How to load multiple same-format files with single batch job?

[françois lacombe]

Hi all,

I'm wondering if it's possible and what's the best way to achieve the loading of multiple files with a Json source to a JDBC sink ?

I'm running Flink 1.7.0

Let's say I have about 1500 files with the same structure (same format, schema, everything) and I want to load them with a *batch* job

Can Flink handle the loading of one and each file in a single source and send data to my JDBC sink?

I wish I can provide the URL of the directory containing my thousand files to the batch source to make it load all of them sequentially.

My sources and sinks are currently available for BatchTableSource, I guess the cost to make them available for streaming would be quite expensive for me for the moment.

Have someone ever done this?

Am I wrong to expect doing so with a batch job?

All the best

François Lacombe

      

 Pensez à la planète, imprimer ce papier que si nécessaire 

[Fabian Hueske-2]

Hi,

You can point a file-based input format to a directory and the input format should read all files in that directory.

That works as well for TableSources that are internally use file-based input formats.

Is that what you are looking for?

Best, Fabian

Am Mo., 28. Jan. 2019 um 17:22 Uhr schrieb françois lacombe <[hidden email]>:

Hi all,

I'm wondering if it's possible and what's the best way to achieve the loading of multiple files with a Json source to a JDBC sink ?

I'm running Flink 1.7.0

Let's say I have about 1500 files with the same structure (same format, schema, everything) and I want to load them with a *batch* job

Can Flink handle the loading of one and each file in a single source and send data to my JDBC sink?

I wish I can provide the URL of the directory containing my thousand files to the batch source to make it load all of them sequentially.

My sources and sinks are currently available for BatchTableSource, I guess the cost to make them available for streaming would be quite expensive for me for the moment.

Have someone ever done this?

Am I wrong to expect doing so with a batch job?

All the best

François Lacombe

      

 Pensez à la planète, imprimer ce papier que si nécessaire 

[françois lacombe]

Hi Fabian,

Thank you for this input.

This is interesting.

With such an input format, will all the file will be loaded in memory before to be processed or will all be streamed?

All the best

François

Le mar. 29 janv. 2019 à 22:20, Fabian Hueske <[hidden email]> a écrit :

Hi,

You can point a file-based input format to a directory and the input format should read all files in that directory.

That works as well for TableSources that are internally use file-based input formats.

Is that what you are looking for?

Best, Fabian

Am Mo., 28. Jan. 2019 um 17:22 Uhr schrieb françois lacombe <[hidden email]>:

Hi all,

I'm wondering if it's possible and what's the best way to achieve the loading of multiple files with a Json source to a JDBC sink ?

I'm running Flink 1.7.0

Let's say I have about 1500 files with the same structure (same format, schema, everything) and I want to load them with a *batch* job

Can Flink handle the loading of one and each file in a single source and send data to my JDBC sink?

I wish I can provide the URL of the directory containing my thousand files to the batch source to make it load all of them sequentially.

My sources and sinks are currently available for BatchTableSource, I guess the cost to make them available for streaming would be quite expensive for me for the moment.

Have someone ever done this?

Am I wrong to expect doing so with a batch job?

All the best

François Lacombe

      

 Pensez à la planète, imprimer ce papier que si nécessaire 

      

 Pensez à la planète, imprimer ce papier que si nécessaire 

[Fabian Hueske-2]

Hi,

The files will be read in a streaming fashion.

Typically files are broken down into processing splits that are distributed to tasks for reading.

How a task reads a file split depends on the implementation, but usually the format reads the split as a stream and does not read the split as a whole before emitting records.

Best,

Fabian

Am Mo., 4. Feb. 2019 um 12:06 Uhr schrieb françois lacombe <[hidden email]>:

Hi Fabian,

Thank you for this input.

This is interesting.

With such an input format, will all the file will be loaded in memory before to be processed or will all be streamed?

All the best

François

Le mar. 29 janv. 2019 à 22:20, Fabian Hueske <[hidden email]> a écrit :

Hi,

You can point a file-based input format to a directory and the input format should read all files in that directory.

That works as well for TableSources that are internally use file-based input formats.

