Question about reading ORC file in Flink

[ShuQi]

Hi Guys,

The Flink version is 1.9.0. I use OrcTableSource to read ORC file in HDFS and the job is executed successfully, no any exception or error. But some fields(such as tagIndustry) are always null, actually these fields are not null. I can read these fields by direct reading it. Below is my code:

//main

         final ParameterTool params = ParameterTool.fromArgs(args);

        final ExecutionEnvironment env = ExecutionEnvironment.getExecutionEnvironment();

        env.getConfig().setGlobalJobParameters(params);

        Configuration config = new Configuration();

        OrcTableSource orcTableSource = OrcTableSource

            .builder()

            .path(params.get("input"))

            .forOrcSchema(TypeDescription.fromString(TYPE_INFO))

            .withConfiguration(config)

            .build();

        DataSet<Row> dataSet = orcTableSource.getDataSet(env);

        DataSet<Tuple2<String, Integer>> counts = dataSet.flatMap(new Tokenizer()).groupBy(0).sum(1);

//read field

public void flatMap(Row row, Collector<Tuple2<String, Integer>> out) {

            String content = ((String) row.getField(6));

            String tagIndustry = ((String) row.getField(35));

                LOGGER.info("arity: " + row.getArity());

                LOGGER.info("content: " + content);

                LOGGER.info("tagIndustry: " + tagIndustry);

                LOGGER.info("===========================");

            if (Strings.isNullOrEmpty(content) || Strings.isNullOrEmpty(tagIndustry) || !tagIndustry.contains("FD001")) {

                return;

            }

            // normalize and split the line

            String[] tokens = content.toLowerCase().split("\\W+");

            // emit the pairs

            for (String token : tokens) {

                if (token.length() > 0) {

                    out.collect(new Tuple2<>(token, 1));

                }

            }

        }

Thanks for your help!

QiShu

 

[Fabian Hueske-2]

Hi QiShu,

It might be that Flink's OrcInputFormat has a bug.

Can you open a Jira issue to report the problem? 

In order to be able to fix this, we need as much information as possible.

It would be great if you could create a minimal example of an ORC file and a program that reproduces the issue.

If that's not possible, we need the schema of an Orc file that cannot be correctly read.

Thanks,

Fabian

Am Mo., 23. Sept. 2019 um 11:40 Uhr schrieb ShuQi <[hidden email]>:

Hi Guys,

The Flink version is 1.9.0. I use OrcTableSource to read ORC file in HDFS and the job is executed successfully, no any exception or error. But some fields(such as tagIndustry) are always null, actually these fields are not null. I can read these fields by direct reading it. Below is my code:

//main

         final ParameterTool params = ParameterTool.fromArgs(args);

        final ExecutionEnvironment env = ExecutionEnvironment.getExecutionEnvironment();

        env.getConfig().setGlobalJobParameters(params);

        Configuration config = new Configuration();

        OrcTableSource orcTableSource = OrcTableSource

            .builder()

            .path(params.get("input"))

            .forOrcSchema(TypeDescription.fromString(TYPE_INFO))

            .withConfiguration(config)

            .build();

        DataSet<Row> dataSet = orcTableSource.getDataSet(env);

        DataSet<Tuple2<String, Integer>> counts = dataSet.flatMap(new Tokenizer()).groupBy(0).sum(1);

//read field

public void flatMap(Row row, Collector<Tuple2<String, Integer>> out) {

            String content = ((String) row.getField(6));

            String tagIndustry = ((String) row.getField(35));

                LOGGER.info("arity: " + row.getArity());

                LOGGER.info("content: " + content);

                LOGGER.info("tagIndustry: " + tagIndustry);

                LOGGER.info("===========================");

            if (Strings.isNullOrEmpty(content) || Strings.isNullOrEmpty(tagIndustry) || !tagIndustry.contains("FD001")) {

                return;

            }

            // normalize and split the line

            String[] tokens = content.toLowerCase().split("\\W+");

            // emit the pairs

            for (String token : tokens) {

                if (token.length() > 0) {

                    out.collect(new Tuple2<>(token, 1));

                }

            }

        }

Thanks for your help!

QiShu

 

[ShuQi]

Hi Fabian,

After debugging in local mode, I found that Flink orc connector is no problem, but some fields in our schema is in capital form，so these fields can not be matched.

But the program directly read orc file using includeColumns method, which will use equalsIgnoreCase to match the column, so it can read the fields.

Thanks for your Help!

Qi Shu

在 2019年9月24日，下午4:36，Fabian Hueske <[hidden email]> 写道：

Hi QiShu,

It might be that Flink's OrcInputFormat has a bug.

Can you open a Jira issue to report the problem? 

In order to be able to fix this, we need as much information as possible.

It would be great if you could create a minimal example of an ORC file and a program that reproduces the issue.

If that's not possible, we need the schema of an Orc file that cannot be correctly read.

