscha@apache.org>
wrote:

> Hi,
> the reason why the last element might never be emitted is the way the
> ascending timestamp extractor works. I'll try and explain with an example.
>
> Let's say we have a window size of 2 milliseconds, elements arrive starting
> with timestamp 0, window begin timestamp is inclusive, end timestamp is
> exclusive:
>
> Element 0, Timestamp 0 (at this point the watermark is -1)
> Element 1, Timestamp 1 (at this point the watermark is 0)
> Element 2, Timestamp 1 (at this point the watermark is still 0)
> Element 3, Timestamp 2 (at this point the watermark is 1)
>
> now we can process the window (0, 2) because we know from the watermark
> that no elements can arrive for that window anymore. The window contains
> elements 0,1,2
>
> Element 4, Timestamp 3 (at this point the watermark is 2)
> Element 5, Timestamp 4 (at this point the watermark is 3)
>
> now we can process window (2, 4). The window contains elements 3,4.
>
> At this point, we have Element 5 sitting in internal buffers for window (4,
> 6) but if we don't receive further elements the watermark will never
> advance and we will never process that window.
>
> If, however, we get new elements at some point the watermark advances and
> we don't have a problem. That's what I meant when I said that you shouldn't
> have a problem if data keeps continuously arriving.
>
> Cheers,
> Aljoscha
>
>
> On Tue, 28 Jun 2016 at 17:14 Vinay Patil <vinay18.patil@gmail.com> wrote:
>
> > Hi Aljoscha,
> >
> > Thanks a lot for your inputs.
> >
> > I still did not get you when you say you will not face this issue in case
> > of continuous stream, lets consider the following example :
> > Assume that the stream runs continuously from Monday  to Friday, and on
> > Friday it stops after 5.00 PM , will I still face this issue ?
> >
> > I am actually not able to understand how it will differ in real time
> > streams.
> >
> > Regards,
> > Vinay Patil
> >
> > On Tue, Jun 28, 2016 at 5:07 PM, Aljoscha Krettek <aljoscha@apache.org>
> > wrote:
> >
> > > Hi,
> > > ingestion time can only be used if you don't care about the timestamp
> in
> > > the elements. So if you have those you should probably use event time.
> > >
> > > If your timestamps really are strictly increasing then the ascending
> > > extractor is good. And if you have a continuous stream of incoming
> > elements
> > > you will not see the behavior of not getting the last elements.
> > >
> > > By the way, when using Kafka you can also embed the timestamp extractor
> > > directly in the Kafka consumer. This is described here:
> > >
> > >
> >
> https://ci.apache.org/projects/flink/flink-docs-master/apis/streaming/connectors/kafka.html#kafka-consumers-and-timestamp-extractionwatermark-emission
> > >
> > > Cheers,
> > > Aljoscha
> > >
> > > On Tue, 28 Jun 2016 at 11:44 Vinay Patil <vinay18.patil@gmail.com>
> > wrote:
> > >
> > > > Hi Aljoscha,
> > > >
> > > > Thank you for your response.
> > > > So do you suggest to use different approach for extracting timestamp
> > (as
> > > > given in document) instead of AscendingTimeStamp Extractor ?
> > > > Is that the reason I am seeing this unexpected behaviour ? in case of
> > > > continuous stream I would not see any data loss ?
> > > >
> > > > Also assuming that the records are always going to be in order ,
> which
> > is
> > > > the best approach : Ingestion Time or Event Time ?
> > > >
> > > >
> > > >
> > > > Regards,
> > > > Vinay Patil
> > > >
> > > > On Tue, Jun 28, 2016 at 2:41 PM, Aljoscha Krettek <
> aljoscha@apache.org
> > >
> > > > wrote:
> > > >
> > > > > Hi,
> > > > > first regarding tumbling windows: even if you have 5 minute windows
> > it
> > > > can
> > > > > happen that elements that are only seconds apart go into different
> > > > windows.
> > > > > Consider the following case:
> > > > >
> > > > > |                x | x                 |
> > > > >
> > > > > These are two 5-mintue windows and the two elements are only
> seconds
> > > > apart
> > > > > but go into different windows because windows are aligned to epoch.