Is that what you are looking for?

Best, Fabian

Am Mo., 28. Jan. 2019 um 17:22 Uhr schrieb françois lacombe <[hidden email]>:

Hi all,

I'm wondering if it's possible and what's the best way to achieve the loading of multiple files with a Json source to a JDBC sink ?

I'm running Flink 1.7.0

Let's say I have about 1500 files with the same structure (same format, schema, everything) and I want to load them with a *batch* job

Can Flink handle the loading of one and each file in a single source and send data to my JDBC sink?

I wish I can provide the URL of the directory containing my thousand files to the batch source to make it load all of them sequentially.

My sources and sinks are currently available for BatchTableSource, I guess the cost to make them available for streaming would be quite expensive for me for the moment.

Have someone ever done this?

Am I wrong to expect doing so with a batch job?

All the best

François Lacombe

      

 Pensez à la planète, imprimer ce papier que si nécessaire 

      

 Pensez à la planète, imprimer ce papier que si nécessaire 

[françois lacombe]

Thank you Fabian,

That's good, I'll go for a custom File input stream.

All the best

François

Le lun. 4 févr. 2019 à 12:10, Fabian Hueske <[hidden email]> a écrit :

Hi,

The files will be read in a streaming fashion.

Typically files are broken down into processing splits that are distributed to tasks for reading.

How a task reads a file split depends on the implementation, but usually the format reads the split as a stream and does not read the split as a whole before emitting records.

Best,

Fabian

Am Mo., 4. Feb. 2019 um 12:06 Uhr schrieb françois lacombe <[hidden email]>:

Hi Fabian,

Thank you for this input.

This is interesting.

With such an input format, will all the file will be loaded in memory before to be processed or will all be streamed?

All the best

François

Le mar. 29 janv. 2019 à 22:20, Fabian Hueske <[hidden email]> a écrit :

Hi,

You can point a file-based input format to a directory and the input format should read all files in that directory.

That works as well for TableSources that are internally use file-based input formats.

Is that what you are looking for?

Best, Fabian

Am Mo., 28. Jan. 2019 um 17:22 Uhr schrieb françois lacombe <[hidden email]>:

Hi all,

I'm wondering if it's possible and what's the best way to achieve the loading of multiple files with a Json source to a JDBC sink ?

I'm running Flink 1.7.0

Let's say I have about 1500 files with the same structure (same format, schema, everything) and I want to load them with a *batch* job

Can Flink handle the loading of one and each file in a single source and send data to my JDBC sink?

I wish I can provide the URL of the directory containing my thousand files to the batch source to make it load all of them sequentially.

My sources and sinks are currently available for BatchTableSource, I guess the cost to make them available for streaming would be quite expensive for me for the moment.

Have someone ever done this?

Am I wrong to expect doing so with a batch job?

All the best

François Lacombe

      

 Pensez à la planète, imprimer ce papier que si nécessaire 

      

 Pensez à la planète, imprimer ce papier que si nécessaire 

      

 Pensez à la planète, imprimer ce papier que si nécessaire 

[françois lacombe]

Hi Fabian,

I've got issues for a custom InputFormat implementation with my existing code.

Is this can be used in combination with a BatchTableSource custom source?

As I understand your solution, I should move my source to implementations like :

tableEnvironment
  .connect(...)
  .withFormat(...)
  .withSchema(...)
  .inAppendMode()
  .registerTableSource("MyTable")

right?

I currently have a BatchTableSource class which produce a DataSet<Row> from a single geojson file.

This doesn't sound compatible with a custom InputFormat, don't you?

Thanks in advance for any addition hint, all the best

François

Le lun. 4 févr. 2019 à 12:10, Fabian Hueske <[hidden email]> a écrit :

Hi,

The files will be read in a streaming fashion.

Typically files are broken down into processing splits that are distributed to tasks for reading.

How a task reads a file split depends on the implementation, but usually the format reads the split as a stream and does not read the split as a whole before emitting records.

Best,

Fabian

Am Mo., 4. Feb. 2019 um 12:06 Uhr schrieb françois lacombe <[hidden email]>:

Hi Fabian,

Thank you for this input.

