Stream conversion

[Sane Lee]

Dear all,

I want to convert the data from each window of stream to dataset. What is the best way to do that?  So, while streaming, at the end of each window I want to convert those data to dataset and possible apply dataset transformations to it.

Any suggestions?

-best

-sane

[Matthias J. Sax-2]

Hi Sane,

Currently, DataSet and DataStream API a strictly separated. Thus, this
is not possible at the moment.

What kind of operation do you want to perform on the data of a window?
Why do you want to convert the data into a data set?

-Matthias

On 02/04/2016 10:11 AM, Sane Lee wrote:

> Dear all,
>
> I want to convert the data from each window of stream to dataset. What
> is the best way to do that?  So, while streaming, at the end of each
> window I want to convert those data to dataset and possible apply
> dataset transformations to it.
> Any suggestions?
>
> -best
> -sane

[Jeyhun Karimov]

Hi Matthias,

This need not to be necessarily in api functions. I just want to get a roadmap to add this functionality. Should I save each window's data into disk and create a new dataset environment in parallel? Or change trigger functionality maybe? 

I have large windows. As I asked in previous question, in flink the problem with large windows (that data inside windows may not fit in memory) will be solved. So, after getting the data of window, I want to do more than the functions in stream api. Therefore I need to convert it to dataset. Any roadmap would be appreciated.

On Thu, Feb 4, 2016 at 10:23 AM Matthias J. Sax <[hidden email]> wrote:

Hi Sane,

Currently, DataSet and DataStream API a strictly separated. Thus, this
is not possible at the moment.

What kind of operation do you want to perform on the data of a window?
Why do you want to convert the data into a data set?

-Matthias

On 02/04/2016 10:11 AM, Sane Lee wrote:
> Dear all,
>
> I want to convert the data from each window of stream to dataset. What
> is the best way to do that?  So, while streaming, at the end of each
> window I want to convert those data to dataset and possible apply
> dataset transformations to it.
> Any suggestions?
>
> -best
> -sane

[Sane Lee]

I have also, similar scenario. Any suggestion would be appreciated. 

On Thu, Feb 4, 2016 at 10:29 AM Jeyhun Karimov <[hidden email]> wrote:

Hi Matthias,

This need not to be necessarily in api functions. I just want to get a roadmap to add this functionality. Should I save each window's data into disk and create a new dataset environment in parallel? Or change trigger functionality maybe? 

I have large windows. As I asked in previous question, in flink the problem with large windows (that data inside windows may not fit in memory) will be solved. So, after getting the data of window, I want to do more than the functions in stream api. Therefore I need to convert it to dataset. Any roadmap would be appreciated.

On Thu, Feb 4, 2016 at 10:23 AM Matthias J. Sax <[hidden email]> wrote:

Hi Sane,

Currently, DataSet and DataStream API a strictly separated. Thus, this
is not possible at the moment.

What kind of operation do you want to perform on the data of a window?
Why do you want to convert the data into a data set?

-Matthias

On 02/04/2016 10:11 AM, Sane Lee wrote:
> Dear all,
>
> I want to convert the data from each window of stream to dataset. What
> is the best way to do that?  So, while streaming, at the end of each
> window I want to convert those data to dataset and possible apply
> dataset transformations to it.
> Any suggestions?
>
> -best
> -sane

[Stephan Ewen]

Hi!

If I understand you correctly, what you are looking for is a kind of periodic batch job, where the input data for each batch is a large window.

We have actually thought about this kind of application before. It is not on the short term road map that we shared a few weeks ago, but I think it will come to Flink in the mid-term (that would be in some months or so), it is asked for quite frequently.

Implementing this as a core feature is a bit of effort. A mock that writes out the windows and triggers a batch job sounds not too difficult, actually.

Greetings,

Stephan

On Thu, Feb 4, 2016 at 10:30 AM, Sane Lee <[hidden email]> wrote:

I have also, similar scenario. Any suggestion would be appreciated. 

On Thu, Feb 4, 2016 at 10:29 AM Jeyhun Karimov <[hidden email]> wrote:

Hi Matthias,

This need not to be necessarily in api functions. I just want to get a roadmap to add this functionality. Should I save each window's data into disk and create a new dataset environment in parallel? Or change trigger functionality maybe? 

I have large windows. As I asked in previous question, in flink the problem with large windows (that data inside windows may not fit in memory) will be solved. So, after getting the data of window, I want to do more than the functions in stream api. Therefore I need to convert it to dataset. Any roadmap would be appreciated.

On Thu, Feb 4, 2016 at 10:23 AM Matthias J. Sax <[hidden email]> wrote:

Hi Sane,

Currently, DataSet and DataStream API a strictly separated. Thus, this
is not possible at the moment.

What kind of operation do you want to perform on the data of a window?
Why do you want to convert the data into a data set?

-Matthias

On 02/04/2016 10:11 AM, Sane Lee wrote:
> Dear all,
>
> I want to convert the data from each window of stream to dataset. What
> is the best way to do that?  So, while streaming, at the end of each
> window I want to convert those data to dataset and possible apply
> dataset transformations to it.
> Any suggestions?
>
> -best
> -sane

[rmetzger0]

I'm wondering which kind of transformations you want to apply to the window you cannot apply with the DataStream API?

Would it be sufficient for you to have the windows as files in HDFS and then run batch jobs against the windows on disk? If so, you could use our filesystem sink, which creates files bucketed by certain time-windows.