> > > > >
> > > > > Now, for the ascending timestamp extractor. The reason this can
> > behave
> > > in
> > > > > unexpected ways is that it emits a watermark that is "last
> timestamp
> > -
> > > > 1",
> > > > > i.e. if it has seen timestamp t it can only emit watermark t-1
> > because
> > > > > there might be other elements with timestamp t arriving. If you
> have
> > a
> > > > > continuous stream of elements you wouldn't notice this. Only in
> this
> > > > > constructed example does it become visible.
> > > > >
> > > > > Cheers,
> > > > > Aljoscha
> > > > >
> > > > > On Tue, 28 Jun 2016 at 06:04 Vinay Patil <vinay18.patil@gmail.com>
> > > > wrote:
> > > > >
> > > > > > Hi,
> > > > > >
> > > > > > Following is the timestamp I am getting from DTO, here is the
> > > timestamp
> > > > > > difference between the two records :
> > > > > > 1466115892162154279
> > > > > > 1466116026233613409
> > > > > >
> > > > > > So the time difference is roughly 3 min, even if I apply the
> window
> > > of
> > > > > 5min
> > > > > > , I am not getting the last record (last timestamp value above),
> > > > > > using ascending timestamp extractor for generating the timestamp
> > > > > (assuming
> > > > > > that the timestamp are always in order)
> > > > > >
> > > > > > I was at-least expecting data to reach the co-group function.
> > > > > > What could be the reason for the data loss ? The data we are
> > getting
> > > is
> > > > > > critical, hence we cannot afford to loose any data
> > > > > >
> > > > > >
> > > > > > Regards,
> > > > > > Vinay Patil
> > > > > >
> > > > > > On Mon, Jun 27, 2016 at 11:31 PM, Vinay Patil <
> > > vinay18.patil@gmail.com
> > > > >
> > > > > > wrote:
> > > > > >
> > > > > > > Just an update, when I keep IngestionTime and remove the
> > timestamp
> > > I
> > > > am
> > > > > > > generating, I am getting all the records, but for Event
Time I
> am
> > > > > getting
> > > > > > > one less record, I checked the Time Difference between
two
> > records,
> > > > it
> > > > > > is 3
> > > > > > > min, I tried keeping the window time to 5 mins, but that
even
> did
> > > not
> > > > > > work.
> > > > > > >
> > > > > > > Even when I try assigning timestamp for IngestionTime,
I get
> one
> > > > record
> > > > > > > less, so should I safely use Ingestion Time or is it always
> > > advisable
> > > > > to
> > > > > > > use EventTime ?
> > > > > > >
> > > > > > > Regards,
> > > > > > > Vinay Patil
> > > > > > >
> > > > > > > On Mon, Jun 27, 2016 at 8:16 PM, Vinay Patil <
> > > > vinay18.patil@gmail.com>
> > > > > > > wrote:
> > > > > > >
> > > > > > >> Hi ,
> > > > > > >>
> > > > > > >> Actually I am only publishing 5 messages each to two
different
> > > kafka
> > > > > > >> topics (using Junit), even if I keep the window to
500 seconds
> > the
> > > > > > result
> > > > > > >> is same.
> > > > > > >>
> > > > > > >> I am not understanding why it is not sending the 5th
element
> to
> > > > > co-group
> > > > > > >> operator even when the keys are same.
> > > > > > >>
> > > > > > >> I actually cannot share the actual client code.
> > > > > > >> But this is what the streams look like :
> > > > > > >> sourceStream.coGroup(destStream)
> > > > > > >> here the sourceStream and destStream is actually
> > > Tuple2<String,DTO>
> > > > ,
> > > > > > and
> > > > > > >> the ElementSelector returns tuple.f0 which is the key.
> > > > > > >>
> > > > > > >> I am generating the timestamp based on a field from
the DTO
> > which
> > > is
> > > > > > >> guaranteed to be in order.
> > > > > > >>
> > > > > > >> Will using the triggers help here ?