This is interesting.

With such an input format, will all the file will be loaded in memory before to be processed or will all be streamed?

All the best

François

Le mar. 29 janv. 2019 à 22:20, Fabian Hueske <[hidden email]> a écrit :

Hi,

You can point a file-based input format to a directory and the input format should read all files in that directory.

That works as well for TableSources that are internally use file-based input formats.

Is that what you are looking for?

Best, Fabian

Am Mo., 28. Jan. 2019 um 17:22 Uhr schrieb françois lacombe <[hidden email]>:

Hi all,

I'm wondering if it's possible and what's the best way to achieve the loading of multiple files with a Json source to a JDBC sink ?

I'm running Flink 1.7.0

Let's say I have about 1500 files with the same structure (same format, schema, everything) and I want to load them with a *batch* job

Can Flink handle the loading of one and each file in a single source and send data to my JDBC sink?

I wish I can provide the URL of the directory containing my thousand files to the batch source to make it load all of them sequentially.

My sources and sinks are currently available for BatchTableSource, I guess the cost to make them available for streaming would be quite expensive for me for the moment.

Have someone ever done this?

Am I wrong to expect doing so with a batch job?

All the best

François Lacombe

      

 Pensez à la planète, imprimer ce papier que si nécessaire 

      

 Pensez à la planète, imprimer ce papier que si nécessaire 

      

 Pensez à la planète, imprimer ce papier que si nécessaire 

[Fabian Hueske-2]

H François,

The TableEnvironment.connect() method can only be used if you provide (quite a bit) more code.

It requires a TableSourceFactory and handling of all the properties that are defined in the other builder methods. See [1].

I would recommend to either register the BatchTableSource directly (tEnv.registerTableSource()) or get a DataSet (via env.createSource()) and register the DataSet as a Table (tEnv.registerDataSet()).

Best, Fabian

[1] https://ci.apache.org/projects/flink/flink-docs-release-1.7/dev/table/sourceSinks.html#define-a-tablefactory

Am Mo., 11. Feb. 2019 um 21:09 Uhr schrieb françois lacombe <[hidden email]>:

Hi Fabian,

I've got issues for a custom InputFormat implementation with my existing code.

Is this can be used in combination with a BatchTableSource custom source?

As I understand your solution, I should move my source to implementations like :

tableEnvironment
  .connect(...)
  .withFormat(...)
  .withSchema(...)
  .inAppendMode()
  .registerTableSource("MyTable")

right?

I currently have a BatchTableSource class which produce a DataSet<Row> from a single geojson file.

This doesn't sound compatible with a custom InputFormat, don't you?

Thanks in advance for any addition hint, all the best

François

Le lun. 4 févr. 2019 à 12:10, Fabian Hueske <[hidden email]> a écrit :

Hi,

The files will be read in a streaming fashion.

Typically files are broken down into processing splits that are distributed to tasks for reading.

How a task reads a file split depends on the implementation, but usually the format reads the split as a stream and does not read the split as a whole before emitting records.

Best,

Fabian

Am Mo., 4. Feb. 2019 um 12:06 Uhr schrieb françois lacombe <[hidden email]>:

Hi Fabian,

Thank you for this input.

This is interesting.

With such an input format, will all the file will be loaded in memory before to be processed or will all be streamed?

All the best

François

Le mar. 29 janv. 2019 à 22:20, Fabian Hueske <[hidden email]> a écrit :

Hi,

You can point a file-based input format to a directory and the input format should read all files in that directory.

That works as well for TableSources that are internally use file-based input formats.

Is that what you are looking for?

Best, Fabian

Am Mo., 28. Jan. 2019 um 17:22 Uhr schrieb françois lacombe <[hidden email]>:

Hi all,

I'm wondering if it's possible and what's the best way to achieve the loading of multiple files with a Json source to a JDBC sink ?

I'm running Flink 1.7.0

Let's say I have about 1500 files with the same structure (same format, schema, everything) and I want to load them with a *batch* job

Can Flink handle the loading of one and each file in a single source and send data to my JDBC sink?

I wish I can provide the URL of the directory containing my thousand files to the batch source to make it load all of them sequentially.

