Spark java.lang.NoSuchMethodError for Univocity CSV Parser setDelimiter method

问题

I'm trying to run a Scala Spark job that uses the Univocity CSV Parser and after upgrading to support a String delimiter (vs only character), I'm getting the following error when I run my jar in the cluster. Running it locally in my IDEA IDE produces expected results with no errors.

ERROR yarn.ApplicationMaster: User class threw exception: java.lang.NoSuchMethodError: com.univocity.parsers.csv.CsvFormat.setDelimiter(Ljava/lang/String;)V
java.lang.NoSuchMethodError: com.univocity.parsers.csv.CsvFormat.setDelimiter(Ljava/lang/String;)V

I've tried the following: Eliminated all conflicting univocity parsers by examining the dependency tree via this: mvn dependency:tree -Dverbose -Dincludes=com.univocity:univocity-parsers which yields:

[INFO] --- maven-dependency-plugin:2.8:tree (default-cli) @ preval ---
[INFO] dataqa:preval:jar:1.0-SNAPSHOT
[INFO] \- com.univocity:univocity-parsers:jar:2.8.2:compile

I also tried to set the spark.executor.userClassPathFirst=true config when running the spark job, with no change in behavior.

Here is the dependency section in my pom.xml:

<dependency>
            <groupId>org.scala-lang</groupId>
            <artifactId>scala-library</artifactId>
            <version>2.11.12</version>
        </dependency>
        <!--
            Spark library. spark-core_2.xx must match the scala language version
        -->
        <dependency>
            <groupId>org.apache.spark</groupId>
            <artifactId>spark-core_2.11</artifactId>
            <version>2.0.0</version>
        </dependency>

        <!--
            Spark SQL library. spark-sql_2.xx must match the scala language version
        -->
        <dependency>
            <groupId>org.apache.spark</groupId>
            <artifactId>spark-sql_2.11</artifactId>
            <version>2.0.0</version>
            <exclusions>
                <exclusion>  <!-- declare the exclusion here -->
                    <groupId>com.univocity</groupId>
                    <artifactId>univocity-parsers</artifactId>
                </exclusion>
            </exclusions>
        </dependency>

        <!--
            Library to make REST API call
        -->
        <dependency>
            <groupId>com.typesafe.play</groupId>
            <artifactId>play-ahc-ws-standalone_2.11</artifactId>
            <version>2.0.0-M1</version>
        </dependency>


        <!--
            Parses delimited files
        -->
        <dependency>
            <groupId>com.univocity</groupId>
            <artifactId>univocity-parsers</artifactId>
            <version>2.8.2</version>
            <type>jar</type>
        </dependency>

        <!-- https://mvnrepository.com/artifact/com.googlecode.json-simple/json-simple -->
        <dependency>
            <groupId>com.googlecode.json-simple</groupId>
            <artifactId>json-simple</artifactId>
            <version>1.1.1</version>
        </dependency>

I wonder if Spark has a built-in dependency that is overriding my version (2.8 is the first version to support String arguments. Previously, it only supported character).

Any insights?

回答1:

A bit late, but if using --conf spark.driver.extraClassPath and spark.executor.extraClassPath is an option, please see my response here.

回答2:

After much time spent troubleshooting, I found a solution. I had to use the maven-shade-plugin as described here https://www.cloudera.com/documentation/enterprise/5-13-x/topics/spark_building.html#relocation

Here is the relevant portion of code I had to add to the maven-shade-plugin definition in my pom.xml:

<plugin>
    <groupId>org.apache.maven.plugins</groupId>
    <artifactId>maven-shade-plugin</artifactId>
    <version>3.2.1</version>
    <executions>
        <execution>
            <phase>package</phase>
            <goals>
                <goal>shade</goal>
            </goals>
            <configuration>

                <!-- other non-revelant filters etc omitted for brevity -->

                <relocations>
                    <!-- used to make sure there are no conflicts between the univocity parser version used here and the one that is bundled with spark -->
                    <relocation>
                        <pattern>com.univocity.parsers</pattern>
                        <shadedPattern>com.shaded.parsers</shadedPattern>
                    </relocation>
                </relocations>
            </configuration>
        </execution>
    </executions>
</plugin>

来源：https://stackoverflow.com/questions/57247793/spark-java-lang-nosuchmethoderror-for-univocity-csv-parser-setdelimiter-method

标签

apache-spark

univocity