> > > > > > >>
> > > > > > >>
> > > > > > >> Regards,
> > > > > > >> Vinay Patil
> > > > > > >>
> > > > > > >> *+91-800-728-4749*
> > > > > > >>
> > > > > > >> On Mon, Jun 27, 2016 at 7:05 PM, Aljoscha Krettek <
> > > > > aljoscha@apache.org>
> > > > > > >> wrote:
> > > > > > >>
> > > > > > >>> Hi,
> > > > > > >>> what timestamps are you assigning? Is it guaranteed
that all
> of
> > > > them
> > > > > > >>> would
> > > > > > >>> fall into the same 30 second window?
> > > > > > >>>
> > > > > > >>> The issue with duplicate printing in the ElementSelector
is
> > > > strange?
> > > > > > >>> Could
> > > > > > >>> you post a more complete code example so that I
can reproduce
> > the
> > > > > > >>> problem?
> > > > > > >>>
> > > > > > >>> Cheers,
> > > > > > >>> Aljoscha
> > > > > > >>>
> > > > > > >>> On Mon, 27 Jun 2016 at 13:21 Vinay Patil <
> > > vinay18.patil@gmail.com>
> > > > > > >>> wrote:
> > > > > > >>>
> > > > > > >>> > Hi ,
> > > > > > >>> >
> > > > > > >>> > I am able to get the matching and non-matching
elements.
> > > > > > >>> >
> > > > > > >>> > However when I am unit testing the code ,
I am getting one
> > > record
> > > > > > less
> > > > > > >>> > inside the overriden cogroup function.
> > > > > > >>> > Testing the following way :
> > > > > > >>> >
> > > > > > >>> > 1) Insert 5 messages into local kafka topic
(test1)
> > > > > > >>> > 2) Insert different 5 messages into local
kafka topic
> (test2)
> > > > > > >>> > 3) Consume 1) and 2) and I have two different
kafka
> streams
> > > > > > >>> > 4) Generate ascending timestamp(using Event
Time) for both
> > > > streams
> > > > > > and
> > > > > > >>> > create key(String)
> > > > > > >>> >
> > > > > > >>> > Now till 4) I am able to get all the records
(checked by
> > > printing
> > > > > the
> > > > > > >>> > stream in text file)
> > > > > > >>> >
> > > > > > >>> > However when I send the stream to co-group
operator, I am
> > > > receiving
> > > > > > one
> > > > > > >>> > less record, using the following syntax:
> > > > > > >>> >
> > > > > > >>> > sourceStream.coGroup(destStream)
> > > > > > >>> > .where(new ElementSelector())
> > > > > > >>> > .equalTo(new ElementSelector())
> > > > > > >>> > .window(TumblingEventTimeWindows.of(Time.seconds(30)))
> > > > > > >>> > .apply(new JoinStreams);
> > > > > > >>> >
> > > > > > >>> > Also in the Element Selector I have inserted
a sysout, I am
> > > > getting
> > > > > > 20
> > > > > > >>> > sysouts instead of 10 (10 sysouts for source
and 10 for
> dest
> > > > > stream)
> > > > > > >>> >
> > > > > > >>> > Unable to understand why one record is coming
less to
> > co-group
> > > > > > >>> >
> > > > > > >>> >
> > > > > > >>> >
> > > > > > >>> > Regards,
> > > > > > >>> > Vinay Patil
> > > > > > >>> >
> > > > > > >>> > On Wed, Jun 15, 2016 at 1:39 PM, Fabian Hueske
<
> > > > fhueske@gmail.com>
> > > > > > >>> wrote:
> > > > > > >>> >
> > > > > > >>> > > Can you add a flag to each element emitted
by the
> > > > CoGroupFunction
> > > > > > >>> that
> > > > > > >>> > > indicates whether it was joined or not?
> > > > > > >>> > > Then you can use split to distinguish
between both cases
> > and
> > > > > handle
> > > > > > >>> both
> > > > > > >>> > > streams differently.