My sources and sinks are currently available for BatchTableSource, I guess the cost to make them available for streaming would be quite expensive for me for the moment.

Have someone ever done this?

Am I wrong to expect doing so with a batch job?

All the best

François Lacombe

      

 Pensez à la planète, imprimer ce papier que si nécessaire 

      

 Pensez à la planète, imprimer ce papier que si nécessaire 

      

 Pensez à la planète, imprimer ce papier que si nécessaire 

[françois lacombe]

Hi Fabian,

After a bit more documentation reading I have a better understanding of how InputFormat interface works.

Indeed I've better to wrap a custom InputFormat implementation in my source.

This article helps a lot

https://brewing.codes/2017/02/06/implementing-flink-batch-data-connector/

connect() will be for a next sprint

All the best

François

Le ven. 15 févr. 2019 à 09:37, Fabian Hueske <[hidden email]> a écrit :

H François,

The TableEnvironment.connect() method can only be used if you provide (quite a bit) more code.

It requires a TableSourceFactory and handling of all the properties that are defined in the other builder methods. See [1].

I would recommend to either register the BatchTableSource directly (tEnv.registerTableSource()) or get a DataSet (via env.createSource()) and register the DataSet as a Table (tEnv.registerDataSet()).

Best, Fabian

[1] https://ci.apache.org/projects/flink/flink-docs-release-1.7/dev/table/sourceSinks.html#define-a-tablefactory

Am Mo., 11. Feb. 2019 um 21:09 Uhr schrieb françois lacombe <[hidden email]>:

Hi Fabian,

I've got issues for a custom InputFormat implementation with my existing code.

Is this can be used in combination with a BatchTableSource custom source?

As I understand your solution, I should move my source to implementations like :

tableEnvironment
  .connect(...)
  .withFormat(...)
  .withSchema(...)
  .inAppendMode()
  .registerTableSource("MyTable")

right?

I currently have a BatchTableSource class which produce a DataSet<Row> from a single geojson file.

This doesn't sound compatible with a custom InputFormat, don't you?

Thanks in advance for any addition hint, all the best

François

Le lun. 4 févr. 2019 à 12:10, Fabian Hueske <[hidden email]> a écrit :

Hi,

The files will be read in a streaming fashion.

Typically files are broken down into processing splits that are distributed to tasks for reading.

How a task reads a file split depends on the implementation, but usually the format reads the split as a stream and does not read the split as a whole before emitting records.

Best,

Fabian

Am Mo., 4. Feb. 2019 um 12:06 Uhr schrieb françois lacombe <[hidden email]>:

Hi Fabian,

Thank you for this input.

This is interesting.

With such an input format, will all the file will be loaded in memory before to be processed or will all be streamed?

All the best

François

Le mar. 29 janv. 2019 à 22:20, Fabian Hueske <[hidden email]> a écrit :

Hi,

You can point a file-based input format to a directory and the input format should read all files in that directory.

That works as well for TableSources that are internally use file-based input formats.

Is that what you are looking for?

Best, Fabian

Am Mo., 28. Jan. 2019 um 17:22 Uhr schrieb françois lacombe <[hidden email]>:

Hi all,

I'm wondering if it's possible and what's the best way to achieve the loading of multiple files with a Json source to a JDBC sink ?

I'm running Flink 1.7.0

Let's say I have about 1500 files with the same structure (same format, schema, everything) and I want to load them with a *batch* job

Can Flink handle the loading of one and each file in a single source and send data to my JDBC sink?

I wish I can provide the URL of the directory containing my thousand files to the batch source to make it load all of them sequentially.

My sources and sinks are currently available for BatchTableSource, I guess the cost to make them available for streaming would be quite expensive for me for the moment.

Have someone ever done this?

Am I wrong to expect doing so with a batch job?

All the best

François Lacombe

      

 Pensez à la planète, imprimer ce papier que si nécessaire 

      

 Pensez à la planète, imprimer ce papier que si nécessaire 

      

 Pensez à la planète, imprimer ce papier que si nécessaire 

      

 Pensez à la planète, imprimer ce papier que si nécessaire