I have observed that a Flink-on-Tez test job stalls in two cases on the
Travis CI server.
https://travis-ci.org/StephanEwen/incubator-flink/jobs/62302207
It looks like a shuffle fetch is simply not continuing, but freezing. The
stack traces suggest at a first glance that this is actually a Tez is
Hi!
The *collection execution* runs the program simply as functions over Java
collections. It is single threaded, always local, and does not use any
Flink memory management, serialization, or so. It is designed to be very
lightweight and is tailored towards very small problems.
The *cluster mode*
Hello,
Thanks Andra for the gaussian sequence generation. It is a little
tricky, i just leave this part for future work.
I meet another problem in AffinityPropogation algorithm. I write a few
test code for it.
https://github.com/joey001/flink/blob/ap_add/flink-staging/flink-gelly/src/test/ja
Awesome work, thanks Max ! =)
- Henry
On Mon, May 11, 2015 at 9:56 AM, Maximilian Michels wrote:
> We're now on Git for our website!
>
> Instructions for changing the website have been updated in the "How to
> contribute" guide:
>
> http://flink.apache.org/how-to-contribute.html#contribute-chang
Henry Saputra created FLINK-2005:
Summary: Remove dependencies on Record APIs for flink-jdbc module
Key: FLINK-2005
URL: https://issues.apache.org/jira/browse/FLINK-2005
Project: Flink
Issue
Stephan Ewen created FLINK-2004:
---
Summary: Memory leack in presence of failed checkpoints in
KafkaSource
Key: FLINK-2004
URL: https://issues.apache.org/jira/browse/FLINK-2004
Project: Flink
Is
Thanks Max!
Happy to have the website on Git :-)
2015-05-11 18:56 GMT+02:00 Maximilian Michels :
> We're now on Git for our website!
>
> Instructions for changing the website have been updated in the "How to
> contribute" guide:
>
>
> http://flink.apache.org/how-to-contribute.html#contribute-chan
Hi,
This was the exact need that motivated me to rework the windowing and
introduce the StreamWindow abstraction which can hold any metadata that
represents the current window.
At this moment it only contains a unique id but this could be extended
easily.
When the user created a windoweddatastrea
Hi,
I'll try to make it quick this time. I think we need to make
information about the event time of an element and information about
windows in which it resides accessible to the user. A simple example
would be the aggregation of some user behaviour, for example:
in = clickSource()
analysedData
Watermarks also don't need to flush buffers, they can actually simply queue
in as special stream records, if we want to.
On Tue, May 12, 2015 at 2:53 PM, Gyula Fóra wrote:
> Its actually a very different mechanism as watermarks will not block the
> computations
>
> On Tue, May 12, 2015 at 2:48 P
What Stephan mentioned is exactly how I'm planning to implement it,
yes. How do the barriers work with chained tasks and OutputSelectorS?
Or is there no special-case code required?
On Tue, May 12, 2015 at 2:53 PM, Gyula Fóra wrote:
> Its actually a very different mechanism as watermarks will not
Its actually a very different mechanism as watermarks will not block the
computations
On Tue, May 12, 2015 at 2:48 PM, Matthias J. Sax <
mj...@informatik.hu-berlin.de> wrote:
> Hi,
>
> I don't understand why we need the same machnism twice in the code...
> Could checkpoing barrieres and low water
Hi,
I don't understand why we need the same machnism twice in the code...
Could checkpoing barrieres and low watermarks be unified (or one build
on-top/by-using the other)
-Matthias
On 05/12/2015 02:47 PM, Gyula Fóra wrote:
> Hi,
>
> Checkpoint barriers are handled directly on top of the netwo
I would like to refrain from adding additional tasks as much as possible.
I agree with Gyula that extending the reader to track watermarks and call a
handler whenever the watermark advances would be a nice way to implement
this.
On Tue, May 12, 2015 at 2:40 PM, Aljoscha Krettek
wrote:
> Hi Folk
Hi,
Checkpoint barriers are handled directly on top of the network layer and
you are right they work similarly, by blocking input channels until it gets
the barrier from all of them.
A way of implementing this on the operator level would be by adding a way
to ask the inputreader the channel index
Hi Folks,
as I said in the subject. How will this work? I'm in the process about
thinking how to implement low watermarks in Streaming. I'm thinking
that the implementation should be quite similar to how the
checkpointing barriers will be implemented since they also flush out
stuff.
Now I'm wonder
In streaming API you can get efficiency as only some data changes.