> > > > > > >>> > >
> > > > > > >>> > > Best, Fabian
> > > > > > >>> > >
> > > > > > >>> > > 2016-06-15 6:45 GMT+02:00 Vinay Patil
<
> > > vinay18.patil@gmail.com
> > > > >:
> > > > > > >>> > >
> > > > > > >>> > > > Hi Jark,
> > > > > > >>> > > >
> > > > > > >>> > > > I am able to get the non-matching
elements in a stream
> :,
> > > > > > >>> > > >
> > > > > > >>> > > > Of-course we can collect the matching
elements in the
> > same
> > > > > stream
> > > > > > >>> as
> > > > > > >>> > > well,
> > > > > > >>> > > > however I want to perform additional
operations on the
> > > joined
> > > > > > >>> stream
> > > > > > >>> > > before
> > > > > > >>> > > > writing it to S3, so I would have
to include a separate
> > > join
> > > > > > >>> operator
> > > > > > >>> > for
> > > > > > >>> > > > the same two streams, right ?
> > > > > > >>> > > > Correct me if I am wrong.
> > > > > > >>> > > >
> > > > > > >>> > > > I have pasted the dummy code which
collects the
> > > non-matching
> > > > > > >>> records (i
> > > > > > >>> > > > have to perform this on the actual
data, correct me if
> I
> > am
> > > > > dong
> > > > > > >>> > wrong).
> > > > > > >>> > > >
> > > > > > >>> > > > sourceStream.coGroup(destStream).where(new
> > > > > > >>> > ElementSelector()).equalTo(new
> > > > > > >>> > > > ElementSelector())
> > > > > > >>> > > > .window(TumblingEventTimeWindows.of(Time.seconds(30)))
> > > > > > >>> > > > .apply(new CoGroupFunction<Integer,
Integer,
> Integer>() {
> > > > > > >>> > > >
> > > > > > >>> > > > private static final long serialVersionUID
=
> > > > > > 6408179761497497475L;
> > > > > > >>> > > >
> > > > > > >>> > > > @Override
> > > > > > >>> > > > public void coGroup(Iterable<Integer>
paramIterable,
> > > > > > >>> Iterable<Integer>
> > > > > > >>> > > > paramIterable1,
> > > > > > >>> > > > Collector<Integer> paramCollector)
throws Exception {
> > > > > > >>> > > > long exactSizeIfKnown =
> > > > > > >>> > > paramIterable.spliterator().getExactSizeIfKnown();
> > > > > > >>> > > > long exactSizeIfKnown2 =
> > > > > > >>> > > > paramIterable1.spliterator().getExactSizeIfKnown();
> > > > > > >>> > > > if(exactSizeIfKnown == 0 ) {
> > > > > > >>> > > >
> paramCollector.collect(paramIterable1.iterator().next());
> > > > > > >>> > > > } else if (exactSizeIfKnown2 ==
0) {
> > > > > > >>> > > >
> paramCollector.collect(paramIterable.iterator().next());
> > > > > > >>> > > > }
> > > > > > >>> > > > }
> > > > > > >>> > > > }).print();
> > > > > > >>> > > >
> > > > > > >>> > > > Regards,
> > > > > > >>> > > > Vinay Patil
> > > > > > >>> > > >
> > > > > > >>> > > >
> > > > > > >>> > > > On Tue, Jun 14, 2016 at 1:37 PM,
Vinay Patil <
> > > > > > >>> vinay18.patil@gmail.com>
> > > > > > >>> > > > wrote:
> > > > > > >>> > > >
> > > > > > >>> > > > > You are right, debugged it
for all elements , I can
> do
> > > that
> > > > > > now.
> > > > > > >>> > > > > Thanks a lot.
> > > > > > >>> > > > >
> > > > > > >>> > > > > Regards,
> > > > > > >>> > > > > Vinay Patil
> > > > > > >>> > > > >
> > > > > > >>> > > > > On Tue, Jun 14, 2016 at 11:56
AM, Jark Wu <
> > > > > > >>> > wuchong.wc@alibaba-inc.com>
> > > > > > >>> > > > > wrote:
> > > > > > >>> > > > >
> > > > > > >>> > > > >> In `coGroup(Iterable<Integer>
iter1,
> Iterable<Integer>
> > > > > iter2,
> > > > > > >>> > > > >> Collector<Integer>
out)` ,   when both iter1 and
> iter2
> > > are
> > > > > not
> > > > > > >>> > empty,
> > > > > > >>> > > it
> > > > > > >>> > > > >> means they are matched
elements from both stream.