On Thu, Feb 4, 2016 at 11:33 AM, Stephan Ewen <[hidden email]> wrote:

Hi!

If I understand you correctly, what you are looking for is a kind of periodic batch job, where the input data for each batch is a large window.

We have actually thought about this kind of application before. It is not on the short term road map that we shared a few weeks ago, but I think it will come to Flink in the mid-term (that would be in some months or so), it is asked for quite frequently.

Implementing this as a core feature is a bit of effort. A mock that writes out the windows and triggers a batch job sounds not too difficult, actually.

Greetings,

Stephan

On Thu, Feb 4, 2016 at 10:30 AM, Sane Lee <[hidden email]> wrote:

I have also, similar scenario. Any suggestion would be appreciated. 

On Thu, Feb 4, 2016 at 10:29 AM Jeyhun Karimov <[hidden email]> wrote:

Hi Matthias,

This need not to be necessarily in api functions. I just want to get a roadmap to add this functionality. Should I save each window's data into disk and create a new dataset environment in parallel? Or change trigger functionality maybe? 

I have large windows. As I asked in previous question, in flink the problem with large windows (that data inside windows may not fit in memory) will be solved. So, after getting the data of window, I want to do more than the functions in stream api. Therefore I need to convert it to dataset. Any roadmap would be appreciated.

On Thu, Feb 4, 2016 at 10:23 AM Matthias J. Sax <[hidden email]> wrote:

Hi Sane,

Currently, DataSet and DataStream API a strictly separated. Thus, this
is not possible at the moment.

What kind of operation do you want to perform on the data of a window?
Why do you want to convert the data into a data set?

-Matthias

On 02/04/2016 10:11 AM, Sane Lee wrote:
> Dear all,
>
> I want to convert the data from each window of stream to dataset. What
> is the best way to do that?  So, while streaming, at the end of each
> window I want to convert those data to dataset and possible apply
> dataset transformations to it.
> Any suggestions?
>
> -best
> -sane

[Jeyhun Karimov]

For example, I will do aggregate operations with other windows (n-window aggregations) that are already outputted.

I tried your suggestion and used filesystem sink, outputted to HDFS.

 I got k files in HDFS directory where k is the number of parallelism (I used single machine).

These files get bigger (new records are appended) as stream continues. Because they are (outputted files) not closed and file size is changed regularly, would this cause some problems while processing data with dataset api or hadoop or another library?

On Thu, Feb 4, 2016 at 2:14 PM Robert Metzger <[hidden email]> wrote:

I'm wondering which kind of transformations you want to apply to the window you cannot apply with the DataStream API?

Would it be sufficient for you to have the windows as files in HDFS and then run batch jobs against the windows on disk? If so, you could use our filesystem sink, which creates files bucketed by certain time-windows.

On Thu, Feb 4, 2016 at 11:33 AM, Stephan Ewen <[hidden email]> wrote:

Hi!

If I understand you correctly, what you are looking for is a kind of periodic batch job, where the input data for each batch is a large window.

We have actually thought about this kind of application before. It is not on the short term road map that we shared a few weeks ago, but I think it will come to Flink in the mid-term (that would be in some months or so), it is asked for quite frequently.

Implementing this as a core feature is a bit of effort. A mock that writes out the windows and triggers a batch job sounds not too difficult, actually.

Greetings,

Stephan

On Thu, Feb 4, 2016 at 10:30 AM, Sane Lee <[hidden email]> wrote:

I have also, similar scenario. Any suggestion would be appreciated. 

On Thu, Feb 4, 2016 at 10:29 AM Jeyhun Karimov <[hidden email]> wrote:

Hi Matthias,

This need not to be necessarily in api functions. I just want to get a roadmap to add this functionality. Should I save each window's data into disk and create a new dataset environment in parallel? Or change trigger functionality maybe? 

I have large windows. As I asked in previous question, in flink the problem with large windows (that data inside windows may not fit in memory) will be solved. So, after getting the data of window, I want to do more than the functions in stream api. Therefore I need to convert it to dataset. Any roadmap would be appreciated.

On Thu, Feb 4, 2016 at 10:23 AM Matthias J. Sax <[hidden email]> wrote:

Hi Sane,

Currently, DataSet and DataStream API a strictly separated. Thus, this
is not possible at the moment.

What kind of operation do you want to perform on the data of a window?
Why do you want to convert the data into a data set?

-Matthias

On 02/04/2016 10:11 AM, Sane Lee wrote:
> Dear all,
>
> I want to convert the data from each window of stream to dataset. What
> is the best way to do that?  So, while streaming, at the end of each
> window I want to convert those data to dataset and possible apply
> dataset transformations to it.
> Any suggestions?
>
> -best
> -sane

[Ufuk Celebi]

> On 05 Feb 2016, at 08:56, Jeyhun Karimov <[hidden email]> wrote:
>
> For example, I will do aggregate operations with other windows (n-window aggregations) that are already outputted.
> I tried your suggestion and used filesystem sink, outputted to HDFS.
>  I got k files in HDFS directory where k is the number of parallelism (I used single machine).
> These files get bigger (new records are appended) as stream continues. Because they are (outputted files) not closed and file size is changed regularly, would this cause some problems while processing data with dataset api or hadoop or another library?

I think you have used the plain file sink and Robert was referring to the rolling HDFS file sink [1] This will bucket your data in different directories like this: /base/path/{date-time}/part-{parallel-task}-{count}

– Ufuk

[1] https://ci.apache.org/projects/flink/flink-docs-master/apis/streaming/connectors/hdfs.html