如何处理 Apache Flink / AWS Kinesis 连接器中的无效 JSON？

Question

我有一个简单的 Apache Flink (PyFlink) 应用程序，它使用官方 flink 连接器和 Flink TableAPI 从 AWS Kinesis 流读取数据 https://nightlies.apache.org/flink/flink-docs-master/docs/connectors /数据流/运动/

from pyflink.table import EnvironmentSettings, TableEnvironment
import os
import json

# 1. Creates a Table Environment
env_settings = EnvironmentSettings.in_streaming_mode()
table_env = TableEnvironment.create(env_settings)

statement_set = table_env.create_statement_set()

APPLICATION_PROPERTIES_FILE_PATH = "/etc/flink/application_properties.json"  # on kda

is_local = (
    True if os.environ.get("IS_LOCAL") else False
)  # set this env var in your local environment

if is_local:
    # only for local, overwrite variable to properties and pass in your jars delimited by a semicolon (;)
    APPLICATION_PROPERTIES_FILE_PATH = "application_properties.json"  # local

    CURRENT_DIR = os.path.dirname(os.path.realpath(__file__))
    table_env.get_config().get_configuration().set_string(
        "pipeline.jars",
        "file:///" + CURRENT_DIR + "/lib/flink-sql-connector-kinesis-1.15.2.jar",
    )


def get_application_properties():
    if os.path.isfile(APPLICATION_PROPERTIES_FILE_PATH):
        with open(APPLICATION_PROPERTIES_FILE_PATH, "r") as file:
            contents = file.read()
            properties = json.loads(contents)
            return properties
    else:
        print('A file at "{}" was not found'.format(APPLICATION_PROPERTIES_FILE_PATH))


def property_map(props, property_group_id):
    for prop in props:
        if prop["PropertyGroupId"] == property_group_id:
            return prop["PropertyMap"]

def create_source_table(table_name, stream_name, region, stream_initpos):
    return """ CREATE TABLE {0} (
                ticker VARCHAR(6),
                price DOUBLE,
                event_time TIMESTAMP(3),
                WATERMARK FOR event_time AS event_time - INTERVAL '5' SECOND

              )
              PARTITIONED BY (ticker)
              WITH (
                'connector' = 'kinesis',
                'stream' = '{1}',
                'aws.region' = '{2}',
                'scan.stream.initpos' = '{3}',
                'format' = 'json',
                'json.timestamp-format.standard' = 'ISO-8601'
              ) """.format(
        table_name, stream_name, region, stream_initpos
    )


def create_sink_table(table_name, stream_name, region, stream_initpos):
    return """ CREATE TABLE {0} (
                ticker VARCHAR(6),
                price DOUBLE,
                event_time TIMESTAMP(3),
                WATERMARK FOR event_time AS event_time - INTERVAL '5' SECOND

              )
              PARTITIONED BY (ticker)
              WITH (
                'connector' = 'kinesis',
                'stream' = '{1}',
                'aws.region' = '{2}',
                'sink.partitioner-field-delimiter' = ';',
                'sink.batch.max-size' = '100',
                'format' = 'json',
                'json.timestamp-format.standard' = 'ISO-8601'
              ) """.format(
        table_name, stream_name, region
    )


def create_print_table(table_name, stream_name, region, stream_initpos):
    return """ CREATE TABLE {0} (
                ticker VARCHAR(6),
                price DOUBLE,
                event_time TIMESTAMP(3),
                WATERMARK FOR event_time AS event_time - INTERVAL '5' SECOND

              )
              WITH (
                'connector' = 'print'
              ) """.format(
        table_name, stream_name, region, stream_initpos
    )

def main():
    # Application Property Keys
    input_property_group_key = "consumer.config.0"
    producer_property_group_key = "producer.config.0"

    input_stream_key = "input.stream.name"
    input_region_key = "aws.region"
    input_starting_position_key = "flink.stream.initpos"

    output_stream_key = "output.stream.name"
    output_region_key = "aws.region"

    # tables
    input_table_name = "input_table"
    output_table_name = "output_table"

    # get application properties
    props = get_application_properties()

    input_property_map = property_map(props, input_property_group_key)
    output_property_map = property_map(props, producer_property_group_key)

    input_stream = input_property_map[input_stream_key]
    input_region = input_property_map[input_region_key]
    stream_initpos = input_property_map[input_starting_position_key]

    output_stream = output_property_map[output_stream_key]
    output_region = output_property_map[output_region_key]

    # 2. Creates a source table from a Kinesis Data Stream
    table_env.execute_sql(
        create_source_table(input_table_name, input_stream, input_region, stream_initpos)
    )

    # 3. Creates a sink table writing to a Kinesis Data Stream
    table_env.execute_sql(
        create_sink_table(output_table_name, output_stream, output_region, stream_initpos)
    )

    # 4. Inserts the source table data into the sink table
    table_result = table_env.execute_sql("INSERT INTO {0} SELECT * FROM {1}"
                                         .format(output_table_name, input_table_name))

    if is_local:
        table_result.wait()
    else:
        # get job status through TableResult
        print(table_result.get_job_client().get_job_status())


if __name__ == "__main__":
    main() 

问题是 Kinesis 流中偶尔会出现格式错误的 JSON，这会导致

java.lang.RuntimeException: java.io.IOException: Failed to deserialize JSON

需要提出的例外情况。如何配置应用程序/连接器以忽略 Kinesis 流中格式错误的条目并继续？

我尝试使用标准 try/ except 来处理此问题，但是由于此异常可能是在预打包的 Flink/Kinesis 连接器 .jar 中引起的，因此这没有任何效果。

Answer 1

您可以在 Kinesis 源（表 API）连接器中使用以下配置：

json.ignore-parse-errors

为

true

。

我不记得来源，但我使用了该属性。这里使用 Flink 1.18。