> > > > > > >>> > > > >> When one of iter1 and iter2
is empty , it means that
> > > they
> > > > > are
> > > > > > >>> > > unmatched.
> > > > > > >>> > > > >>
> > > > > > >>> > > > >>
> > > > > > >>> > > > >> - Jark Wu (wuchong)
> > > > > > >>> > > > >>
> > > > > > >>> > > > >> > 在 2016年6月14日,下午12:46,Vinay
Patil <
> > > > vinay18.patil@gmail.com
> > > > > >
> > > > > > >>> 写道:
> > > > > > >>> > > > >> >
> > > > > > >>> > > > >> > Hi Matthias ,
> > > > > > >>> > > > >> >
> > > > > > >>> > > > >> > I did not get you,
even if we use Co-Group we have
> > to
> > > > > apply
> > > > > > >>> it on
> > > > > > >>> > a
> > > > > > >>> > > > key
> > > > > > >>> > > > >> >
> > > > > > >>> > > > >> > sourceStream.coGroup(destStream)
> > > > > > >>> > > > >> > .where(new ElementSelector())
> > > > > > >>> > > > >> > .equalTo(new ElementSelector())
> > > > > > >>> > > > >> >
> > .window(TumblingEventTimeWindows.of(Time.seconds(30)))
> > > > > > >>> > > > >> > .apply(new CoGroupFunction<Integer,
Integer,
> > > Integer>()
> > > > {
> > > > > > >>> > > > >> > private static final
long serialVersionUID =
> > > > > > >>> 6408179761497497475L;
> > > > > > >>> > > > >> >
> > > > > > >>> > > > >> > @Override
> > > > > > >>> > > > >> > public void coGroup(Iterable<Integer>
> paramIterable,
> > > > > > >>> > > Iterable<Integer>
> > > > > > >>> > > > >> > paramIterable1,
> > > > > > >>> > > > >> > Collector<Integer>
paramCollector) throws
> Exception
> > {
> > > > > > >>> > > > >> > Iterator<Integer>
iterator =
> > paramIterable.iterator();
> > > > > > >>> > > > >> > while(iterator.hasNext())
{
> > > > > > >>> > > > >> > }
> > > > > > >>> > > > >> > }
> > > > > > >>> > > > >> > });
> > > > > > >>> > > > >> >
> > > > > > >>> > > > >> > when I debug this
,only the matched element from
> > both
> > > > > stream
> > > > > > >>> will
> > > > > > >>> > > come
> > > > > > >>> > > > >> in
> > > > > > >>> > > > >> > the coGroup function.
> > > > > > >>> > > > >> >
> > > > > > >>> > > > >> > What I want is how
do I check for unmatched
> elements
> > > > from
> > > > > > both
> > > > > > >>> > > streams
> > > > > > >>> > > > >> and
> > > > > > >>> > > > >> > write it to sink.
> > > > > > >>> > > > >> >
> > > > > > >>> > > > >> > Regards,
> > > > > > >>> > > > >> > Vinay Patil
> > > > > > >>> > > > >> >
> > > > > > >>> > > > >> > *+91-800-728-4749*
> > > > > > >>> > > > >> >
> > > > > > >>> > > > >> > On Tue, Jun 14, 2016
at 2:07 AM, Matthias J. Sax <
> > > > > > >>> > mjsax@apache.org>
> > > > > > >>> > > > >> wrote:
> > > > > > >>> > > > >> >
> > > > > > >>> > > > >> >> You need to do
an outer-join. However, there is
> no
> > > > > build-in
> > > > > > >>> > support
> > > > > > >>> > > > for
> > > > > > >>> > > > >> >> outer-joins yet.