Thanks,

Fabian

Am Mo., 23. Sept. 2019 um 11:40 Uhr schrieb ShuQi <[hidden email]>:

Hi Guys,

The Flink version is 1.9.0. I use OrcTableSource to read ORC file in HDFS and the job is executed successfully, no any exception or error. But some fields(such as tagIndustry) are always null, actually these fields are not null. I can read these fields by direct reading it. Below is my code:

//main

         final ParameterTool params = ParameterTool.fromArgs(args);

        final ExecutionEnvironment env = ExecutionEnvironment.getExecutionEnvironment();

        env.getConfig().setGlobalJobParameters(params);

        Configuration config = new Configuration();

        OrcTableSource orcTableSource = OrcTableSource

            .builder()

            .path(params.get("input"))

            .forOrcSchema(TypeDescription.fromString(TYPE_INFO))

            .withConfiguration(config)

            .build();

        DataSet<Row> dataSet = orcTableSource.getDataSet(env);

        DataSet<Tuple2<String, Integer>> counts = dataSet.flatMap(new Tokenizer()).groupBy(0).sum(1);

//read field

public void flatMap(Row row, Collector<Tuple2<String, Integer>> out) {

            String content = ((String) row.getField(6));

            String tagIndustry = ((String) row.getField(35));

                LOGGER.info("arity: " + row.getArity());

                LOGGER.info("content: " + content);

                LOGGER.info("tagIndustry: " + tagIndustry);

                LOGGER.info("===========================");

            if (Strings.isNullOrEmpty(content) || Strings.isNullOrEmpty(tagIndustry) || !tagIndustry.contains("FD001")) {

                return;

            }

            // normalize and split the line

            String[] tokens = content.toLowerCase().split("\\W+");

            // emit the pairs

            for (String token : tokens) {

                if (token.length() > 0) {

                    out.collect(new Tuple2<>(token, 1));

                }

            }

        }

Thanks for your help!

QiShu

 

[Fabian Hueske-2]

Thank you very much for coming back and reporting the good news! :-)

If you think that there is something that we can do to improve Flink's ORC input format, for example log a warning, please open a Jira.

Thank you,

Fabian

Am Mi., 25. Sept. 2019 um 05:14 Uhr schrieb 163 <[hidden email]>:

Hi Fabian,

After debugging in local mode, I found that Flink orc connector is no problem, but some fields in our schema is in capital form，so these fields can not be matched.

But the program directly read orc file using includeColumns method, which will use equalsIgnoreCase to match the column, so it can read the fields.

Thanks for your Help!

Qi Shu

在 2019年9月24日，下午4:36，Fabian Hueske <[hidden email]> 写道：

Hi QiShu,

It might be that Flink's OrcInputFormat has a bug.

Can you open a Jira issue to report the problem? 

In order to be able to fix this, we need as much information as possible.

It would be great if you could create a minimal example of an ORC file and a program that reproduces the issue.

If that's not possible, we need the schema of an Orc file that cannot be correctly read.

Thanks,

Fabian

Am Mo., 23. Sept. 2019 um 11:40 Uhr schrieb ShuQi <[hidden email]>:

Hi Guys,

The Flink version is 1.9.0. I use OrcTableSource to read ORC file in HDFS and the job is executed successfully, no any exception or error. But some fields(such as tagIndustry) are always null, actually these fields are not null. I can read these fields by direct reading it. Below is my code:

//main

         final ParameterTool params = ParameterTool.fromArgs(args);

        final ExecutionEnvironment env = ExecutionEnvironment.getExecutionEnvironment();

        env.getConfig().setGlobalJobParameters(params);

        Configuration config = new Configuration();

        OrcTableSource orcTableSource = OrcTableSource

            .builder()

            .path(params.get("input"))

            .forOrcSchema(TypeDescription.fromString(TYPE_INFO))

            .withConfiguration(config)

            .build();

        DataSet<Row> dataSet = orcTableSource.getDataSet(env);

        DataSet<Tuple2<String, Integer>> counts = dataSet.flatMap(new Tokenizer()).groupBy(0).sum(1);

//read field

public void flatMap(Row row, Collector<Tuple2<String, Integer>> out) {

            String content = ((String) row.getField(6));

            String tagIndustry = ((String) row.getField(35));

                LOGGER.info("arity: " + row.getArity());

                LOGGER.info("content: " + content);

                LOGGER.info("tagIndustry: " + tagIndustry);

                LOGGER.info("===========================");

            if (Strings.isNullOrEmpty(content) || Strings.isNullOrEmpty(tagIndustry) || !tagIndustry.contains("FD001")) {

                return;

            }

            // normalize and split the line

            String[] tokens = content.toLowerCase().split("\\W+");

            // emit the pairs

            for (String token : tokens) {

                if (token.length() > 0) {

                    out.collect(new Tuple2<>(token, 1));

                }

            }

        }

Thanks for your help!

QiShu