I am looking into moving window calculations on real time market data to
generate trading signals
--
View this message in context:
http://apache-flink-mailing-list-archive.1008284.n3.nabble.com/Gelly-and-ML-for-Streaming-tp556
PS. Is there a particular reason why the APIs are stacked above each other
in the picture (ML on top of Gelly on top of the Table API)? I was actually
picturing the three next to each other...
2015-05-12 12:08 GMT+02:00 Alexander Alexandrov <
alexander.s.alexand...@gmail.com>:
> I suggest to chan
The problem is that the the StreamIterationHead is not created, because
only IterativeDataStream.transform(...) can create it. groupBy() on an
IterativeDataStream does not call transform(), therefore the exception. All
methods of DataStream that is supported for iterations and do not call
transform
I suggest to change the layout of the bottom half in the following way
(will solve the alignment issue):
- 2 column layout in 1:1 ratio for *Getting Started*, 1st column with the
text and the download button, second column with the maven code snippets
- 2 column layout in 1:1 ratio for the *Recen
Hi,
some UnsupportedOperationExceptions are required, because Storm
interfaces are implement but Flink cannot support those functionality.
Some other are "not yet implemented" once.
A few other of them could be removed (in case an interface in not
implemented, but only mimicked), by removing the
Theodore Vasiloudis created FLINK-2003:
--
Summary: Building on some encrypted filesystems leads to "File
name too long" error
Key: FLINK-2003
URL: https://issues.apache.org/jira/browse/FLINK-2003
Péter Szabó created FLINK-2002:
--
Summary: Iterative test fails when ran with other tests in the
same environment
Key: FLINK-2002
URL: https://issues.apache.org/jira/browse/FLINK-2002
Project: Flink
Every vote counts. :D
On Tue, May 12, 2015 at 11:04 AM, Matthias J. Sax
wrote:
> I like it. Not sure if my vote counts ;)
>
> On 05/12/2015 07:18 AM, Aljoscha Krettek wrote:
>> My proposal for the runtime classes (per my Pull Request is this):
>>
>> StreamTask: base of streaming tasks, the task
Hi again Suminda!
The stream ML api will be along the lines of the batch ML library and will have
some interesting features. We consider re-using as much as possible from the
batch ML (e.g. the same data structures and general abstractions etc.). Faye
and Martha (CCed) are looking into stream
I like it. Not sure if my vote counts ;)
On 05/12/2015 07:18 AM, Aljoscha Krettek wrote:
> My proposal for the runtime classes (per my Pull Request is this):
>
> StreamTask: base of streaming tasks, the task is the AbstractInvokable
> that runs in the TaskManager and invokes stream operators
> O
I think starting with Ufuk's draft, de-cluttering it is a good start.
- Having a more prominent tag line
- Add some pictures for features back
- align the content as per Felix' suggestion
That should give us a faster quick overview, but keep some serious content
in the starting page.
I especi
+1 to merge it to contrib.
We should hold a path open that
- inactive code gets dropped from contrib after a while
- very active code may gets its own project (inside or outside the Flink
codebase, whatever fits better)
On Tue, May 12, 2015 at 9:52 AM, Márton Balassi
wrote:
> The purpose of
Hi,
Thank you for starting the discussion Marton!
I would really like to merge the storm compat to our source repo. I think
that code which is not merged there will not get enough attention.
I'm against splitting flink-contrib into small maven modules. I totally
understand your reasoning (mixed
Good points raised by Stephan, Felix, and Volker.
Do you think we can achieve this by iterating on the new design? One thing
we can indeed do is de-clutter a bit. Apart from that, what high level
points would you like to see in the frontpage?
On Tue, May 12, 2015 at 6:16 AM, Markl, Volker, Prof.
Hi Suminda,
indeed this is a very exciting idea and we have been working on both Gelly
streaming and ML streaming for a while here in Stockholm.
Daniel has been looking into graph streaming for his thesis together with
Paris and myself.
We have evaluated existing streaming models and algorithms a
The purpose of flink-contrib currently is to hold contributions to the
project that we do not consider part of the core flink functionality, but
provide useful tools around it. In general code placed here has to meet
less requirements in terms of covering all corner cases if it provides a
nice solu
Dear Suminda,
Thanks for the suggestion.
Some folks at Stockholm have started implementing these, but are a bit
quite about it. Paris, Vasia, Gyula and Daniel Bali can tell you more on
the topic just to name a few.
Best,
Marton
On Tue, May 12, 2015 at 7:37 AM, sirinath wrote:
> It would be go
33 matches
Mail list logo