> > > > > > >>> > > > >> >>
> > > > > > >>> > > > >> >> You can use Window-CoGroup
to implement the
> > > outer-join
> > > > as
> > > > > > an
> > > > > > >>> own
> > > > > > >>> > > > >> operator.
> > > > > > >>> > > > >> >>
> > > > > > >>> > > > >> >>
> > > > > > >>> > > > >> >> -Matthias
> > > > > > >>> > > > >> >>
> > > > > > >>> > > > >> >> On 06/13/2016
06:53 PM, Vinay Patil wrote:
> > > > > > >>> > > > >> >>> Hi,
> > > > > > >>> > > > >> >>>
> > > > > > >>> > > > >> >>> I have a question
regarding the join operation,
> > > > consider
> > > > > > the
> > > > > > >>> > > > following
> > > > > > >>> > > > >> >>> dummy example:
> > > > > > >>> > > > >> >>>
> > > > > > >>> > > > >> >>> StreamExecutionEnvironment
env =
> > > > > > >>> > > > >> >>>
> > > StreamExecutionEnvironment.getExecutionEnvironment();
> > > > > > >>> > > > >> >>>
> > > > > > >>> >
> > > > env.setStreamTimeCharacteristic(TimeCharacteristic.IngestionTime);
> > > > > > >>> > > > >> >>> DataStreamSource<Integer>
sourceStream =
> > > > > > >>> > > > >> >>> env.fromElements(10,20,23,25,30,33,102,18);
> > > > > > >>> > > > >> >>> DataStreamSource<Integer>
destStream =
> > > > > > >>> > > > >> >> env.fromElements(20,30,40,50,60,10);
> > > > > > >>> > > > >> >>>
> > > > > > >>> > > > >> >>> sourceStream.join(destStream)
> > > > > > >>> > > > >> >>> .where(new
ElementSelector())
> > > > > > >>> > > > >> >>> .equalTo(new
ElementSelector())
> > > > > > >>> > > > >> >>>
> > > > > > .window(TumblingEventTimeWindows.of(Time.milliseconds(10)))
> > > > > > >>> > > > >> >>> .apply(new
JoinFunction<Integer, Integer,
> > > Integer>() {
> > > > > > >>> > > > >> >>>
> > > > > > >>> > > > >> >>> private static
final long serialVersionUID = 1L;
> > > > > > >>> > > > >> >>>
> > > > > > >>> > > > >> >>> @Override
> > > > > > >>> > > > >> >>> public Integer
join(Integer paramIN1, Integer
> > > > paramIN2)
> > > > > > >>> throws
> > > > > > >>> > > > >> Exception
> > > > > > >>> > > > >> >> {
> > > > > > >>> > > > >> >>> return paramIN1;
> > > > > > >>> > > > >> >>> }
> > > > > > >>> > > > >> >>> }).print();
> > > > > > >>> > > > >> >>>
> > > > > > >>> > > > >> >>> I perfectly
get the elements that are matching
> in
> > > both
> > > > > the
> > > > > > >>> > > streams,
> > > > > > >>> > > > >> >> however
> > > > > > >>> > > > >> >>> my requirement
is to write these matched
> elements
> > > and
> > > > > also
> > > > > > >>> the
> > > > > > >>> > > > >> unmatched
> > > > > > >>> > > > >> >>> elements to
sink(S3)
> > > > > > >>> > > > >> >>>
> > > > > > >>> > > > >> >>> How do I get
the unmatched elements from each
> > > stream ?
> > > > > > >>> > > > >> >>>
> > > > > > >>> > > > >> >>> Regards,
> > > > > > >>> > > > >> >>> Vinay Patil
> > > > > > >>> > > > >> >>>
> > > > > > >>> > > > >> >>
> > > > > > >>> > > > >> >>
> > > > > > >>> > > > >>
> > > > > > >>> > > > >>
> > > > > > >>> > > > >
> > > > > > >>> > > >
> > > > > > >>> > >
> > > > > > >>> >
> > > > > > >>>
> > > > > > >>
> > > > > > >>
> > > > > > >
> > > > > >
> > > > >
> > > >
> > >
> >
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message